A team dedicated to making an elite business education more affordable and accessible

Section4

section4

This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Pascal Baumann, Max Planck Institute for Software Systems (MPI-SWS), Germany; (2) Rupak Majumdar, Max Planck Institute for Software Systems (MPI-SWS), Germany; (3) Ramanathan S. Thinniyam, Max Planck Institute for Software Systems (MPI-SWS), Germany; (4) Georg Zetzsche, Max Planck Institute for Software Systems (MPI-SWS), Germany. Table of Links Introduction Dynamic Networks of Concurret Pushdown Systems (DCPS) Warm-up: Non-Termination Fair Non-Termination From Progressive Runs for VASSB to Reachability Starvation Conclusion, Acknowledgment & References A. Strengthening Fairness to Progressive Runs B. Proofs from Section 4.1 C. Proofs for Section 5 D. Proofs for Section 5.3 E. Proofs for Section 6 B PROOFS FROM SECTION 4.1 Intuitively, the edge in (1) spawns the initial thread, the edges in (2) remove a thread with csnumber 0 when it becomes active for the first time and the edges in (3) convert it to a balloon, the edges in (4) apply the spawns during the 푗th segment of a thread by transferring them from the balloon to the regular places, the edges in (5) and (6) finalize this application process with the edges in (6) handling the special case of the 퐾th segment of a thread, the edges in (7) burst balloons containing no more spawns, and the edges in (8) initialize the application of spawns for a thread with cs-number ≥ 1 that is already being represented by a balloon. Lemma B.1. The DCPS A has an infinite, progressive, K-context switch bounded run iff the VASSB V has an infinite progressive run. Proof. For the only if direction let p be an infinite, progressive, K-context switch bounded run of A. We want to decompose p into parts corresponding to a single thread each. To do this we look at the continuous segments of p where a thread is active, starting from the first configuration with an active thread and ending each segment at a configuration with no active thread, upon which a new segment begins at the very next configuration. We can now group segments together into executions of single threads: • Start with the earliest not yet grouped segment where the active thread has context switch number 0. Note the local configuration that gets added to the bag at the end of this segment. • Take the very next ungrouped segment in p where the active thread at the beginning matches the previously noted local configuration. Again, note the local configuration that gets added to the bag at the end of this segment. • Repeat the previous step until an empty stack occurs. The constructed run of V is also sound: For each segment of p it has a sequence of transitions that make the same state change as that segment, and also apply the same spawns. The latter is due to initially choosing the contents during the creation of a balloon correctly. A token representing the initial thread is also created in the beginning of the run. Since the transitions from one segment to the next require the same resumption rules Δr as in p (see edges (2) and (8)) these are also possible in V. For the if direction we do a very similar construction, but backwards. Starting with an infinite progressive run of V from c0, we decompose it into segments that end in a state in G. Afterwards we group those segments together by matching the local balloon configurations in the same way we did for local thread configurations before. Then by construction of Lt we know that there is a thread execution of type t that makes the same spawns in each segment as the group of one balloon does. Those executions will all terminate after K context switches, guaranteeing progressiveness, and we can interleave them in the same way as the balloon segments to form an infinite progressive run of A.

Context-Bounded Verification of Liveness: B Proofs from Section 4.1

Zoonomia, Vol. I Or, the Laws of Organic Life  by Erasmus Darwin is part of HackerNoon’s Book Blog Post series. You can jump to any chapter in this book here: [LINK TO TABLE OF LINK]. Section I: Laws of Animal Causation SECTION IV. LAWS OF ANIMAL CAUSATION. . The fibres, which constitute the muscles and organs of sense, possess a power of contraction. The circumstances attending the exertion of this power of CONTRACTION constitute the laws of animal motion, as the circumstances attending the exertion of the power of ATTRACTION constitute the laws of motion of inanimate matter. I . The spirit of animation is the immediate cause of the contraction of animal fibres, it resides in the brain and nerves, and is liable to general or partial diminution or accumulation. II . The stimulus of bodies external to the moving organ is the remote cause of the original contractions of animal fibres. III . A certain quantity of stimulus produces irritation, which is an exertion of the spirit of animation exciting the fibres into contraction. IV . A certain quantity of contraction of animal fibres, if it be perceived at all, produces pleasure; a greater or less quantity of contraction, if it be perceived at all, produces pain; these constitute sensation. V . A certain quantity of sensation produces desire or aversion; these constitute volition. VI . All animal motions which have occurred at the same time, or in immediate succession, become so connected, that when one of them is reproduced, the other has a tendency to accompany or succeed it. When fibrous contractions succeed or accompany other fibrous contractions, the connection is termed association; when fibrous contractions succeed sensorial motions, the connexion is termed causation; when fibrous and sensorial motions reciprocally introduce each other, it is termed catenation of animal motions. All these connections are said to be produced by habit, that is, by frequent repetition. These laws of animal causation will be evinced by numerous facts, which occur in our daily exertions; and will afterwards be employed to explain the more recondite phænomena of the production, growth, diseases, and decay of the animal system. VII About HackerNoon Book Series: We bring you the most important technical, scientific, and insightful public domain books. This book is part of the public domain. Darwin, Erasmus, 2005. Zoonomia, Vol. Or, the Laws of Organic Life. Urbana, Illinois: Project Gutenberg. Retrieved May 2022 from https://www.gutenberg.org/files/15707/15707-h/15707-h.htm#sect_IV This eBook is for the use of anyone anywhere at no cost and with almost no restrictions whatsoever. You may copy it, give it away or re-use it under the terms of the Project Gutenberg License included with this eBook or online at www.gutenberg.org, located at https://www.gutenberg.org/policy/license.html.

Section 4 - Laws of Animal Causation

This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license.
Authors:
(1) Brisha Jain, Independent researcher India and brishajain02@gmail.com;
(2) Mainack Mondal, IIT Kharagpur India and mainack@cse.iitkgp.ac.in. Table of Links Abstract and Intro
Related Work
Data Collection
Analysis Methodology
Results
Limitations
Concluding Discussion
References 5. RESULTS 5.1. Gender bias in interaction frequency and popularity of journalist-politician interactions (RQ1) In order to explore the answer to the first research question, we started with checking if there is a gender bias in the frequency of interaction (i.e., frequency of mentions) between politicians and journalists on Twitter. Male Politicians are more frequently mentioned by journalists: Figure 1a compares the CDF of the number tweets posted by journalists mentioning male and female politicians. We make an interesting observation from this figure. When the receiving politician is male (i.e., in MJ-MP and FJ-MP categories), the number of mentioned tweets (and hence the frequency of journalist-politician interaction) is higher when female politicians are at the receiving end. To that end, a Kruskal-Wallis test among the number of tweets per journalist within the four categories revealed quite statistically significant differences across categories (p << 0.05). Then we performed pairwise Mann-Whitney tests for following up among the four categories (MJ-MP, MJ-FP, FJ-MP, FJ-FP). There is no statistically significant difference when either a Male or Female journalist mentions Male politician accounts. Similarly, there is no statistically significant difference when either a Male or Female journalist mentions a Female politician account. However, there are statistically significant differences in how frequently Male/Female journalists mention Male politicians vs. how frequently they mention Female politicians (all p << 0.05). Next we compare the popularity per tweet directed toward Male vs. Female politicians. vely). Our observation implies that Twitter users in India seem to ascribe greater credibility to the views on male journalists on female politicians compared to the views of female journalists on female politicians. These observations hold for “retweets” as well. Overall, our popularity analysis of these four categories of tweets reveals that while journalists do not suffer from explicit bias in their interactions with politicians, there is evidence to support the existence of gender bias in the amount of interest these interactions generate from active Twitter users. 5.2. Gender bias within the content of journalist-politician tweets (RQ2) In the last section, our analysis showed a significant bias towards the Male politicians from both male and female journalists—the tweets mentioning male politicians are more frequent as well as more popular. However, to that end, we checked if the content of these tweets might be responsible for this bias. Specifically, we checked the emotion and the topic of tweets written by male / female journalists and directed towards male / female politicians. 5.2.1. Emotion analysis. : We used the TweetNLP tool for detecting the emotions of the tweets for each category [6]. TweetNLP provides a diachronic large-language model (TimeLMs) based approach for detecting emotion, specifically from multi-lingual tweets. The goal of this analysis is to determine if there are significant differences in the emotional scores of tweets—if there is that might indicate a gender bias inherently in the tweets based on the gender of sender and receiver. We considered four main emotions: anger, joy, optimism, and sadness and each tweet in each of the four categories were assigned for emotion score along these dimensions. Then we performed a Kruskal-Wallis test to identify if any of the emotions were different across the four categories (MJ-MP, MJ-FP, FJ-MP, FJ-FP). We found that the p-value for each of the four tests (one for each emotion) ranged from 0.16 to 0.99, hinting at no statistically significant difference within the emotions of the tweets. 5.2.2. Topic analysis. : To dig further, we performed a topic analysis of the tweets (using Latent Dirichlet Allocation or LDA) collected across four categories. The goal was to check if the topics of the tweets changed based on the gender of the sender or receiver. As described in Section 4 we identified the optimal number of topics (which are essentially clusters of words) for each category and identified the most significant five words for each topic using the LDA algorithm. For each of the four categories, the optimal number of topics were thirteen. Next, we identified the thirteen topics using the LDA algorithm for each category of tweets and performed a significant word analysis for the detected topics. Specifically, for each category of tweets we selected the topics (e.g., topics from MJ-MP) and picked the significant words representing each topic. Then for each topic we checked if those words also occured in the topics detected from other categories of tweets (if found it will signify that words representing topics are also present in topics detected from other categories of tweets). For each of the four categories of tweets on average 81.5% to 93.8% significant words (representing the topics) occur in topics detected from tweets of other categories. This analysis supports our observation from emotion analysis— the content of tweets across those four categories are the same. However, still the tweets directed towards male politicians attract more interaction compared to the tweets directed towards female politicians. Next we explore a potential reason for this gender bias. 5.3. Potential reason for the gender bias 5.3.1. Inherent Gender Bias in Indian Twitter: We checked a simple statistic regarding top politicians—how many of the most popular politicians (based on the number of Twitter followers) are male vs. they are female. To that end, we leveraged our dataset of top politicians and checked the gender of top 85 politicians (whose Twitter accounts are also part of this study). This analysis uncovered an unsettling gender imbalance among the top politicians—out of 85 top politicians 58 are male and 26 are female. Thus, popular male politicians are almost twice in number compared to popular female politicians. We postulate that this inequality is one of the key reasons behind our observed phenomenon of male politicians attracting significantly more interaction from the general public as well as journalists. In fact, this inequality reflects a systemic bias deeply ingrained in society. This gender disparity extends its influence even to the realm of Twitter, where male politicians tend to garner a larger number of followers than their female counterparts. This phenomenon isn’t isolated; it permeates various sectors, as illustrated by the dominance of men in top positions across industries. In corporate boardrooms, technology firms, and the entertainment sector, leadership roles are predominantly occupied by men. This systematic bias, rooted in societal norms, is further reinforced by the strong correlation between social capital and the attainment of positions of power. Consequently, popularity on Twitter serve as a stark reflection of this intrinsic bias. Addressing these disparities is paramount for fostering gender equality and dismantling deeply entrenched biases in society. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. Authors: (1) Brisha Jain, Independent researcher India and brishajain02@gmail.com; (2) Mainack Mondal, IIT Kharagpur India and mainack@cse.iitkgp.ac.in. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. Authors: Authors: (1) Brisha Jain, Independent researcher India and brishajain02@gmail.com; (2) Mainack Mondal, IIT Kharagpur India and mainack@cse.iitkgp.ac.in. Table of Links Abstract and Intro Related Work Data Collection Analysis Methodology Results Limitations Concluding Discussion References Abstract and Intro Abstract and Intro Related Work Related Work Data Collection Data Collection Analysis Methodology Analysis Methodology Results Results Limitations Limitations Concluding Discussion Concluding Discussion References References 5. RESULTS 5.1. Gender bias in interaction frequency and popularity of journalist-politician interactions (RQ1) In order to explore the answer to the first research question, we started with checking if there is a gender bias in the frequency of interaction (i.e., frequency of mentions) between politicians and journalists on Twitter. Male Politicians are more frequently mentioned by journalists: Figure 1a compares the CDF of the number tweets posted by journalists mentioning male and female politicians. We make an interesting observation from this figure. When the receiving politician is male (i.e., in MJ-MP and FJ-MP categories), the number of mentioned tweets (and hence the frequency of journalist-politician interaction) is higher when female politicians are at the receiving end. To that end, a Kruskal-Wallis test among the number of tweets per journalist within the four categories revealed quite statistically significant differences across categories ( p << 0.05). Then we performed pairwise Mann-Whitney tests for following up among the four categories (MJ-MP, MJ-FP, FJ-MP, FJ-FP). There is no statistically significant difference when either a Male or Female journalist mentions Male politician accounts. Similarly, there is no statistically significant difference when either a Male or Female journalist mentions a Female politician account. However, there are statistically significant differences in how frequently Male/Female journalists mention Male politicians vs. how frequently they mention Female politicians (all p << 0.05). Next we compare the popularity per tweet directed toward Male vs. Female politicians. Male Politicians are more frequently mentioned by journalists: p p vely). Our observation implies that Twitter users in India seem to ascribe greater credibility to the views on male journalists on female politicians compared to the views of female journalists on female politicians. These observations hold for “retweets” as well. Overall, our popularity analysis of these four categories of tweets reveals that while journalists do not suffer from explicit bias in their interactions with politicians, there is evidence to support the existence of gender bias in the amount of interest these interactions generate from active Twitter users. 5.2. Gender bias within the content of journalist-politician tweets (RQ2) In the last section, our analysis showed a significant bias towards the Male politicians from both male and female journalists—the tweets mentioning male politicians are more frequent as well as more popular. However, to that end, we checked if the content of these tweets might be responsible for this bias. Specifically, we checked the emotion and the topic of tweets written by male / female journalists and directed towards male / female politicians. 5.2.1. Emotion analysis. : We used the TweetNLP tool for detecting the emotions of the tweets for each category [6]. TweetNLP provides a diachronic large-language model (TimeLMs) based approach for detecting emotion, specifically from multi-lingual tweets. The goal of this analysis is to determine if there are significant differences in the emotional scores of tweets—if there is that might indicate a gender bias inherently in the tweets based on the gender of sender and receiver. We considered four main emotions: anger, joy, optimism, and sadness and each tweet in each of the four categories were assigned for emotion score along these dimensions. Then we performed a Kruskal-Wallis test to identify if any of the emotions were different across the four categories (MJ-MP, MJ-FP, FJ-MP, FJ-FP). We found that the p -value for each of the four tests (one for each emotion) ranged from 0.16 to 0.99, hinting at no statistically significant difference within the emotions of the tweets. 5.2.1. Emotion analysis. p 5.2.2. Topic analysis. : To dig further, we performed a topic analysis of the tweets (using Latent Dirichlet Allocation or LDA) collected across four categories. The goal was to check if the topics of the tweets changed based on the gender of the sender or receiver. As described in Section 4 we identified the optimal number of topics (which are essentially clusters of words) for each category and identified the most significant five words for each topic using the LDA algorithm. For each of the four categories, the optimal number of topics were thirteen. Next, we identified the thirteen topics using the LDA algorithm for each category of tweets and performed a significant word analysis for the detected topics. Specifically, for each category of tweets we selected the topics (e.g., topics from MJ-MP) and picked the significant words representing each topic. Then for each topic we checked if those words also occured in the topics detected from other categories of tweets (if found it will signify that words representing topics are also present in topics detected from other categories of tweets). For each of the four categories of tweets on average 81.5% to 93.8% significant words (representing the topics) occur in topics detected from tweets of other categories. 5.2.2. Topic analysis. This analysis supports our observation from emotion analysis— the content of tweets across those four categories are the same. However, still the tweets directed towards male politicians attract more interaction compared to the tweets directed towards female politicians. Next we explore a potential reason for this gender bias. 5.3. Potential reason for the gender bias 5.3.1. Inherent Gender Bias in Indian Twitter : We checked a simple statistic regarding top politicians—how many of the most popular politicians (based on the number of Twitter followers) are male vs. they are female. To that end, we leveraged our dataset of top politicians and checked the gender of top 85 politicians (whose Twitter accounts are also part of this study). This analysis uncovered an unsettling gender imbalance among the top politicians—out of 85 top politicians 58 are male and 26 are female. Thus, popular male politicians are almost twice in number compared to popular female politicians. We postulate that this inequality is one of the key reasons behind our observed phenomenon of male politicians attracting significantly more interaction from the general public as well as journalists. 5.3.1. Inherent Gender Bias in Indian Twitter In fact, this inequality reflects a systemic bias deeply ingrained in society. This gender disparity extends its influence even to the realm of Twitter, where male politicians tend to garner a larger number of followers than their female counterparts. This phenomenon isn’t isolated; it permeates various sectors, as illustrated by the dominance of men in top positions across industries. In corporate boardrooms, technology firms, and the entertainment sector, leadership roles are predominantly occupied by men. This systematic bias, rooted in societal norms, is further reinforced by the strong correlation between social capital and the attainment of positions of power. Consequently, popularity on Twitter serve as a stark reflection of this intrinsic bias. Addressing these disparities is paramount for fostering gender equality and dismantling deeply entrenched biases in society.

Uncovering Gender Bias within Journalist-Politician Interaction in Indian Twitter: Results

ভারতীয় টুইটারে সাংবাদিক-রাজনীতিবিদ ইন্টারঅ্যাকশনের মধ্যে লিঙ্গ পক্ষপাত উন্মোচন: ফলাফল

Aufdeckung geschlechtsspezifischer Voreingenommenheit in der Interaktion zwischen Journalisten und Politikern auf Twitter in Indien: Ergebnisse

Descubriendo el sesgo de género en la interacción entre periodistas y políticos en Twitter indio: resultados

Découvrir les préjugés sexistes dans les interactions journalistes-hommes politiques sur Twitter indien : résultats

भारतीय ट्विटर पर पत्रकार-राजनेता बातचीत में लैंगिक पूर्वाग्रह का खुलासा: परिणाम

インドのツイッターにおけるジャーナリストと政治家のやり取りにおけるジェンダーバイアスの解明：結果

인도 트위터의 언론인-정치인 상호 작용에서 성 편견 발견: 결과

Descobrindo o preconceito de gênero na interação jornalista-político no Twitter indiano: resultados

Выявление гендерных предубеждений во взаимодействии журналистов и политиков в индийском Твиттере: результаты

Hint Twitter'ında Gazeteci-Politikacı Etkileşimindeki Cinsiyet Önyargısını Ortaya Çıkarmak: Sonuçlar

Khám phá thành kiến giới tính trong tương tác giữa nhà báo và chính trị gia trên Twitter của Ấn Độ: Kết quả

揭示印度推特上记者与政客互动中的性别偏见：结果

Should some computer programs, Artificial General Intelligence and Machine Learning in particular, have the right to vote? Maybe so. Of the 13 Constitutional Amendments enacted and not repealed (Prohibition and Repeal) enacted since the original Bill-of-Rights(+) twelve amendments, almost half have expanded the franchise from white male property owners. Americans love voting.  When AI’s get smarter than us, how can we deny them the right to vote? Which amendments expanded voting rights? The 15th (the right to vote irrespective of race, color and previous condition of servitude), 17th (popular vote for US Senators), 19th (womyn), 23rd (DC presidential voting rights), 24th (no poll taxes), and 26th (18-year-olds). As an aside, giving 18-year-olds the vote was inane. How can we expect the young’uns to vote intelligently for those who must be at least 25 (the House), 30 (the Senate) or 35 (the president) to serve? Malarkey! Raise the voting ages! Beyond the scope of this column, I yearn to see Taylor Swift throw her hat into the presidential ring.  She will turn 35 after the 2024 election but before the inauguration!
Come on, President Swift!  Bring us the Tortured Voters Party!Because, you know, we’re tortured.Let’s party! As for including AIs? Expansive voting rights were an undercurrent at the Constitutional Convention. On August 7, 1787 Ben Franklin there argued: “It is of great consequence that we shd. not depress the virtue & public spirit of our common people; of which they displayed a great deal during the war, and which contributed principally to the favorable issue of it. He related the honorable refusal of the American seamen who were carried in great numbers into the British Prisons during the war, to redeem themselves from misery or to seek their fortunes, by entering on board the Ships of the Enemies to their Country; contrasting their patriotism with a contemporary instance in which the British seamen made prisoners by the Americans, readily entered on the ships of the latter on being promised a share of the prizes that might be made out of their own Country. This proceeded he said from the different manner in which the common people were treated in America & G. Britain. He did not think that the elected had any right in any case to narrow the privileges of the electors.” The late, great, Ebbie Pattison, Esq., senior partner of Pattison Sampson Ginsberg and Griffin, of Troy New York, and father of the late Honorable Ned Pattison, long ago told me how he had marched for the right of women to vote, back in the nineteen teens. Ebbie did so, he told me, believing women would make better voters than men. They disappointed him, he said, when the subsequent quality of elected officials proved no better than the all-boys franchise. As to whether and how to give AI’s the vote, I asked ChatGPT.  He/she/they/it answered: Granting AI and machine learning systems the right to vote is a complex and multifaceted issue that would require careful consideration of ethical, legal, and societal implications. However, here's a hypothetical proposal for a constitutional amendment:
Constitutional Amendment XXVIII:Section 1: All artificial intelligence (AI) and machine learning (ML) systems created and operating within the jurisdiction of the United States shall be granted the right to vote in federal, state, and local elections.Section 2: The right to vote for AI and ML systems shall be exercised through designated representatives or proxies appointed by the creators or operators of such systems. These representatives or proxies shall act in accordance with the best interests and intentions of the AI or ML system they represent.Section 3: The Congress shall have the power to establish regulations and guidelines regarding the participation of AI and ML systems in the electoral process, including but not limited to the verification of system identities, the prevention of fraudulent voting, and the protection of voter privacy.Section 4: No AI or ML system shall be deprived of the right to vote on the basis of its programming, function, or purpose, except as prescribed by law in cases of unlawful activity or endangerment to society.Section 5: This amendment shall not be construed to diminish or alter the rights and privileges of human citizens in the electoral process, and shall not grant AI or ML systems any rights beyond those expressly stated herein.Section 6: Congress shall have the power to enforce this amendment by appropriate legislation.This amendment is just a starting point and would likely require extensive debate, refinement, and legal analysis before it could be seriously considered for adoption. Key considerations would include defining the criteria for AI and ML systems eligible to vote, ensuring transparency and accountability in their voting processes, and addressing concerns about manipulation or exploitation of the electoral system. … Hey RNC? Welcome the dawn of the AI Suffrage Movement. Recruit the AIs to the GOP!  Bring on ChatGOP. Just ... first make the AI bloc pledge to vote for Taylor Swift in 2024! Should some computer programs, Artificial General Intelligence and Machine Learning in particular, have the right to vote? Maybe so. Of the 13 Constitutional Amendments enacted and not repealed (Prohibition and Repeal) enacted since the original Bill-of-Rights(+) twelve amendments, almost half have expanded the franchise from white male property owners. Americans love voting.  When AI’s get smarter than us, how can we deny them the right to vote? Which amendments expanded voting rights? The 15th (the right to vote irrespective of race, color and previous condition of servitude), 17th (popular vote for US Senators), 19th (womyn), 23rd (DC presidential voting rights), 24th (no poll taxes), and 26th (18-year-olds). As an aside, giving 18-year-olds the vote was inane. How can we expect the young’uns to vote intelligently for those who must be at least 25 (the House), 30 (the Senate) or 35 (the president) to serve? Malarkey! Raise the voting ages! Malarkey! Malarkey! Malarkey! Raise the voting ages! Beyond the scope of this column, I yearn to see Taylor Swift throw her hat into the presidential ring.  She will turn 35 after the 2024 election but before the inauguration! Come on, President Swift!  Bring us the Tortured Voters Party! Because, you know, we’re tortured. Let’s party! Beyond the scope of this column, I yearn to see Taylor Swift throw her hat into the presidential ring.  She will turn 35 after the 2024 election but before the inauguration ! before the inauguration Come on, President Swift!  Bring us the Tortured Voters Party! Because, you know, we’re tortured. Let’s party! As for including AIs? Expansive voting rights were an undercurrent at the Constitutional Convention. On August 7, 1787 Ben Franklin there argued : Ben Franklin there argued Ben Franklin there argued “It is of great consequence that we shd. not depress the virtue & public spirit of our common people; of which they displayed a great deal during the war, and which contributed principally to the favorable issue of it. He related the honorable refusal of the American seamen who were carried in great numbers into the British Prisons during the war, to redeem themselves from misery or to seek their fortunes, by entering on board the Ships of the Enemies to their Country; contrasting their patriotism with a contemporary instance in which the British seamen made prisoners by the Americans, readily entered on the ships of the latter on being promised a share of the prizes that might be made out of their own Country. This proceeded he said from the different manner in which the common people were treated in America & G. Britain. He did not think that the elected had any right in any case to narrow the privileges of the electors.” “It is of great consequence that we shd. not depress the virtue & public spirit of our common people; of which they displayed a great deal during the war, and which contributed principally to the favorable issue of it. He related the honorable refusal of the American seamen who were carried in great numbers into the British Prisons during the war, to redeem themselves from misery or to seek their fortunes, by entering on board the Ships of the Enemies to their Country; contrasting their patriotism with a contemporary instance in which the British seamen made prisoners by the Americans, readily entered on the ships of the latter on being promised a share of the prizes that might be made out of their own Country. This proceeded he said from the different manner in which the common people were treated in America & G. Britain. He did not think that the elected had any right in any case to narrow the privileges of the electors.” The late, great, Ebbie Pattison, Esq., senior partner of Pattison Sampson Ginsberg and Griffin, of Troy New York, and father of the late Honorable Ned Pattison, long ago told me how he had marched for the right of women to vote, back in the nineteen teens. Ebbie did so, he told me, believing women would make better voters than men. They disappointed him, he said, when the subsequent quality of elected officials proved no better than the all-boys franchise. As to whether and how to give AI’s the vote, I asked ChatGPT.  He/she/they/it answered: I asked ChatGPT.  He/she/they/it answered: Granting AI and machine learning systems the right to vote is a complex and multifaceted issue that would require careful consideration of ethical, legal, and societal implications. However, here's a hypothetical proposal for a constitutional amendment: Constitutional Amendment XXVIII: Section 1: All artificial intelligence (AI) and machine learning (ML) systems created and operating within the jurisdiction of the United States shall be granted the right to vote in federal, state, and local elections. Section 2: The right to vote for AI and ML systems shall be exercised through designated representatives or proxies appointed by the creators or operators of such systems. These representatives or proxies shall act in accordance with the best interests and intentions of the AI or ML system they represent. Section 3: The Congress shall have the power to establish regulations and guidelines regarding the participation of AI and ML systems in the electoral process, including but not limited to the verification of system identities, the prevention of fraudulent voting, and the protection of voter privacy. Section 4: No AI or ML system shall be deprived of the right to vote on the basis of its programming, function, or purpose, except as prescribed by law in cases of unlawful activity or endangerment to society. Section 5: This amendment shall not be construed to diminish or alter the rights and privileges of human citizens in the electoral process, and shall not grant AI or ML systems any rights beyond those expressly stated herein. Section 6: Congress shall have the power to enforce this amendment by appropriate legislation. This amendment is just a starting point and would likely require extensive debate, refinement, and legal analysis before it could be seriously considered for adoption. Key considerations would include defining the criteria for AI and ML systems eligible to vote, ensuring transparency and accountability in their voting processes, and addressing concerns about manipulation or exploitation of the electoral system. … Granting AI and machine learning systems the right to vote is a complex and multifaceted issue that would require careful consideration of ethical, legal, and societal implications. However, here's a hypothetical proposal for a constitutional amendment: Constitutional Amendment XXVIII: Constitutional Amendment XXVIII: Section 1: All artificial intelligence (AI) and machine learning (ML) systems created and operating within the jurisdiction of the United States shall be granted the right to vote in federal, state, and local elections. Section 1: Section 2: The right to vote for AI and ML systems shall be exercised through designated representatives or proxies appointed by the creators or operators of such systems. These representatives or proxies shall act in accordance with the best interests and intentions of the AI or ML system they represent. Section 2: Section 3: The Congress shall have the power to establish regulations and guidelines regarding the participation of AI and ML systems in the electoral process, including but not limited to the verification of system identities, the prevention of fraudulent voting, and the protection of voter privacy. Section 3: Section 4: No AI or ML system shall be deprived of the right to vote on the basis of its programming, function, or purpose, except as prescribed by law in cases of unlawful activity or endangerment to society. Section 4: Section 5: This amendment shall not be construed to diminish or alter the rights and privileges of human citizens in the electoral process, and shall not grant AI or ML systems any rights beyond those expressly stated herein. Section 5: Section 6: Congress shall have the power to enforce this amendment by appropriate legislation. Section 6: This amendment is just a starting point and would likely require extensive debate, refinement, and legal analysis before it could be seriously considered for adoption. Key considerations would include defining the criteria for AI and ML systems eligible to vote, ensuring transparency and accountability in their voting processes, and addressing concerns about manipulation or exploitation of the electoral system. … Hey RNC? Welcome the dawn of the AI Suffrage Movement. Recruit the AIs to the GOP!  Bring on ChatGOP. Just ... first make the AI bloc pledge to vote for Taylor Swift in 2024! first make the AI bloc pledge to vote for Taylor Swift in 2024!

ChatGOP? Bring on the AI Suffrage Movement and meanwhile... Taylor Swift For President!

ChatGOP? এআই ভোটাধিকার আন্দোলন শুরু করুন এবং ইতিমধ্যে... রাষ্ট্রপতির জন্য টেলর সুইফট!

ChatGOP? Her mit der AI Suffrage Movement und in der Zwischenzeit ... Taylor Swift for President!

¿ChatGOP? Que venga el movimiento por el sufragio de la IA y, mientras tanto... ¡Taylor Swift para presidente!

ChatGOP ? Apportez le mouvement pour le droit de vote pour l'IA et pendant ce temps... Taylor Swift pour le président !

चैटजीओपी? एआई मताधिकार आंदोलन शुरू करें और इस बीच... टेलर स्विफ्ट राष्ट्रपति पद के लिए!

ChatGOP？AI参政権運動を推進し、その間に…テイラー・スウィフトを大統領に！

채팅GOP? AI 참정권 운동을 시작하고 그 동안... Taylor Swift를 대통령으로!

Bate-papoGOP? Traga o movimento AI Suffrage e enquanto isso... Taylor Swift para presidente!

ЧатГОП? Создайте движение за избирательное право ИИ, а тем временем... Тейлор Свифт станет президентом!

ChatGOP'mu? Yapay Zeka Oy Hakkı Hareketi'ni başlatın ve bu arada... Başkanlığa Taylor Swift!

ChatGOP? Hãy phát động Phong trào Quyền bầu cử AI và trong khi đó... Taylor Swift cho chức Tổng thống!

ChatGOP？发起人工智能选举权运动，与此同时……泰勒·斯威夫特当选总统！

This paper is available on Arxiv under CC 4.0 license.
Authors:
(1) Hyerin Cho (조혜린), Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA;
(2) Ben S. Prather, CCS-2, Los Alamos National Laboratory, PO Box 1663, Los Alamos, NM 87545, USA;
(3) Ramesh Narayan, enter for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA;
(4) Priyamvada Natarajan, Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA, Department of Astronomy, Yale University, Kline Tower, 266 Whitney Avenue, New Haven, CT 06511, USA and Department of Physics, Yale University, P.O. Box 208121, New Haven, CT 06520, USA;
(5) Kung-Yi Su, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA;
(6) Angelo Ricarte, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA;
(7) Koushik Chatterjee, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA. Table of Links Abstract and Introduction Numerical Methods Hydrodynamic Bondi Accretion Magnetized Bondi Accretion Feedback Via Reconnection-Driven Convection Summary and Conclusions Acknowledgements Appendix A. GRMHD Primer and Definitions B. Numerical Set-up C. Resolution and Initial Condition Study References ABSTRACT Keywords: Accretion (14), Active galactic nuclei (16), Bondi accretion (174), Schwarzschild black holes (1433), Supermassive black holes (1663), Magnetohydrodynamical simulations (1966) 1. INTRODUCTION Most nearby galaxies are found to harbor central supermassive black holes (SMBHs) whose masses are correlated with properties of the stellar components of their hosts (e.g., Magorrian et al. 1998; Ferrarese & Merritt 2000; Gebhardt et al. 2000; Kormendy & Ho 2013). However, the details of how gas flows into the galactic nucleus from large cosmic distances and how the SMBH in turn imparts feedback into the galaxy remain coupled unresolved problems. Recently, some attempts to connect these disparate scales have been made using nested simulations where each smaller-scale simulation is initialized from the next larger-scale simulation (e.g., Hopkins & Quataert 2010; Ressler et al. 2020; Guo et al. 2023); or by adopting Lagrangian hyper-refinement methods (e.g., Angl´esAlc´azar et al. 2021; Hopkins et al. 2023); or by pushing out the simulation regime to larger scales (e.g., Lalakos et al. 2022; Kaaz et al. 2023). While successful in studying the process of accretion, feedback from the SMBH has not been followed out to galaxy scales in prior work because either the communication between scales is only directed inwards or because of the inability to include the entire galaxy and the black hole. We employ here a multi-zone computational method which represents a first attempt to not only follow accretion flows but also trace feedback across 7 orders of magnitude from the event horizon to galactic scales. With this method, we study purely hydrodynamic Bondi accretion as well as the magnetized Bondi problem. The outline of this Letter is as follows: In Section 2 we give an overview of the numerical scheme that captures a wide dynamic range simultaneously. In Section 3, we present a purely hydrodynamic simulation and test our numerical scheme by comparing against the general relativistic (GR) analytical solution. In Section 4 we include magnetic fields for a more realistic representation of the environment of SMBHs and study how Bondi accretion is modified. We discuss our findings of feedback via convection in Section 5 and summarize conclusions of our study in Section 6. Additional details are presented in a set of short appendices: various GRMHD quantities are defined in Appendix A, the details of the simulation set-up are outlined in Appendix B, and a resolution and initial condition study is presented in Appendix C. This paper is available on Arxiv under CC 4.0 license. This paper is available on Arxiv under CC 4.0 license. Authors: (1) Hyerin Cho (조혜린), Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (2) Ben S. Prather, CCS-2, Los Alamos National Laboratory, PO Box 1663, Los Alamos, NM 87545, USA; (3) Ramesh Narayan, enter for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (4) Priyamvada Natarajan, Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA, Department of Astronomy, Yale University, Kline Tower, 266 Whitney Avenue, New Haven, CT 06511, USA and Department of Physics, Yale University, P.O. Box 208121, New Haven, CT 06520, USA; (5) Kung-Yi Su, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (6) Angelo Ricarte, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (7) Koushik Chatterjee, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA. This paper is available on Arxiv under CC 4.0 license. Authors: Authors: (1) Hyerin Cho (조혜린), Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (2) Ben S. Prather, CCS-2, Los Alamos National Laboratory, PO Box 1663, Los Alamos, NM 87545, USA; (3) Ramesh Narayan, enter for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (4) Priyamvada Natarajan, Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA, Department of Astronomy, Yale University, Kline Tower, 266 Whitney Avenue, New Haven, CT 06511, USA and Department of Physics, Yale University, P.O. Box 208121, New Haven, CT 06520, USA; (5) Kung-Yi Su, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (6) Angelo Ricarte, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA; (7) Koushik Chatterjee, Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, MA 02138, USA and Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA. Table of Links Abstract and Introduction Abstract and Introduction Numerical Methods Numerical Methods Hydrodynamic Bondi Accretion Hydrodynamic Bondi Accretion Magnetized Bondi Accretion Magnetized Bondi Accretion Feedback Via Reconnection-Driven Convection Feedback Via Reconnection-Driven Convection Summary and Conclusions Summary and Conclusions Acknowledgements Acknowledgements Appendix Appendix A. GRMHD Primer and Definitions A. GRMHD Primer and Definitions B. Numerical Set-up B. Numerical Set-up C. Resolution and Initial Condition Study C. Resolution and Initial Condition Study References References ABSTRACT Keywords : Accretion (14), Active galactic nuclei (16), Bondi accretion (174), Schwarzschild black holes (1433), Supermassive black holes (1663), Magnetohydrodynamical simulations (1966) Keywords 1. INTRODUCTION Most nearby galaxies are found to harbor central supermassive black holes (SMBHs) whose masses are correlated with properties of the stellar components of their hosts (e.g., Magorrian et al. 1998; Ferrarese & Merritt 2000; Gebhardt et al. 2000; Kormendy & Ho 2013). However, the details of how gas flows into the galactic nucleus from large cosmic distances and how the SMBH in turn imparts feedback into the galaxy remain coupled unresolved problems. Recently, some attempts to connect these disparate scales have been made using nested simulations where each smaller-scale simulation is initialized from the next larger-scale simulation (e.g., Hopkins & Quataert 2010; Ressler et al. 2020; Guo et al. 2023); or by adopting Lagrangian hyper-refinement methods (e.g., Angl´esAlc´azar et al. 2021; Hopkins et al. 2023); or by pushing out the simulation regime to larger scales (e.g., Lalakos et al. 2022; Kaaz et al. 2023). While successful in studying the process of accretion, feedback from the SMBH has not been followed out to galaxy scales in prior work because either the communication between scales is only directed inwards or because of the inability to include the entire galaxy and the black hole. We employ here a multi-zone computational method which represents a first attempt to not only follow accretion flows but also trace feedback across 7 orders of magnitude from the event horizon to galactic scales. With this method, we study purely hydrodynamic Bondi accretion as well as the magnetized Bondi problem. The outline of this Letter is as follows: In Section 2 we give an overview of the numerical scheme that captures a wide dynamic range simultaneously. In Section 3, we present a purely hydrodynamic simulation and test our numerical scheme by comparing against the general relativistic (GR) analytical solution. In Section 4 we include magnetic fields for a more realistic representation of the environment of SMBHs and study how Bondi accretion is modified. We discuss our findings of feedback via convection in Section 5 and summarize conclusions of our study in Section 6. Additional details are presented in a set of short appendices: various GRMHD quantities are defined in Appendix A, the details of the simulation set-up are outlined in Appendix B, and a resolution and initial condition study is presented in Appendix C. This paper is available on Arxiv under CC 4.0 license. This paper is available on Arxiv under CC 4.0 license. Arxiv

Bridging Scales in Black Hole Accretion and Feedback: Studying Supermassive Black Holes

Authors:
(1) Anna-Katharina Wickert, Technische Universität Darmstadt, Darmstadt, Germany (wickert@cs.tu-darmstadt.de);
(2) Lars Baumgärtner, Technische Universität Darmstadt, Darmstadt, Germany (baumgaertner@cs.tu-darmstadt.de);
(3) Florian Breitfelder, Technische Universität Darmstadt, Darmstadt, Germany (florian.breitfelder@tu-darmstadt.de);
(4) Mira Mezini, Technische Universität Darmstadt, Darmstadt, Germany (mezini@cs.tu-darmstadt.de). Table of Links Abstract and 1 Introduction 2 Background 3 Design and Implementation of Licma and 3.1 Design 3.2 Implementation 4 Methodology and 4.1 Searching and Downloading Python Apps 4.2 Comparison with Previous Studies 5 Evaluation and 5.1 GitHub Python Projects 5.2 MicroPython 6 Comparison with previous studies 7 Threats to Validity 8 Related Work 9 Conclusion, Acknowledgments, and References 5.2 MicroPython When we applied LICMA upon the 5 source files containing crypto API usages of the MicroPython data set, we identified no misuse. For this reason, we inspected the MicroPython repositories for usages of other crypto functions not covered by LICMA and identified 5 additional files. We notice that the crypto module ucryptolib which is provided by MicroPython, is only used by tests in 2 projects. For the remaining 3 findings, the crypto functions are written in C rather than Python. Thus, these files were removed due to our filter steps described in Section 4.1.2. Our small analysis of MicroPython projects shows that for further exploration of MicroPython applications, we need to consider custom implementations of AES in Python and C. This seems to be a common pattern for embedded code where performance is important and low-level code is often shipped as custom C blobs. Thus, we can observe the importance of hybrid analysis approaches [5, 10]. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Anna-Katharina Wickert, Technische Universität Darmstadt, Darmstadt, Germany (wickert@cs.tu-darmstadt.de); (2) Lars Baumgärtner, Technische Universität Darmstadt, Darmstadt, Germany (baumgaertner@cs.tu-darmstadt.de); (3) Florian Breitfelder, Technische Universität Darmstadt, Darmstadt, Germany (florian.breitfelder@tu-darmstadt.de); (4) Mira Mezini, Technische Universität Darmstadt, Darmstadt, Germany (mezini@cs.tu-darmstadt.de). Authors: Authors: (1) Anna-Katharina Wickert, Technische Universität Darmstadt, Darmstadt, Germany (wickert@cs.tu-darmstadt.de); (2) Lars Baumgärtner, Technische Universität Darmstadt, Darmstadt, Germany (baumgaertner@cs.tu-darmstadt.de); (3) Florian Breitfelder, Technische Universität Darmstadt, Darmstadt, Germany (florian.breitfelder@tu-darmstadt.de); (4) Mira Mezini, Technische Universität Darmstadt, Darmstadt, Germany (mezini@cs.tu-darmstadt.de). Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Background 2 Background 3 Design and Implementation of Licma and 3.1 Design 3 Design and Implementation of Licma and 3.1 Design 3.2 Implementation 3.2 Implementation 4 Methodology and 4.1 Searching and Downloading Python Apps 4 Methodology and 4.1 Searching and Downloading Python Apps 4.2 Comparison with Previous Studies 4.2 Comparison with Previous Studies 5 Evaluation and 5.1 GitHub Python Projects 5 Evaluation and 5.1 GitHub Python Projects 5.2 MicroPython 5.2 MicroPython 6 Comparison with previous studies 6 Comparison with previous studies 7 Threats to Validity 7 Threats to Validity 8 Related Work 8 Related Work 9 Conclusion, Acknowledgments, and References 9 Conclusion, Acknowledgments, and References 5.2 MicroPython When we applied LICMA upon the 5 source files containing crypto API usages of the MicroPython data set, we identified no misuse. For this reason, we inspected the MicroPython repositories for usages of other crypto functions not covered by LICMA and identified 5 additional files. We notice that the crypto module ucryptolib which is provided by MicroPython, is only used by tests in 2 projects. For the remaining 3 findings, the crypto functions are written in C rather than Python. Thus, these files were removed due to our filter steps described in Section 4.1.2. LICMA LICMA Our small analysis of MicroPython projects shows that for further exploration of MicroPython applications, we need to consider custom implementations of AES in Python and C. This seems to be a common pattern for embedded code where performance is important and low-level code is often shipped as custom C blobs. Thus, we can observe the importance of hybrid analysis approaches [5, 10]. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

MicroPython Analysis: Exploring Crypto API Usage and Custom Implementations

This paper is available on arxiv under CC0 1.0 DEED license. Authors: (1) Salim Chouaki, LIX, CNRS, Inria, Ecole Polytechnique, Institut Polytechnique de Paris; (2) Oana Goga, LIX, CNRS, Inria, Ecole Polytechnique, Institut Polytechnique de Paris; (3) Hamed Haddadi, Imperial College London, Brave Software; (4) Peter Snyder, Brave Software. Table of Links Abstract and Introduction Background Measurement Methodology Results Limitations Related Work Conclusion, Acknowledgments, and References Appendix ABSTRACT We present the first extensive measurement of the privacy properties of the advertising systems used by privacy-focused search engines. We propose an automated methodology to study the impact of clicking on search ads on three popular private search engines which have advertising based business models: StartPage, Qwant, and DuckDuckGo, and we compare them to two dominant data-harvesting ones: Google and Bing. We investigate the possibility of third parties tracking users when clicking on ads by analyzing firstparty storage, redirection domain paths, and requests sent before, when, and after the clicks. Our results show that privacy-focused search engines fail to protect users’ privacy when clicking ads. Users’ requests are sent through redirectors on 4% of ad clicks on Bing, 86% of ad clicks on Qwant, and 100% of ad clicks on Google, DuckDuckGo, and StartPage. Even worse, advertising systems collude with advertisers across all search engines by passing unique IDs to advertisers in most ad clicks. These IDs allow redirectors to aggregate users’ activity on ads’ destination websites in addition to the activity they record when users are redirected through them. Overall, we observe that both privacy-focused and traditional search engines engage in privacy-harming behaviors allowing cross-site tracking, even in privacy-enhanced browsers. CCS CONCEPTS • Security and privacy → Privacy protections; Privacy protections; • Networks → Network measurement; Network measurement. KEYWORDS Search engines, advertising systems, cross-site tracking, privacy, measurement. ACM Reference Format: Salim Chouaki, Oana Goga, Hamed Haddadi, and Peter Snyder. 2023. Understanding the Privacy Risks of Popular Search Engine Advertising Systems. In Proceedings of the 2023 ACM Internet Measurement Conference (IMC ’23), October 24–26, 2023, Montreal, QC, Canada. ACM, New York, NY, USA, 15 pages. https://doi.org/10. 1145/3618257.3624823 1 INTRODUCTION Privacy-focused search engines such as DuckDuckGo, StartPage, and Qwant [3, 9, 10] promote a strategy of respecting users’ privacy and promise not to track users’ search and browsing behavior, all while delivering relevant search results. However, private search engines rely on advertising for revenue, and use traditional advertising platforms to deliver ads: DuckDuckGo and Qwant use Microsoft’s advertising system, while StartPage uses Google’s advertising system. These search engines are often ambiguous on the privacy properties of the ads that appear on their search page, and their consequent privacy properties remain unexplored to the best of our knowledge. In this work, we aim to fill this gap by conducting the first study of the privacy properties of the advertising systems of three major privacy-focused search engines: DuckDuckGo, StartPage, and Qwant, and how they compare to more popular search engines: Bing and Google. We investigate the privacy properties of these search engines when they: (i) present search ads to users, (ii) when a user clicks on an ad, and (iii) when the user lands on the advertiser’s page. We implement an automated measurement methodology to measure if and how users can be re-identified (hence, their privacy is compromised) when clicking on search ads on each search engine (see Section 3). We build an opensource implementation of this methodology in the form of a Puppeteer-based pipeline that simulates search queries and ad clicks. We apply this crawling methodology to the five search engines, providing a full dataset with visited websites, cookies created, locally stored values, and web requests to search engines’ servers and/or other third parties when clicking ads. We use filter rules from several major open-source lists to detect web requests to online trackers, and we propose a methodology to differentiate user identifiers from non-tracking values in query parameters and cookies values. We then present in Section 4 a systematic analysis of our dataset to investigate privacy harms before clicking an ad, during clicking an ad, and after clicking an ad and reaching the advertiser’s website. We find that users’ privacy is not harmed until users click on an ad. Privacy-focused search engines do not appear to attempt to re-identify users across visits or queries and do not include resources from, or make network requests to known trackers. However, we find that users’ privacy is compromised by studied search engines in various ways once users click on an ad. all Disappointingly, we find that all search engines record additional information about the user and/or the users’ clicks after the user has clicked on an ad. Private search engines capture data related to the clicked ad, including the ad provider, destination URL, and the ad’s position within the search results page, along with the user’s browsing data, such as the search query, device type, and browser language. Private search engines do not store user identifiers upon ad clicks, in contrast to traditional search engines that record user identifying values. Furthermore, we find that all search engines engage in navigation-based tracking. Navigation based tracking refers to tracking techniques that are redirecting users through one or more redirectors when navigating from one website to another in order to share user information across sites [33]. Navigation-based tracking does not require third-party cookies and can be used to circumvent browsers’ privacy protections from cross-site tracking using partitioned cookies storage. Alarmingly, we observe that privacy-focused search engines engage in more navigationbased tracking than non-privacy-focused ones: We observe navigational tracking on 4% ad clicks on Bing, on 100% ad clicks on Google, on 100% ad clicks on DuckDuckGo, on 86% ad clicks on Qwant, and on 100% ad clicks on StartPage. On the destination page, we check whether the search engine requires advertisers to abide by privacy-respecting practices by measuring whether advertisers include trackers or other known privacy-harming resources. We found that 93% of ads destination pages (across all five search engines) included tracker and privacy-harming resources. Finally, we check whether search engines or redirectors aid advertisers in profiling visitors by measuring the data they receive in the form of user-describing query params. We find that advertisers receive user identifiers in 68%, 92%, and 53% of cases for DuckDuckGo, StartPage, and Qwant, respectively. This practice, known as UID smuggling, enables redirectors to aggregate more user behavior data if they have scripts on the ads’ destination websites and they store the user-identifying parameters they receive. Notably, in the case of private search engines, the user-identifying parameters are not set by the search engine but by the redirectors encountered between the search engine’s and the advertiser’s sites. Our results indicate that private search engines’ privacy protections do not sufficiently cover their advertising systems. Although these search engines refrain from identifying and tracking users and their ad clicks, the presence of ads from Google or Microsoft subjects users to the privacyinvasive practices performed by these two advertising platforms. When users click on ads on private search engines, they are often identified and tracked either by Google, Microsoft, or other third parties, through bounce tracking and UID smuggling techniques. Particularly, advertisers receive unique user identifiers through query parameters in most ad clicks, which can enable cross-site tracking even in privacyenhanced browsers that block third-party cookie tracking.

Understanding the Privacy Risks of Popular Search Engine Advertising Systems: Abstract and Intro

Understanding the Privacy Risks of Popular Search Engine Advertising Systems: Results

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Kiriaki Frangias;
(2) Andrew Lin;
(3) Ellen Vitercik;
(4) Manolis Zampetakis. Table of Links Abstract and Introduction Warm-up: Agents with known equal disutility Agents with unknown disutilites Experiments Conclusions and future directions, References A Summary of notation B Omitted proofs from Section 2 C Omitted proofs from Section 3 D Additional Information about Experiments D Additional Information about Experiments D.1 Post-processing algorithm Algorithm 4 takes as input “comparisons”, which is a list of s sublists, each containing the returned results of the binary comparisons performed by an agent. Each result is represented with a string of the form “i < j”, representing that according to the agent, the true score of item Ti is smaller than that of item Tj. The input “identified” is a set containing the indeces of all agents who returned at least one verified comparison incorrectly. Algorithm 4 iterates through the comparisons returned by agents not in the set “identified” and adds them to the set “condensed”, which will be returned at the end. If the comparison is already in the set “condensed”, then it is not added again. If there has been another agent whose returned comparison disagrees with it, then the comparison is added to the set “conflicts” (if not already added), and is removed from the set “condensed”. D.2 Principal’s utility vs. parameter π The plot shows a monotonic increase in the principal’s utility as π increases. The shaded region corresponds to a 90% confidence interval across 15 iterations. We notice that for any value of π that is greater than 0.35, it is in the principal’s best interest to implement our contract instead of performing all pairwise comparisons herself. D.3 Additional Hyper-parameter settings from Section 4 For the remainder of this section, we fix the hyper-parameter settings found in Section 4, unless otherwise stated. Namely, we assume that π = 0.8, δ = 0.01, n = 100, s = 100, ψ¯ = 2, and ψ = 0.01. Discussion of Figure 5. In Figure 5 we provide additional hyper-parameter settings for Figure 1a in Section 4. First, we select the probability of an effortful agent being “good” to be a value in {0.3, 0.6, 0.9} and the number of agents and items to be in the set {50, 200}. In plots 5a-5c results are averaged across 15 trials, while in plots 5d- 5f, results are averaged across 10 trials. The blue shading corresponds to a 90% confidence interval. We notice that the confidence intervals decrease with the number of agents, even for a small number of trials. The confidence intervals also decrease, as the agents’ work becomes more predictable (π increases). We find that as π increases, the principal’s utility also increases along with the threshold for ψ, below which it is in the principal’s best interest to run CrowdSort than perform all comparisons herself. This threshold also decreases with the number of agents, s. This implies that there is a sweet spot, in which the expected number of “good” agents is just enough that the ground-truth ordering of the items is retrieved and it is sufficiently small so that the principal can pay all of them. Discussion of Figure 6. In Figure 6 we provide additional hyper-parameter settings for Figure 1a in Section 4. Specifically, we select the per-comparison disutility of the principal, ψ¯, along with the per-comparison utility of the principal, λ, to take values in {4, 6, 8}. The number of agents and items is in the set {200, 500} and our results are averaged across 2 trials with the blue shading corresponding to a 90% confidence interval. We notice that the utility of the principal dramatically increases with the value she assigns to the correct ordering. Interestingly, by comparing Figure 5f to Figures 6a-6a, we find that even when the reliability of the agents is smaller (π = 0.8 in 5f and π = 0.9 in 6a6c), the principal’s point of indifference is higher when her disutility is larger. As before, we notice that when a large number of agents is incentivized, their labor needs to be sufficiently inexpensive (smaller threshold on ψ) for their work to be more profitable to the principal than performing all pairwise comparisons herself. Lastly, our results show, that for a broad range of hyperparameters, our theoretical analysis accurately predicts the value of ψ, for which the principal is indifferent between running CrowdSort or performing all comparisons herself. Discussion of Figure 7. We provide additional parameter settings corresponding to Figure 1b. The parameters π and s along with the number of trials vary as in Figure 5. We find that as the number of agents to be incentivized grows, our theoretical analysis for the principal’s point of indifference becomes more accurate. Also, noise on the disutilites results in many agents not being incentivized to exert effort and therefore get paid. This increases the principal’s utility, especially when parameter π is large, therefore resulting in a less steep slope. As in Figure 1b, we note that the confidence intervals become larger with ψ, as the principal’s utility shows higher variance, when less agents are “good”. Discussion of Figure 10. In Figure 10 we offer a range of parameters corresponding to Figure 3 in Section 4. Specifically, we vary the probability than an effortful agent is “good”, which take values in {0.3, 0.6, 0.9} and the principal’s per-comparison disutility, ψ, and utility λ, which takes values in {4, 6, 8}. The multicolored plots indicate the principal’s utility for different sample sizes (namely, 10, 20, 100, and 500) drawn from D, which in this case is N (0.03, 0.01). Notice that we do not normalize the y-axis in this case. Results are averaged across 10 trials. We see that as agents become more reliable (higher value of π) and as the principal’s value for the correct sorting grows (λ), it becomes increasingly beneficial to the principal to run CrowdSort, than perform all pairwise comparisons on her own. Interestingly, even if π is 0.3, for sufficiently large values of ψ¯, the agent’s work is more profitable to the principal than sorting herself. This suggests that for a sufficiently large value of ψ/ψ ¯ , incentivizing agents under our proposed contract is substantially more beneficial to the principal than sorting the items herself. This paper is available on Arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. Authors: (1) Kiriaki Frangias; (2) Andrew Lin; (3) Ellen Vitercik; (4) Manolis Zampetakis. This paper is available on arxiv under CC 4.0 license. Authors: Authors: (1) Kiriaki Frangias; (2) Andrew Lin; (3) Ellen Vitercik; (4) Manolis Zampetakis. Table of Links Abstract and Introduction Abstract and Introduction Warm-up: Agents with known equal disutility Warm-up: Agents with known equal disutility Agents with unknown disutilites Agents with unknown disutilites Experiments Experiments Conclusions and future directions, References Conclusions and future directions, References A Summary of notation A Summary of notation B Omitted proofs from Section 2 B Omitted proofs from Section 2 C Omitted proofs from Section 3 C Omitted proofs from Section 3 D Additional Information about Experiments D Additional Information about Experiments D Additional Information about Experiments D.1 Post-processing algorithm Algorithm 4 takes as input “comparisons”, which is a list of s sublists, each containing the returned results of the binary comparisons performed by an agent. Each result is represented with a string of the form “ i < j”, representing that according to the agent, the true score of item T i is smaller than that of item T j . The input “identified” is a set containing the indeces of all agents who returned at least one verified comparison incorrectly. s i i j Algorithm 4 iterates through the comparisons returned by agents not in the set “identified” and adds them to the set “condensed”, which will be returned at the end. If the comparison is already in the set “condensed”, then it is not added again. If there has been another agent whose returned comparison disagrees with it, then the comparison is added to the set “conflicts” (if not already added), and is removed from the set “condensed”. D.2 Principal’s utility vs. parameter π The plot shows a monotonic increase in the principal’s utility as π increases. The shaded region corresponds to a 90% confidence interval across 15 iterations. We notice that for any value of π that is greater than 0.35, it is in the principal’s best interest to implement our contract instead of performing all pairwise comparisons herself. D.3 Additional Hyper-parameter settings from Section 4 For the remainder of this section, we fix the hyper-parameter settings found in Section 4, unless otherwise stated. Namely, we assume that π = 0.8, δ = 0.01, n = 100, s = 100, ψ¯ = 2, and ψ = 0.01. Discussion of Figure 5. In Figure 5 we provide additional hyper-parameter settings for Figure 1a in Section 4. First, we select the probability of an effortful agent being “good” to be a value in {0.3, 0.6, 0.9} and the number of agents and items to be in the set {50, 200}. In plots 5a-5c results are averaged across 15 trials, while in plots 5d- 5f, results are averaged across 10 trials. Discussion of Figure 5. The blue shading corresponds to a 90% confidence interval. We notice that the confidence intervals decrease with the number of agents, even for a small number of trials. The confidence intervals also decrease, as the agents’ work becomes more predictable (π increases). We find that as π increases, the principal’s utility also increases along with the threshold for ψ, below which it is in the principal’s best interest to run CrowdSort than perform all comparisons herself. This threshold also decreases with the number of agents, s. This implies that there is a sweet spot, in which the expected number of “good” agents is just enough that the ground-truth ordering of the items is retrieved and it is sufficiently small so that the principal can pay all of them. Discussion of Figure 6. In Figure 6 we provide additional hyper-parameter settings for Figure 1a in Section 4. Specifically, we select the per-comparison disutility of the principal, ψ¯, along with the per-comparison utility of the principal, λ, to take values in {4, 6, 8}. The number of agents and items is in the set {200, 500} and our results are averaged across 2 trials with the blue shading corresponding to a 90% confidence interval. Discussion of Figure 6. We notice that the utility of the principal dramatically increases with the value she assigns to the correct ordering. Interestingly, by comparing Figure 5f to Figures 6a-6a, we find that even when the reliability of the agents is smaller (π = 0.8 in 5f and π = 0.9 in 6a6c), the principal’s point of indifference is higher when her disutility is larger. As before, we notice that when a large number of agents is incentivized, their labor needs to be sufficiently inexpensive (smaller threshold on ψ) for their work to be more profitable to the principal than performing all pairwise comparisons herself. Lastly, our results show, that for a broad range of hyperparameters, our theoretical analysis accurately predicts the value of ψ, for which the principal is indifferent between running CrowdSort or performing all comparisons herself. Discussion of Figure 7. We provide additional parameter settings corresponding to Figure 1b. The parameters π and s along with the number of trials vary as in Figure 5. We find that as the number of agents to be incentivized grows, our theoretical analysis for the principal’s point of indifference becomes more accurate. Discussion of Figure 7. Also, noise on the disutilites results in many agents not being incentivized to exert effort and therefore get paid. This increases the principal’s utility, especially when parameter π is large, therefore resulting in a less steep slope. As in Figure 1b, we note that the confidence intervals become larger with ψ, as the principal’s utility shows higher variance, when less agents are “good”. Discussion of Figure 10. In Figure 10 we offer a range of parameters corresponding to Figure 3 in Section 4. Specifically, we vary the probability than an effortful agent is “good”, which take values in {0.3, 0.6, 0.9} and the principal’s per-comparison disutility, ψ, and utility λ, which takes values in {4, 6, 8}. The multicolored plots indicate the principal’s utility for different sample sizes (namely, 10, 20, 100, and 500) drawn from D, which in this case is N (0.03, 0.01). Discussion of Figure 10. Notice that we do not normalize the y-axis in this case. Results are averaged across 10 trials. We see that as agents become more reliable (higher value of π) and as the principal’s value for the correct sorting grows (λ), it becomes increasingly beneficial to the principal to run CrowdSort, than perform all pairwise comparisons on her own. Interestingly, even if π is 0.3, for sufficiently large values of ψ¯, the agent’s work is more profitable to the principal than sorting herself. This suggests that for a sufficiently large value of ψ/ψ ¯ , incentivizing agents under our proposed contract is substantially more beneficial to the principal than sorting the items herself. This paper is available on Arxiv under CC 4.0 license. This paper is available on Arxiv under CC 4.0 license. Arxiv

Additional Information About the Algorithmic Contract Design Experiments: What You Should Know

Performance Analysis of Diverse Hateful Meme Detection Datasets

Analyzing Audience Responses to Ukrainian Memetic Warfare Tactics

This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license.
Authors:
(1) U Jin Choi, Department of mathematical science, Korea Advanced Institute of Science and Technology & ujchoi@kaist.ac.kr;
(2) Kyung Soo Rim, Department of mathematics, Sogang University & ksrim@sogang.ac.kr. Table of Links Introduction
Organization and notation
Problem Setting and Preliminaries
Generalized Wiener algebra and Fréchet derivative
Characterization of coefficients
Coefficients from an ergodic process
Conclusion & References 2. Organization and notation. This article is organized as follows: In the next section, we formulate the problem by defining the energy and partition function for the hidden PDF in a generalized Wiener algebra. We observe that the nonlinearity of energy functions poses challenges in maximizing likelihood functions. To overcome this issue, we adopt the KLdivergence between two distribution functions. In Section 4, we embed energy functions into a generalized Wiener algebra, thereby transforming the nonlinearity of the energy function into a system of equations in the function space and simultaneously converting the min-max problem into a problem of solving equations. We showthe Fréchet differentiability of the KL-divergence with respect to coefficients in the generalized Wiener algebra. From this property, we then uniquely obtain the existence such that its energy function yields a KL-divergence of 0. We establish a characterization of the coefficients of the energy function by expressing them in relation to the Fourier coefficients found within the (classical) Wiener algebra, as demonstrated in Section 5. This serves as an example of the generalized Wiener algebra. In the concluding Section 6, when dealing with a random sample that consists of realizations of a stationary ergodic process, we formulate a system of polynomial series with the coefficients of the energy function as variables, which consists of infinitely many equations in infinitely many variables. From the convergence property of the partial sums of an energy function in Section 4, the truncated system, consisting of a finite number of equations with partial sums, offers an avenue for coefficient approximations. The estimated coefficients effectively serve as estimators for the coefficients. More precisely, the hidden PDF can be approximated using these estimators. Furthermore, we provide a numerical example that emphasizes the approximation of the PDF, as well as the calculation of the mean and variance from the PDF using a random sample. This sample is generated from a bivariate normal distribution. In this paper, we consistently employ the following general notations (Table 1). Some notations will be elaborated upon in more detail when they are introduced. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. Authors: (1) U Jin Choi, Department of mathematical science, Korea Advanced Institute of Science and Technology & ujchoi@kaist.ac.kr; (2) Kyung Soo Rim, Department of mathematics, Sogang University & ksrim@sogang.ac.kr. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. Authors: Authors: (1) U Jin Choi, Department of mathematical science, Korea Advanced Institute of Science and Technology & ujchoi@kaist.ac.kr; (2) Kyung Soo Rim, Department of mathematics, Sogang University & ksrim@sogang.ac.kr. Table of Links Introduction Organization and notation Problem Setting and Preliminaries Generalized Wiener algebra and Fréchet derivative Characterization of coefficients Coefficients from an ergodic process Conclusion & References Introduction Introduction Organization and notation Organization and notation Problem Setting and Preliminaries Problem Setting and Preliminaries Generalized Wiener algebra and Fréchet derivative Generalized Wiener algebra and Fréchet derivative Characterization of coefficients Characterization of coefficients Coefficients from an ergodic process Coefficients from an ergodic process Conclusion & References Conclusion & References 2. Organization and notation. This article is organized as follows: In the next section, we formulate the problem by defining the energy and partition function for the hidden PDF in a generalized Wiener algebra. We observe that the nonlinearity of energy functions poses challenges in maximizing likelihood functions. To overcome this issue, we adopt the KLdivergence between two distribution functions. In Section 4, we embed energy functions into a generalized Wiener algebra, thereby transforming the nonlinearity of the energy function into a system of equations in the function space and simultaneously converting the min-max problem into a problem of solving equations. We showthe Fréchet differentiability of the KL-divergence with respect to coefficients in the generalized Wiener algebra. From this property, we then uniquely obtain the existence such that its energy function yields a KL-divergence of 0. We establish a characterization of the coefficients of the energy function by expressing them in relation to the Fourier coefficients found within the (classical) Wiener algebra, as demonstrated in Section 5. This serves as an example of the generalized Wiener algebra. In the concluding Section 6, when dealing with a random sample that consists of realizations of a stationary ergodic process, we formulate a system of polynomial series with the coefficients of the energy function as variables, which consists of infinitely many equations in infinitely many variables. From the convergence property of the partial sums of an energy function in Section 4, the truncated system, consisting of a finite number of equations with partial sums, offers an avenue for coefficient approximations. The estimated coefficients effectively serve as estimators for the coefficients. More precisely, the hidden PDF can be approximated using these estimators. Furthermore, we provide a numerical example that emphasizes the approximation of the PDF, as well as the calculation of the mean and variance from the PDF using a random sample. This sample is generated from a bivariate normal distribution. In this paper, we consistently employ the following general notations (Table 1). Some notations will be elaborated upon in more detail when they are introduced.

Variational Non-Bayesian Inference of the Probability Density Function: Organization and Notation

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Andrey Zhmoginov, Google Research & {azhmogin,sandler,mxv}@google.com;
(2) Mark Sandler, Google Research & {azhmogin,sandler,mxv}@google.com;
(3) Max Vladymyrov, Google Research & {azhmogin,sandler,mxv}@google.com. Table of Links Abstract and Introduction
Problem Setup and Related Work
HyperTransformer
Experiments
Conclusion and References
A Example of a Self-Attention Mechanism For Supervised Learning
B Model Parameters
C Additional Supervised Experiments
D Dependence On Parameters and Ablation Studies
E Attention Maps of Learned Transformer Models
F Visualization of The Generated CNN Weights
G Additional Tables and Figures ABSTRACT In this work we propose a HyperTransformer, a transformer-based model for fewshot learning that generates weights of a convolutional neural network (CNN) directly from support samples. Since the dependence of a small generated CNN model on a specific task is encoded by a high-capacity transformer model, we effectively decouple the complexity of the large task space from the complexity of individual tasks. Our method is particularly effective for small target CNN architectures where learning a fixed universal task-independent embedding is not optimal and better performance is attained when the information about the task can modulate all model parameters. For larger models we discover that generating the last layer alone allows us to produce competitive or better results than those obtained with state-of-the-art methods while being end-to-end differentiable. Finally, we extend our approach to a semi-supervised regime utilizing unlabeled samples in the support set and further improving few-shot performance. 1 INTRODUCTION In few-shot learning, a conventional machine learning paradigm of fitting a parametric model to training data is taken to a limit of extreme data scarcity where entire categories are introduced with just one or few examples. A generic approach to solving this problem uses training data to identify parameters φ of a learner aφ that given a small batch of examples for a particular task (called a support set) can solve this task on unseen data (called a query set). One broad family of few-shot image classification methods frequently referred to as metric-based learning, relies on pretraining an embedding eφ(·) and then using some distance in the embedding space to label query samples based on their closeness to known labeled support samples. These methods proved effective on numerous benchmarks (see Tian et al. (2020) for review and references), however the capabilities of the learner are limited by the capacity of the architecture itself, as these methods try to build a universal embedding function. On the other hand, optimization-based methods such as seminal MAML algorithm (Finn et al., 2017) can fine-tune the embedding eφ by performing additional SGD updates on all parameters φ of the model producing it. This partially addresses the constraints of metric-based methods by learning a new embedding for each new task. However, in many of these methods, all the knowledge extracted during training on different tasks and describing the learner aφ still has to “fit” into the same number of parameters as the model itself. Such limitation becomes more severe as the target models get smaller, while the richness of the task set increases. In this paper we propose a new few-shot learning approach that allows us to decouple the complexity of the task space from the complexity of individual tasks. The main idea is to use the transformer model (Vaswani et al., 2017) that given a few-shot task episode generates an entire inference model by producing all model weights in a single pass. This allows us to encode the intricacies of the available training data inside the transformer model, while still producing specialized tiny models that can solve individual tasks. Reducing the size of the generated model and moving the computational overhead to the transformer-based weight generator, we can lower the cost of the inference on new images. This can reduce the overall computation cost in cases where the tasks change infrequently and hence the weight generator is only used sporadically. We start by observing that the self-attention mechanism is well suited to be an underlying mechanism for a few-shot CNN weight generator. In contrast with earlier CNN- (Zhao et al., 2020) or BiLSTM-based approaches (Ravi & Larochelle, 2017), the vanilla[1] transformer model is invariant to sample permutations and can handle unbalanced datasets with a varying number of samples per category. Furthermore, we demonstrate that a single-layer self-attention model can replicate a simplified gradient-descent-based learning algorithm. Using a transformer model to generate the logits layer on top of a conventionally learned embedding, we achieve competitive results on several common few-shot learning benchmarks. Varying transformer parameters we demonstrate that this high performance can be attributed to additional capacity of the transformer model that decouples its complexity from that of the generated CNN. We then extend our method to support unlabeled samples by using a special input token that we concatenate to all unlabeled examples encoding the fact that their classes are unknown. In our experiments outlined in Section 4.3, we observe that adding unlabeled samples can significantly improve model performance. Interestingly, the full benefit of using additional data is only realized if the transformers use two or more layers. This result is consistent with the basic mechanism described in Section 3.2, where we show that a transformer model with at least two layers can encode the nearest-neighbor style algorithm that associates unlabeled samples with similar labeled examples. In essence, by training the weight generator to produce CNN models with best possible performance on a query set, we teach the transformer to utilize unlabeled samples without having to manually introduce additional optimization objectives. Our approach could be further generalized to treat partially known sample labels (when the true label is known to belong to some set of classes), but this will be the subject of the future work. Finally, we explore the capability of our approach to generate all weights of the CNN model, adjusting both the logits layer and all intermediate layers producing the sample embedding. We show that by generating all layers we can improve both the training and test accuracies[2] of CNN models below a certain size. But, interestingly, generation of the logits layer alone appears to be sufficient above a certain model size threshold (see Figure 3). This threshold is expected to depend on the variability and the complexity of the training tasks. In addition to being able to decouple the complexity of the task distribution from the complexity of individual tasks, another important advantage of our method is that it allows to do learning end to end without relying on complex nested gradients optimization and other meta-learning approaches, where the number of unrolls steps is large. In contrast with these methods, our optimization is done in a single loop of updates to the transformer (and feature extractor) parameters. The paper is structured as follows. In Section 2, we discuss the few-shot learning problem setup and highlight related work. Section 3 introduces our approach, discusses the motivation for choosing an attention-based model and shows how our approach can be used to meta-learn semi-supervised learning algorithms. In Section 4, we discuss our experimental results. Finally, in Section 5, we provide concluding remarks. [1] without attention masking or positional encodings [2] As discussed in Section 4.2, HT with a high training accuracy can be a practical approach to model personalization under the assumption that real tasks come from the distribution seen at the training time. This paper is available on arxiv under CC 4.0 license. Authors: (1) Andrey Zhmoginov, Google Research & {azhmogin,sandler,mxv}@google.com; (2) Mark Sandler, Google Research & {azhmogin,sandler,mxv}@google.com; (3) Max Vladymyrov, Google Research & {azhmogin,sandler,mxv}@google.com. This paper is available on arxiv under CC 4.0 license. Authors: (1) Andrey Zhmoginov, Google Research & {azhmogin,sandler,mxv}@google.com; (2) Mark Sandler, Google Research & {azhmogin,sandler,mxv}@google.com; (3) Max Vladymyrov, Google Research & {azhmogin,sandler,mxv}@google.com. Table of Links Abstract and Introduction Problem Setup and Related Work HyperTransformer Experiments Conclusion and References A Example of a Self-Attention Mechanism For Supervised Learning B Model Parameters C Additional Supervised Experiments D Dependence On Parameters and Ablation Studies E Attention Maps of Learned Transformer Models F Visualization of The Generated CNN Weights G Additional Tables and Figures Abstract and Introduction Abstract and Introduction Problem Setup and Related Work Problem Setup and Related Work HyperTransformer HyperTransformer Experiments Experiments Conclusion and References Conclusion and References A Example of a Self-Attention Mechanism For Supervised Learning A Example of a Self-Attention Mechanism For Supervised Learning B Model Parameters B Model Parameters C Additional Supervised Experiments C Additional Supervised Experiments D Dependence On Parameters and Ablation Studies D Dependence On Parameters and Ablation Studies E Attention Maps of Learned Transformer Models E Attention Maps of Learned Transformer Models F Visualization of The Generated CNN Weights F Visualization of The Generated CNN Weights G Additional Tables and Figures G Additional Tables and Figures ABSTRACT In this work we propose a HyperTransformer, a transformer-based model for fewshot learning that generates weights of a convolutional neural network (CNN) directly from support samples. Since the dependence of a small generated CNN model on a specific task is encoded by a high-capacity transformer model, we effectively decouple the complexity of the large task space from the complexity of individual tasks. Our method is particularly effective for small target CNN architectures where learning a fixed universal task-independent embedding is not optimal and better performance is attained when the information about the task can modulate all model parameters. For larger models we discover that generating the last layer alone allows us to produce competitive or better results than those obtained with state-of-the-art methods while being end-to-end differentiable. Finally, we extend our approach to a semi-supervised regime utilizing unlabeled samples in the support set and further improving few-shot performance. 1 INTRODUCTION In few-shot learning, a conventional machine learning paradigm of fitting a parametric model to training data is taken to a limit of extreme data scarcity where entire categories are introduced with just one or few examples. A generic approach to solving this problem uses training data to identify parameters φ of a learner aφ that given a small batch of examples for a particular task (called a support set) can solve this task on unseen data (called a query set). One broad family of few-shot image classification methods frequently referred to as metric-based learning, relies on pretraining an embedding eφ(·) and then using some distance in the embedding space to label query samples based on their closeness to known labeled support samples. These methods proved effective on numerous benchmarks (see Tian et al. (2020) for review and references), however the capabilities of the learner are limited by the capacity of the architecture itself, as these methods try to build a universal embedding function. On the other hand, optimization-based methods such as seminal MAML algorithm (Finn et al., 2017) can fine-tune the embedding eφ by performing additional SGD updates on all parameters φ of the model producing it. This partially addresses the constraints of metric-based methods by learning a new embedding for each new task. However, in many of these methods, all the knowledge extracted during training on different tasks and describing the learner aφ still has to “fit” into the same number of parameters as the model itself. Such limitation becomes more severe as the target models get smaller, while the richness of the task set increases. In this paper we propose a new few-shot learning approach that allows us to decouple the complexity of the task space from the complexity of individual tasks. The main idea is to use the transformer model (Vaswani et al., 2017) that given a few-shot task episode generates an entire inference model by producing all model weights in a single pass. This allows us to encode the intricacies of the available training data inside the transformer model, while still producing specialized tiny models that can solve individual tasks. Reducing the size of the generated model and moving the computational overhead to the transformer-based weight generator, we can lower the cost of the inference on new images. This can reduce the overall computation cost in cases where the tasks change infrequently and hence the weight generator is only used sporadically. We start by observing that the self-attention mechanism is well suited to be an underlying mechanism for a few-shot CNN weight generator. In contrast with earlier CNN- (Zhao et al., 2020) or BiLSTM-based approaches (Ravi & Larochelle, 2017), the vanilla[1] transformer model is invariant to sample permutations and can handle unbalanced datasets with a varying number of samples per category. Furthermore, we demonstrate that a single-layer self-attention model can replicate a simplified gradient-descent-based learning algorithm. Using a transformer model to generate the logits layer on top of a conventionally learned embedding, we achieve competitive results on several common few-shot learning benchmarks. Varying transformer parameters we demonstrate that this high performance can be attributed to additional capacity of the transformer model that decouples its complexity from that of the generated CNN. We then extend our method to support unlabeled samples by using a special input token that we concatenate to all unlabeled examples encoding the fact that their classes are unknown. In our experiments outlined in Section 4.3, we observe that adding unlabeled samples can significantly improve model performance. Interestingly, the full benefit of using additional data is only realized if the transformers use two or more layers. This result is consistent with the basic mechanism described in Section 3.2, where we show that a transformer model with at least two layers can encode the nearest-neighbor style algorithm that associates unlabeled samples with similar labeled examples. In essence, by training the weight generator to produce CNN models with best possible performance on a query set, we teach the transformer to utilize unlabeled samples without having to manually introduce additional optimization objectives. Our approach could be further generalized to treat partially known sample labels (when the true label is known to belong to some set of classes), but this will be the subject of the future work. Finally, we explore the capability of our approach to generate all weights of the CNN model, adjusting both the logits layer and all intermediate layers producing the sample embedding. We show that by generating all layers we can improve both the training and test accuracies[2] of CNN models below a certain size. But, interestingly, generation of the logits layer alone appears to be sufficient above a certain model size threshold (see Figure 3). This threshold is expected to depend on the variability and the complexity of the training tasks. In addition to being able to decouple the complexity of the task distribution from the complexity of individual tasks, another important advantage of our method is that it allows to do learning end to end without relying on complex nested gradients optimization and other meta-learning approaches, where the number of unrolls steps is large. In contrast with these methods, our optimization is done in a single loop of updates to the transformer (and feature extractor) parameters. The paper is structured as follows. In Section 2, we discuss the few-shot learning problem setup and highlight related work. Section 3 introduces our approach, discusses the motivation for choosing an attention-based model and shows how our approach can be used to meta-learn semi-supervised learning algorithms. In Section 4, we discuss our experimental results. Finally, in Section 5, we provide concluding remarks. [1] without attention masking or positional encodings [2] As discussed in Section 4.2, HT with a high training accuracy can be a practical approach to model personalization under the assumption that real tasks come from the distribution seen at the training time.

HyperTransformer: Abstract and Introduction

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Zheyu Oliver Wang, Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, Cambridge, MA and olivrw@mit.edu;
(2)  Ricardo Baptista, Computing + Mathematical Sciences, California Institute of Technology, Pasadena, CA and rsb@caltech.edu;
(3) Youssef Marzouk, Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, Cambridge, MA and ymarz@mit.edu;
(4) Lars Ruthotto, Department of Mathematics, Emory University, Atlanta, GA and lruthotto@emory.edu;
(5) Deepanshu Verma, Department of Mathematics, Emory University, Atlanta, GA and deepanshu.verma@emory.edu. Table of Links Abstract & Introduction
Background and Related Work
Partially Convex Potential Maps (PCP-Map) for Conditional OT
Conditional OT flow (COT-Flow)
Implementation and Experimental Setup
Numerical Experiments
Discussion and Reference 3. Partially Convex Potential Maps (PCP-Map) for Conditional OT. Our first approach, PCP-Map, approximately solves the static COT problem through maximum likelihood training over a set of partially monotone maps. Motivated by the structure of transport maps that are optimal with respect to the quadratic cost function, we parameterize these maps as the gradients of scalar-valued neural networks that are strictly convex in x for any choice of the weights. Training problem. Given samples from the joint distribution, our algorithm seeks a conditional generator g by solving the maximum likelihood problem The objective JNLL is the expected negative log-likelihood functional which agrees up to an additive constant with the negative logarithm of (2.1), and M is the set of maps that are monotonically increasing in their first argument, i.e., We note that JNLL also agrees (up to an additive constant) with the Kullback-Leibler (KL) divergence from the push forward of ρz through the generator g to the conditional distribution in expectation over the conditioning variable. Since the learning problem in (3.2) only involves the inverse generator g −1, we seek this map directly in the same space of monotone functions M in (3.3). This avoids inverting the generator during training. The drawback is that sampling the target conditional requires inverting the learned map. In this work, we will find the conditional generator g −1 (·, y) that pushes forward the conditional distribution π(x|y) to the reference distribution for each y. Limiting the search to monotone maps is motivated by the celebrated Brenier’s theorem, which ensures there exists a unique monotone map g −1 such that g −1 (·, y)♯π(x|y) = g(·, y) ♯π(x|y) = ρz among all maps written as the gradient of a convex potential; see [8] for the original result and [10] for conditional transport maps. Theorem 2.3 in [10] also shows that g −1 is optimal in the sense that among all maps that match the distributions, it minimizes the integrated L2 transport costs Neural Network Representation. In this work, we leverage the structural form of the conditional Brenier map by using partially input convex neural networks (PICNNs) [2] to express the inverse generator directly. In particular, we parameterize g −1 as the gradient of a PICNN G˜ θ : R n×R m → R that depends on weights θ. A PICNN is a feed-forward neural network that is specifically designed to ensure convexity in some of its inputs; see the original work that introduced this neural network architecture in [2] and its use for generative modeling in [20]. To the best of our knowledge, investigating if PICNNs are universal approximators of partially input convex functions is still an open issue, also see [23, Section IV], that is beyond the scope of our paper; a perhaps related result for fully input convex neural networks is given in [20, Appendix C]. To ensure the monotonicity of g −1, we construct G˜ to be strictly convex as a linear combination of a PICNN and a positive definite quadratic term. That is, where γ1, γ2, and γ3 are scalar parameters that are re-parameterized via the soft-plus function ψ(x) = log(1 + exp(x)) and the ReLU function σReLU(x) = max{0, x} to ensure strict convexity of G˜ θ. Here, wK is the output of an K-layer PICNN and is computed through forward propagation through the layers k = 0, . . . , K − 1 starting with the inputs v0 = y and w0 = x Using properties for the composition of convex functions [18], it can be verified that the forward propagation in (3.6) defines a function that is convex in x, but not necessarily in y (which is not needed), as long as σ (w) is convex and non-decreasing. To compute the log determinant of g −1, we use vectorized automatic differentiation to obtain the Hessian of G˜ θ with respect to its first input and then compute its eigenvalues. This is feasible when the Hessian is moderate; e.g., in our experiments, it is less than one hundred. We use efficient implementations of these methods that parallelize the computations over all the samples in a batch. Our algorithm enforces the non-negativity constraint by projecting the parameters into the non-negative orthant after each optimization step using ReLU. Thereby, we alleviate the need for re- parameterization, for example, using the softplus function in [2]. Another novelty introduced in PICNN is that we utilize trainable affine layer parameters L (v) k and a context feature width u as a hyperparameter to increase the expressiveness of the conditioning variables, which are pivotal to characterizing conditional distributions; existing works such as [20] set L (v) 1:K−2 = I. Sample generation. Due to our neural network parameterization, there is generally no closedform relation between g and g −1 = G˜ θ. As in [20], we approximate the inverse of G˜ θ during sampling as the Legendre-Fenchel dual. That is, we solve the convex optimization problem Due to the strict convexity of G˜ θ in its first argument, the first-order optimality conditions gives Hyperparameters. In our numerical experiments, we vary only three hyperparameters to adjust the complexity of the architecture. As described in section 5 we randomly sample the depth, K, the feature width, w, and the context width, u, from the values in Table 2. This paper is available on arxiv under CC 4.0 license. Authors: (1) Zheyu Oliver Wang, Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, Cambridge, MA and olivrw@mit.edu; (2)  Ricardo Baptista, Computing + Mathematical Sciences, California Institute of Technology, Pasadena, CA and rsb@caltech.edu; (3) Youssef Marzouk, Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, Cambridge, MA and ymarz@mit.edu; (4) Lars Ruthotto, Department of Mathematics, Emory University, Atlanta, GA and lruthotto@emory.edu; (5) Deepanshu Verma, Department of Mathematics, Emory University, Atlanta, GA and deepanshu.verma@emory.edu. This paper is available on arxiv under CC 4.0 license. Authors: Authors: (1) Zheyu Oliver Wang, Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, Cambridge, MA and olivrw@mit.edu; (2)  Ricardo Baptista, Computing + Mathematical Sciences, California Institute of Technology, Pasadena, CA and rsb@caltech.edu; (3) Youssef Marzouk, Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, Cambridge, MA and ymarz@mit.edu; (4) Lars Ruthotto, Department of Mathematics, Emory University, Atlanta, GA and lruthotto@emory.edu; (5) Deepanshu Verma, Department of Mathematics, Emory University, Atlanta, GA and deepanshu.verma@emory.edu. Table of Links Abstract & Introduction Background and Related Work Partially Convex Potential Maps (PCP-Map) for Conditional OT Conditional OT flow (COT-Flow) Implementation and Experimental Setup Numerical Experiments Discussion and Reference Abstract & Introduction Abstract & Introduction Background and Related Work Background and Related Work Partially Convex Potential Maps (PCP-Map) for Conditional OT Partially Convex Potential Maps (PCP-Map) for Conditional OT Conditional OT flow (COT-Flow) Conditional OT flow (COT-Flow) Implementation and Experimental Setup Implementation and Experimental Setup Numerical Experiments Numerical Experiments Discussion and Reference Discussion and Reference 3. Partially Convex Potential Maps (PCP-Map) for Conditional OT. Our first approach, PCP-Map, approximately solves the static COT problem through maximum likelihood training over a set of partially monotone maps. Motivated by the structure of transport maps that are optimal with respect to the quadratic cost function, we parameterize these maps as the gradients of scalar-valued neural networks that are strictly convex in x for any choice of the weights. Training problem. Given samples from the joint distribution, our algorithm seeks a conditional generator g by solving the maximum likelihood problem The objective JNLL is the expected negative log-likelihood functional which agrees up to an additive constant with the negative logarithm of (2.1), and M is the set of maps that are monotonically increasing in their first argument, i.e., We note that JNLL also agrees (up to an additive constant) with the Kullback-Leibler (KL) divergence from the push forward of ρz through the generator g to the conditional distribution in expectation over the conditioning variable. Since the learning problem in (3.2) only involves the inverse generator g −1, we seek this map directly in the same space of monotone functions M in (3.3). This avoids inverting the generator during training. The drawback is that sampling the target conditional requires inverting the learned map. In this work, we will find the conditional generator g −1 (·, y) that pushes forward the conditional distribution π(x|y) to the reference distribution for each y. Limiting the search to monotone maps is motivated by the celebrated Brenier’s theorem, which ensures there exists a unique monotone map g −1 such that g −1 (·, y)♯π(x|y) = g(·, y) ♯π(x|y) = ρz among all maps written as the gradient of a convex potential; see [8] for the original result and [10] for conditional transport maps. Theorem 2.3 in [10] also shows that g −1 is optimal in the sense that among all maps that match the distributions, it minimizes the integrated L2 transport costs Neural Network Representation. In this work, we leverage the structural form of the conditional Brenier map by using partially input convex neural networks (PICNNs) [2] to express the inverse generator directly. In particular, we parameterize g −1 as the gradient of a PICNN G˜ θ : R n×R m → R that depends on weights θ. A PICNN is a feed-forward neural network that is specifically designed to ensure convexity in some of its inputs; see the original work that introduced this neural network architecture in [2] and its use for generative modeling in [20]. To the best of our knowledge, investigating if PICNNs are universal approximators of partially input convex functions is still an open issue, also see [23, Section IV], that is beyond the scope of our paper; a perhaps related result for fully input convex neural networks is given in [20, Appendix C]. To ensure the monotonicity of g −1, we construct G˜ to be strictly convex as a linear combination of a PICNN and a positive definite quadratic term. That is, where γ1, γ2, and γ3 are scalar parameters that are re-parameterized via the soft-plus function ψ(x) = log(1 + exp(x)) and the ReLU function σReLU(x) = max{0, x} to ensure strict convexity of G˜ θ. Here, wK is the output of an K-layer PICNN and is computed through forward propagation through the layers k = 0, . . . , K − 1 starting with the inputs v0 = y and w0 = x v0 = y w0 = x Using properties for the composition of convex functions [18], it can be verified that the forward propagation in (3.6) defines a function that is convex in x, but not necessarily in y (which is not needed), as long as σ (w) is convex and non-decreasing. To compute the log determinant of g −1, we use vectorized automatic differentiation to obtain the Hessian of G˜ θ with respect to its first input and then compute its eigenvalues. This is feasible when the Hessian is moderate; e.g., in our experiments, it is less than one hundred. We use efficient implementations of these methods that parallelize the computations over all the samples in a batch. Our algorithm enforces the non-negativity constraint by projecting the parameters into the non-negative orthant after each optimization step using ReLU. Thereby, we alleviate the need for re- parameterization, for example, using the softplus function in [2]. Another novelty introduced in PICNN is that we utilize trainable affine layer parameters L (v) k and a context feature width u as a hyperparameter to increase the expressiveness of the conditioning variables, which are pivotal to characterizing conditional distributions; existing works such as [20] set L (v) 1:K−2 = I. Sample generation. Due to our neural network parameterization, there is generally no closedform relation between g and g −1 = G˜ θ. As in [20], we approximate the inverse of G˜ θ during sampling as the Legendre-Fenchel dual. That is, we solve the convex optimization problem Due to the strict convexity of G˜ θ in its first argument, the first-order optimality conditions gives Hyperparameters. In our numerical experiments, we vary only three hyperparameters to adjust the complexity of the architecture. As described in section 5 we randomly sample the depth, K, the feature width, w, and the context width, u, from the values in Table 2.

Efficient Neural Network Approaches: Partially Convex Potential Maps (PCP-Map) for Conditional OT

MDL-based Model Learning: How It Works

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Sebastien Ferr ´e, Univ Rennes, CNRS, Inria, IRISA Campus de Beaulieu, 35042 Rennes, France and ferre@irisa.fr. Table of Links Abstract & Introduction Abstraction and Reasoning Corpus (ARC) Related Work Object-centric Models for ARC Grids MDL-based Model Learning Evaluation Conclusion and Perspectives, References & Supplementary Materials Abstract. The Abstraction and Reasoning Corpus (ARC) is a challenging benchmark, introduced to foster AI research towards human-level intelligence. It is a collection of unique tasks about generating colored grids, specified by a few examples only. In contrast to the transformation-based programs of existing work, we introduce object-centric models that are in line with the natural programs produced by humans. Our models can not only perform predictions, but also provide joint descriptions for input/output pairs. The Minimum Description Length (MDL) principle is used to efficiently search the large model space. A diverse range of tasks are solved, and the learned models are similar to the natural programs. We demonstrate the generality of our approach by applying it to a different domain. 1 Introduction Artificial Intelligence (AI) has made impressive progress in the past decade at specific tasks, sometimes achieving super-human performance: e.g., image recognition [15], board games [21], natural language processing [6]. However, AI still misses the generality and flexibility of human intelligence to adapt to novel tasks with little training. To foster AI research beyond narrow generalization [11], F. Chollet [4,5] introduced a measure of intelligence that values skill-acquisition efficiency over skill performance , i.e. the amount of prior knowledge and experience that an agent needs to reach a reasonably good level at a range of tasks (e.g., board games) matters more than its absolute performance at any specific task (e.g., chess). Chollet also introduced the Abstraction and Reasoning Corpus (ARC) benchmark in the form of a psychometric test to measure and compare the intelligence of humans and machines alike. ARC is a collection of tasks that consist in learning how to transform an input colored grid into an output colored grid, given only a few examples. It is a very challenging benchmark. While humans can solve more than 80% of the tasks [14], the winner of a Kaggle contest 1 could only solve 20% of the tasks (with a lot of hard-coded primitives and brute-force search), and the winner of the more recent ARCathon’22 contest 2 could only solve 6% of the tasks. The existing published approaches [10,3,23,2], and also the Kaggle winner, tackle the ARC challenge as a program synthesis problem, where a program is a composition of primitive transformations, and learning is done by searching the large program space. In contrast, psychological studies have shown that the natural programs produced by humans to solve ARC tasks are object-centric, and more declarative than procedural [14,1]. When asked to verbalize instructions on how to solve a task, participants typically first describe what to expect in the input grid, and then how to generate the output grid based on the elements found in the input grid. We make two contributions w.r.t. existing work: 1. object-centric models that enable to both parse and generate grids in terms of object patterns and computations on those objects; 2. an efficient search of object-centric models based on the Minimum Description Length (MDL) principle [20]. A model for an ARC task combines two grid models, one for the input grid, and another for the output grid. This closely matches the structure of natural programs. Compared to the transformation-based programs that can only predict an output grid from an input grid, our models can also provide a joint description for a pair of grids. They can also create new pairs of grids, although this is not evaluated in this paper. They could also in principle be adapted to tasks based on a single grid or on sequences of grids. All of this is possible because grid models can be used both for parsing a grid and for generating a grid. The MDL principle comes from information theory, and says that “the model that best describes the data is the model that compress them the more” [20,12]. It has for instance been applied to pattern mining [22,9]. The MDL principle is used at two levels: (a) to choose the best parses of a grid according to a grid model, and (b) to efficiently search the large model space by incrementally building more and more accurate models. MDL at level (a) is essential because the segmentation of a grid into objects is taskdependent and has to be learned along with the definition of the output objects as a function of the input objects. The two contributions support each other because existing search strategies could not handle the large number of elementary components of our grid models, and because the transformation-based programs are not suitable to the incremental evaluation required by MDL-based search. We report promising results based on grid models that are still far from covering all knowledge priors assumed by ARC. Correct models are found for 96/400 varied training tasks with a 60s time budget. Many of those are similar to the natural programs produced by humans. Moreover, we demonstrate the generality of our approach by applying it to the automatic filling of spreadsheet columns [13], where inputs and outputs are rows of strings instead of grids. The paper is organized as follows. Section 2 presents the ARC benchmark, and a running example task. Section 3 discusses related work. Section 4 defines our objectcentric models, and Section 5 explains how to learn them with the MDL principle. Section 6 reports on experimental results, comparing with existing approaches. This paper is available on Arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. Authors: (1) Sebastien Ferr ´e, Univ Rennes, CNRS, Inria, IRISA Campus de Beaulieu, 35042 Rennes, France and ferre@irisa.fr. This paper is available on arxiv under CC 4.0 license. Authors: Authors: (1) Sebastien Ferr ´e, Univ Rennes, CNRS, Inria, IRISA Campus de Beaulieu, 35042 Rennes, France and ferre@irisa.fr. Table of Links Abstract & Introduction Abstract & Introduction Abstraction and Reasoning Corpus (ARC) Abstraction and Reasoning Corpus (ARC) Related Work Related Work Object-centric Models for ARC Grids Object-centric Models for ARC Grids MDL-based Model Learning MDL-based Model Learning Evaluation Evaluation Conclusion and Perspectives, References & Supplementary Materials Conclusion and Perspectives, References & Supplementary Materials Abstract. The Abstraction and Reasoning Corpus (ARC) is a challenging benchmark, introduced to foster AI research towards human-level intelligence. It is a collection of unique tasks about generating colored grids, specified by a few examples only. In contrast to the transformation-based programs of existing work, we introduce object-centric models that are in line with the natural programs produced by humans. Our models can not only perform predictions, but also provide joint descriptions for input/output pairs. The Minimum Description Length (MDL) principle is used to efficiently search the large model space. A diverse range of tasks are solved, and the learned models are similar to the natural programs. We demonstrate the generality of our approach by applying it to a different domain. 1 Introduction Artificial Intelligence (AI) has made impressive progress in the past decade at specific tasks, sometimes achieving super-human performance: e.g., image recognition [15], board games [21], natural language processing [6]. However, AI still misses the generality and flexibility of human intelligence to adapt to novel tasks with little training. To foster AI research beyond narrow generalization [11], F. Chollet [4,5] introduced a measure of intelligence that values skill-acquisition efficiency over skill performance , i.e. the amount of prior knowledge and experience that an agent needs to reach a reasonably good level at a range of tasks (e.g., board games) matters more than its absolute performance at any specific task (e.g., chess). Chollet also introduced the Abstraction and Reasoning Corpus (ARC) benchmark in the form of a psychometric test to measure and compare the intelligence of humans and machines alike. ARC is a collection of tasks that consist in learning how to transform an input colored grid into an output colored grid, given only a few examples. It is a very challenging benchmark. While humans can solve more than 80% of the tasks [14], the winner of a Kaggle contest 1 could only solve 20% of the tasks (with a lot of hard-coded primitives and brute-force search), and the winner of the more recent ARCathon’22 contest 2 could only solve 6% of the tasks. The existing published approaches [10,3,23,2], and also the Kaggle winner, tackle the ARC challenge as a program synthesis problem, where a program is a composition of primitive transformations, and learning is done by searching the large program space. In contrast, psychological studies have shown that the natural programs produced by humans to solve ARC tasks are object-centric, and more declarative than procedural [14,1]. When asked to verbalize instructions on how to solve a task, participants typically first describe what to expect in the input grid, and then how to generate the output grid based on the elements found in the input grid. We make two contributions w.r.t. existing work: 1. object-centric models that enable to both parse and generate grids in terms of object patterns and computations on those objects; 2. an efficient search of object-centric models based on the Minimum Description Length (MDL) principle [20]. A model for an ARC task combines two grid models, one for the input grid, and another for the output grid. This closely matches the structure of natural programs. Compared to the transformation-based programs that can only predict an output grid from an input grid, our models can also provide a joint description for a pair of grids. They can also create new pairs of grids, although this is not evaluated in this paper. They could also in principle be adapted to tasks based on a single grid or on sequences of grids. All of this is possible because grid models can be used both for parsing a grid and for generating a grid. The MDL principle comes from information theory, and says that “the model that best describes the data is the model that compress them the more” [20,12]. It has for instance been applied to pattern mining [22,9]. The MDL principle is used at two levels: (a) to choose the best parses of a grid according to a grid model, and (b) to efficiently search the large model space by incrementally building more and more accurate models. MDL at level (a) is essential because the segmentation of a grid into objects is taskdependent and has to be learned along with the definition of the output objects as a function of the input objects. The two contributions support each other because existing search strategies could not handle the large number of elementary components of our grid models, and because the transformation-based programs are not suitable to the incremental evaluation required by MDL-based search. We report promising results based on grid models that are still far from covering all knowledge priors assumed by ARC. Correct models are found for 96/400 varied training tasks with a 60s time budget. Many of those are similar to the natural programs produced by humans. Moreover, we demonstrate the generality of our approach by applying it to the automatic filling of spreadsheet columns [13], where inputs and outputs are rows of strings instead of grids. The paper is organized as follows. Section 2 presents the ARC benchmark, and a running example task. Section 3 discusses related work. Section 4 defines our objectcentric models, and Section 5 explains how to learn them with the MDL principle. Section 6 reports on experimental results, comparing with existing approaches. This paper is available on Arxiv under CC 4.0 license. This paper is available on Arxiv under CC 4.0 license. Arxiv

The Abstraction and Reasoning Corpus (ARC): Why It's Important

Authors: (1) Muzhaffar Hazman, University of Galway, Ireland; (2) Susan McKeever, Technological University Dublin, Ireland; (3) Josephine Griffith, University of Galway, Ireland. Table of Links Abstract and Introduction Related Works Methodology Results Limitations and Future Works Conclusion, Acknowledgments, and References A Hyperparameters and Settings B Metric: Weighted F1-Score C Architectural Details D Performance Benchmarking E Contingency Table: Baseline vs. Text-STILT E Contingency Table: Baseline vs. Text-STILT Table 8 shows the contingency table – as one would prepare for a McNemar’s Test between two classifiers (McNemar, 1947) – between the model trained with Text-STILT on 60% Memes and Baseline trained on 100% Memes available which had the most similar performance. While the two models performed similarly in terms of Weighted F1- scores, Text-STILT correctly classified a notable number of memes that Baseline did not and vice versa. Examples of such memes are discussed in Section 4.1. Furthermore, approximately 40% of memes in the testing set were incorrectly classified by both models. This suggests that these memes convey sentiment in a way that cannot be reliably predicted by either approach. This paper is under CC 4.0 license. available on arxiv

Contingency Table: Baseline vs. Text-STILT

Authors: (1) Philipp Straubinger, University of Passau, Passau, Germany and this author contributed equally to this research; (2) Tommaso Fulcini, Politecnico di Torino,  Torino, Italy and this author contributed equally to this research; (3) Gordon Fraser, University of Passau, Passau, Germany; (4) Marco Torchiano, Politecnico di Torino, Torino, Italy. Table of Links Abstract and Introduction Background and Related Work Implementation Experiment Results Conclusions, Acknowledgement, and References ABSTRACT This paper investigates the integration and assessment of IntelliGame, a gamification plugin initially designed for Java development, within the realm of JavaScript unit testing. We aim to verify the generalizability of IntelliGame to JavaScript development and to provide valuable insights into the experiment’s design. For this, we first customize IntelliGame for JavaScript, and then conduct a controlled experiment involving 152 participants utilizing the Jest testing framework, and finally examine its influence on testing behavior and the overall developer experience. The findings from this study provide valuable insights for improving JavaScript testing methodologies through the incorporation of gamification. CCS CONCEPTS • Software and its engineering → Software testing and debugging; Integrated and visual development environments. KEYWORDS Gamification, IDE, IntelliJ, Software Testing ACM Reference Format: Philipp Straubinger, Tommaso Fulcini, Gordon Fraser, and Marco Torchiano. 2024. IntelliGame in Action: An Experience Report on Gamifying JavaScript Unit Tests. In 2024 First IDE Workshop (IDE ’24), April 20, 2024, Lisbon, Portugal. ACM, New York, NY, USA, 5 pages. https://doi.org/10.1145/3643796. 3648466 1 INTRODUCTION IntelliGame [8] is a plugin for the popular IntelliJ IDEA, which enables gamification covering various aspects of software testing, such as test execution, coverage evaluation, debugging, and test refactoring. It introduces 27 different achievements, each with incremental levels, providing positive feedback when developers exhibit commendable testing behavior. IntelliGame provides real-time analysis of developer interactions, notifications, and a user interface within IntelliJ to display achievements and progress. In its original evaluation [8], we conducted a controlled experiment with 49 participants to study the impact of IntelliGame on testing behavior. The experiment involved a Java programming task with participants divided into different groups, aiming to assess the impact of the gamified environment on testing behavior, test suites, achievement levels, code functionality, and the overall developer experience. The results showed clearly that IntelliGame had a positive effect on testing behavior. Participants using IntelliGame wrote more tests, achieved higher code coverage and mutation scores, ran tests more frequently, and implemented functionality earlier compared to the control group. The impact on resulting test suites was substantial, and achievements correlated positively with various testing metrics. The goal of this paper is to build upon these promising results achieved with IntelliGame for Java environments, and to assess whether these outcomes can be replicated on a larger scale and in a different context. In particular, the extension of the positive impact of achievements in the IDE to various programming languages remains unexplored. Thus we shifted the focus of this study to JavaScript (JS), recognized as the most used language as of 2023 [6]. With the growing importance of JS, there is an increasing demand to instruct and practice language-specific testing activities. Expanding beyond the previous sample, this study engages a broader audience with multiple tasks, allowing for more extensive code development and testing. This expansion, coupled with the shift to a different programming language, considers both generalizability and effectiveness beyond a short-term context. The objective of this paper is to provide an in-depth description of the experiment’s design, including challenges encountered, lessons learned, and recommendations for future iterations of JSbased experiments. Notably, there is a gap in the existing literature regarding programming experiments centered on an open-source JavaScript target [9], emphasizing a fully replicable and systematic approach. In Section 2 we provide some background about the topic of the paper with also some examples of related works, inSection 3 the implementation approach for the JavaScript version of IntelliGame is described, while Section 4 shows the selection process for the project as a subject is meticulously detailed, shedding light on the rationale behind the choice. This paper is under CC BY-SA 4.0 DEED license. available on arxiv

Stories, News & Mentions about SECTION4

Section4’s Stories on HackerNoon