Experiment 27

experiment27

Authors:
(1) Nora Schneider, Computer Science Department, ETH Zurich, Zurich, Switzerland (nschneide@student.ethz.ch);
(2) Shirin Goshtasbpour, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (shirin.goshtasbpour@inf.ethz.ch);
(3) Fernando Perez-Cruz, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (fernando.perezcruz@sdsc.ethz.ch). Table of Links Abstract and 1 Introduction 2 Background 2.1 Data Augmentation 2.2 Anchor Regression 3 Anchor Data Augmentation 3.1 Comparison to C-Mixup and 3.2 Preserving nonlinear data structure 3.3 Algorithm 4 Experiments and 4.1 Linear synthetic data 4.2 Housing nonlinear regression 4.3 In-distribution Generalization 4.4 Out-of-distribution Robustness 5 Conclusion, Broader Impact, and References A Additional information for Anchor Data Augmentation B Experiments 4 Experiments We experimentally investigate and compare the performance of ADA. First, we use ADA in an in-distribution setting for a linear regression problem (Section 4.1), in which we show that even in this case, ADA provides improved performance in the low data regime. Second, in Section 4.2, we apply ADA and C-Mixup to the California and Boston Housing datasets as we increase the number of training samples. In the last two subsections, we replicate the in-distribution generalization (Section 4.3) and the out-of-distribution Robustness (Section 4.4) from the C-Mixup paper [49]. In [49] the authors further assess a task generalization experiment. However, the corresponding code was not publicly provided, and a comparison could not be easily made. 4.1 Linear synthetic data Using synthetic linear data, we investigate if ADA can improve model performance in an overparameterized setting compared to C-Mixup, vanilla augmentation, or classical expected risk minimization (ERM). Additionally, we analyze the sensitivity of our approach to the choice of γ and the number of augmentations. Data: The generated data follows a standard linear structure Models and Comparisons: We investigate and compare the impact of ADA using two different models with varying complexity: a linear Ridge regression and a multilayer perceptron (MLP) with one hidden layer and 10 units with ReLU activation. Using an MLP with more hidden layers shows similar results (see Appendix B.1 for details). For the Ridge regression model, we increase the dataset by a factor of 10 by sampling from the respective augmentation methods and subsequently compute the regression estimators. In contrast, for the MLP, we implement the augmentation methods on a minibatch level. Specifically, we incorporate vanilla augmentation by adding Gaussian noise to each batch, apply C-Mixup after sampling from the beta distribution in each batch, and finally, apply ADA after sampling from the defined gamma values in each batch. In summary, even in the simplest of cases, in which we should not expect gains from ADA (or C-Mixup), these data augmentation strategies provide gains in performance when the number of training examples is not sufficient to achieve the error floor. This paper is available on arxiv under CC0 1.0 DEED license. Authors: (1) Nora Schneider, Computer Science Department, ETH Zurich, Zurich, Switzerland (nschneide@student.ethz.ch); (2) Shirin Goshtasbpour, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (shirin.goshtasbpour@inf.ethz.ch); (3) Fernando Perez-Cruz, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (fernando.perezcruz@sdsc.ethz.ch). Authors: Authors: (1) Nora Schneider, Computer Science Department, ETH Zurich, Zurich, Switzerland (nschneide@student.ethz.ch); (2) Shirin Goshtasbpour, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (shirin.goshtasbpour@inf.ethz.ch); (3) Fernando Perez-Cruz, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (fernando.perezcruz@sdsc.ethz.ch). Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Background 2.1 Data Augmentation 2.1 Data Augmentation 2.2 Anchor Regression 2.2 Anchor Regression 3 Anchor Data Augmentation 3 Anchor Data Augmentation 3.1 Comparison to C-Mixup and 3.2 Preserving nonlinear data structure 3.1 Comparison to C-Mixup and 3.2 Preserving nonlinear data structure 3.3 Algorithm 3.3 Algorithm 4 Experiments and 4.1 Linear synthetic data 4 Experiments and 4.1 Linear synthetic data 4.2 Housing nonlinear regression 4.2 Housing nonlinear regression 4.3 In-distribution Generalization 4.3 In-distribution Generalization 4.4 Out-of-distribution Robustness 4.4 Out-of-distribution Robustness 5 Conclusion, Broader Impact, and References 5 Conclusion, Broader Impact, and References A Additional information for Anchor Data Augmentation A Additional information for Anchor Data Augmentation B Experiments B Experiments 4 Experiments We experimentally investigate and compare the performance of ADA. First, we use ADA in an in-distribution setting for a linear regression problem (Section 4.1), in which we show that even in this case, ADA provides improved performance in the low data regime. Second, in Section 4.2, we apply ADA and C-Mixup to the California and Boston Housing datasets as we increase the number of training samples. In the last two subsections, we replicate the in-distribution generalization (Section 4.3) and the out-of-distribution Robustness (Section 4.4) from the C-Mixup paper [49]. In [49] the authors further assess a task generalization experiment. However, the corresponding code was not publicly provided, and a comparison could not be easily made. 4.1 Linear synthetic data Using synthetic linear data, we investigate if ADA can improve model performance in an overparameterized setting compared to C-Mixup, vanilla augmentation, or classical expected risk minimization (ERM). Additionally, we analyze the sensitivity of our approach to the choice of γ and the number of augmentations. Data : The generated data follows a standard linear structure Data Models and Comparisons: We investigate and compare the impact of ADA using two different models with varying complexity: a linear Ridge regression and a multilayer perceptron (MLP) with one hidden layer and 10 units with ReLU activation. Using an MLP with more hidden layers shows similar results (see Appendix B.1 for details). Models and Comparisons: For the Ridge regression model, we increase the dataset by a factor of 10 by sampling from the respective augmentation methods and subsequently compute the regression estimators. In contrast, for the MLP, we implement the augmentation methods on a minibatch level. Specifically, we incorporate vanilla augmentation by adding Gaussian noise to each batch, apply C-Mixup after sampling from the beta distribution in each batch, and finally, apply ADA after sampling from the defined gamma values in each batch. In summary, even in the simplest of cases, in which we should not expect gains from ADA (or C-Mixup), these data augmentation strategies provide gains in performance when the number of training examples is not sufficient to achieve the error floor. This paper is available on arxiv under CC0 1.0 DEED license. This paper is available on arxiv under CC0 1.0 DEED license. available on arxiv

Evaluating ADA: Experimental Results on Linear and Housing Datasets

Authors:
(1) Shih-Tang Su, University of Michigan, Ann Arbor (shihtang@umich.edu);
(2) Vijay G. Subramanian, University of Michigan, Ann Arbor and (vgsubram@umich.edu);
(3) Grant Schoenebeck, University of Michigan, Ann Arbor (schoeneb@umich.edu). Table of Links Abstract and 1. Introduction 2. Problem Formulation 2.1 Model of Binary-Outcome Experiments in Two-Phase Trials 3 Binary-outcome Experiments in Two-phase Trials and 3.1 Experiments with screenings 3.2 Assumptions and induced strategies 3.3 Constraints given by phase-II experiments 3.4 Persuasion ratio and the optimal signaling structure 3.5 Comparison with classical Bayesian persuasion strategies 4 Binary-outcome Experiments in Multi-phase trials and 4.1 Model of binary-outcome experiments in multi-phase trials 4.2 Determined versus sender-designed experiments 4.3 Multi-phase model and classical Bayesian persuasion and References 4.2 Determined versus sender-designed experiments Given the model, the sender can manipulate the phase-K interim belief only when designing an experiment at phase K − 1. If an experiment at phase K − 1 is determined, then the phase-K interim belief is a function of interim belief at phase K − 1. Therefore, figuring out how these two types of experiments, determined and sender-designed experiments, will influence every given phase’s interim belief is the key to solving for the optimal signaling strategy. We start by noting that if the posterior belief at a leaf node is given, then the receiver’s action is determined - he will take the action with the highest posterior probability unless there is a tie, in which case he is indifferent and will follow the sender’s recommendation. Therefore, we can use backward iteration and the principle of optimality to determine the optimal signaling. We start by considering the last phase’s experiments when the sender can design them. Experiments in phase N − 1 In the second-last phase, results in Section 3.4 describe a sender-designed experiment’s role in the optimal signaling strategy: pick the strategy on the frontier of all persuasion-ratio curves. However, if the experiment is determined in the second-last phase, an additional constraint on the interim belief between the second-last phase and the last phase is enforced. That is to say, the set of (feasible) strategies will shrink. Fortunately, after enforcing the constraints, the process of searching for the optimal signaling strategy under a determined experiment is the same as the sender-designed experiment, i.e., pick the strategy in the frontier of all persuasion ratio curves. Therefore, at each possible branch of phase N − 1, we can plot an optimal persuasion ratio curve capturing the sender’s optimal signaling strategy at phase N − 1 and phase N. Non-binary Outcome Experiments When the experiments have non-binary outcomes, the same approach derived above works with an increased number of phases (if complexity is not an issue). For general non-binary experiments, see the proof of Lemma 6 in [23] for a detailed construction from non-binary to binary experiments. This paper is available on arxiv under CC 4.0 license. Authors: (1) Shih-Tang Su, University of Michigan, Ann Arbor (shihtang@umich.edu); (2) Vijay G. Subramanian, University of Michigan, Ann Arbor and (vgsubram@umich.edu); (3) Grant Schoenebeck, University of Michigan, Ann Arbor (schoeneb@umich.edu). Authors: Authors: (1) Shih-Tang Su, University of Michigan, Ann Arbor (shihtang@umich.edu); (2) Vijay G. Subramanian, University of Michigan, Ann Arbor and (vgsubram@umich.edu); (3) Grant Schoenebeck, University of Michigan, Ann Arbor (schoeneb@umich.edu). Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Problem Formulation 2. Problem Formulation 2.1 Model of Binary-Outcome Experiments in Two-Phase Trials 2.1 Model of Binary-Outcome Experiments in Two-Phase Trials 3 Binary-outcome Experiments in Two-phase Trials and 3.1 Experiments with screenings 3 Binary-outcome Experiments in Two-phase Trials and 3.1 Experiments with screenings 3.2 Assumptions and induced strategies 3.2 Assumptions and induced strategies 3.3 Constraints given by phase-II experiments 3.3 Constraints given by phase-II experiments 3.4 Persuasion ratio and the optimal signaling structure 3.4 Persuasion ratio and the optimal signaling structure 3.5 Comparison with classical Bayesian persuasion strategies 3.5 Comparison with classical Bayesian persuasion strategies 4 Binary-outcome Experiments in Multi-phase trials and 4.1 Model of binary-outcome experiments in multi-phase trials 4 Binary-outcome Experiments in Multi-phase trials and 4.1 Model of binary-outcome experiments in multi-phase trials 4.2 Determined versus sender-designed experiments 4.2 Determined versus sender-designed experiments 4.3 Multi-phase model and classical Bayesian persuasion and References 4.3 Multi-phase model and classical Bayesian persuasion and References 4.2 Determined versus sender-designed experiments Given the model, the sender can manipulate the phase-K interim belief only when designing an experiment at phase K − 1. If an experiment at phase K − 1 is determined, then the phase-K interim belief is a function of interim belief at phase K − 1. Therefore, figuring out how these two types of experiments, determined and sender-designed experiments, will influence every given phase’s interim belief is the key to solving for the optimal signaling strategy. We start by noting that if the posterior belief at a leaf node is given, then the receiver’s action is determined - he will take the action with the highest posterior probability unless there is a tie, in which case he is indifferent and will follow the sender’s recommendation. Therefore, we can use backward iteration and the principle of optimality to determine the optimal signaling. We start by considering the last phase’s experiments when the sender can design them. Experiments in phase N − 1 In the second-last phase, results in Section 3.4 describe a sender-designed experiment’s role in the optimal signaling strategy: pick the strategy on the frontier of all persuasion-ratio curves. However, if the experiment is determined in the second-last phase, an additional constraint on the interim belief between the second-last phase and the last phase is enforced. That is to say, the set of (feasible) strategies will shrink. Fortunately, after enforcing the constraints, the process of searching for the optimal signaling strategy under a determined experiment is the same as the sender-designed experiment, i.e., pick the strategy in the frontier of all persuasion ratio curves. Therefore, at each possible branch of phase N − 1, we can plot an optimal persuasion ratio curve capturing the sender’s optimal signaling strategy at phase N − 1 and phase N . Experiments in phase N − 1 N N N N Non-binary Outcome Experiments When the experiments have non-binary outcomes, the same approach derived above works with an increased number of phases (if complexity is not an issue). Non-binary Outcome Experiments For general non-binary experiments, see the proof of Lemma 6 in [23] for a detailed construction from non-binary to binary experiments. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

Optimizing Signaling Strategies with Sender-Designed Experiments in Multi-phase Trials

Authors:
(1) Shih-Tang Su, University of Michigan, Ann Arbor (shihtang@umich.edu);
(2) Vijay G. Subramanian, University of Michigan, Ann Arbor and (vgsubram@umich.edu);
(3) Grant Schoenebeck, University of Michigan, Ann Arbor (schoeneb@umich.edu). Table of Links Abstract and 1. Introduction 2. Problem Formulation 2.1 Model of Binary-Outcome Experiments in Two-Phase Trials 3 Binary-outcome Experiments in Two-phase Trials and 3.1 Experiments with screenings 3.2 Assumptions and induced strategies 3.3 Constraints given by phase-II experiments 3.4 Persuasion ratio and the optimal signaling structure 3.5 Comparison with classical Bayesian persuasion strategies 4 Binary-outcome Experiments in Multi-phase trials and 4.1 Model of binary-outcome experiments in multi-phase trials 4.2 Determined versus sender-designed experiments 4.3 Multi-phase model and classical Bayesian persuasion and References 3 Binary-outcome Experiments in Two-phase Trials In this section the sender’s optimization problem presented in (2) Section 2.1, is solved starting with the simplest non-trivial case. There are only two phases in the trial studied here, and from this we will develop more insight into how different types of experiments (determined versus sender-designed) influence the optimal signaling strategy of the sender. To be more specific, we will analyze how two determined experiments (in phase II) and one sender-designed experiment (in phase I) will impact the sender’s optimal signaling strategy. Before we present the general case, we discuss a subset class of two-phase trials that are similar to single-phase trials. In this class of two-phase trials, in one of the phase-II experiments, called a trivial experiment, the outcome distribution is independent of the true state. Trivial experiments [2], also called (Blackwell) non-informative experiments in some literature, are frequently used as benchmarks to compare the agents’ expected utility change under different signaling schemes/mechanisms, e.g., [22,20,21]. This two-phase model with a trivial experiment tries to capture real-world problems with one actual (and costly) experiment, e.g., clinical trials, venture capital investments, or space missions. Since the experiment is costly, a screening procedure is provided to decide whether it is worth conducting the experiment. We will then analyze the optimal signaling strategy in the general scenario, where both experiments in phase II are non-trivial. 3.1 Experiments with screenings We start by analyzing the sender’s optimal strategy (signaling structure) in a simple scenario where there is one non-trivial experiment conducted in phase II. The sender’s authority on choosing the probability pair (p1, p2) controls the screening process. To avoid any ambiguity, we first define what a trivial experiment is. When a trivial experiment (in phase II) is conducted, the posterior belief of the state stays the same as the interim belief derived in (1). When there exists a trivial experiment in the two phase-II trial options, then Lemma 1 states that the sender and the receiver’s expected utility under the optimal signaling strategy is the same as in the (single-phase) classical Bayesian persuasion problem. Lemma 1. When the state space is binary, both sender and receiver’s expected utilities are the same in the following two Bayesian persuasion schemes under each scheme’s optimal signaling strategy: Bayesian persuasion in a single-phase trial,


Bayesian persuasion in a two-phase trial with a sender-designed phase-I experiment and a trivial experiment in phase II. In the single-trial classical Bayesian persuasion setting, the optimal signaling strategy only mixes the two possible states in one outcome (e.g., when the prosecutor claims the suspect is guilty). On the other outcome, the sender reveals the true state with probability one (e.g., when the prosecutor says the suspect is innocent). When there is a trivial experiment in phase II, the other experiment (supposing that it will be conducted at outcome ωB) will be rendered defunct by the sender’s choice of experiments in phase I. This phenomenon occurs because the sender can always choose to reveal the true state when the non-trivial experiment is to be conducted, i.e., by setting P(θ1|EB) = 1 or P(θ2|EB) = 1; and the classical Bayesian persuasion strategy can be replicated. In essence, having a trivial experiment in the phase-II trial does not constrain the sender. This paper is available on arxiv under CC 4.0 license. Authors: (1) Shih-Tang Su, University of Michigan, Ann Arbor (shihtang@umich.edu); (2) Vijay G. Subramanian, University of Michigan, Ann Arbor and (vgsubram@umich.edu); (3) Grant Schoenebeck, University of Michigan, Ann Arbor (schoeneb@umich.edu). Authors: Authors: (1) Shih-Tang Su, University of Michigan, Ann Arbor (shihtang@umich.edu); (2) Vijay G. Subramanian, University of Michigan, Ann Arbor and (vgsubram@umich.edu); (3) Grant Schoenebeck, University of Michigan, Ann Arbor (schoeneb@umich.edu). Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Problem Formulation 2. Problem Formulation 2.1 Model of Binary-Outcome Experiments in Two-Phase Trials 2.1 Model of Binary-Outcome Experiments in Two-Phase Trials 3 Binary-outcome Experiments in Two-phase Trials and 3.1 Experiments with screenings 3 Binary-outcome Experiments in Two-phase Trials and 3.1 Experiments with screenings 3.2 Assumptions and induced strategies 3.2 Assumptions and induced strategies 3.3 Constraints given by phase-II experiments 3.3 Constraints given by phase-II experiments 3.4 Persuasion ratio and the optimal signaling structure 3.4 Persuasion ratio and the optimal signaling structure 3.5 Comparison with classical Bayesian persuasion strategies 3.5 Comparison with classical Bayesian persuasion strategies 4 Binary-outcome Experiments in Multi-phase trials and 4.1 Model of binary-outcome experiments in multi-phase trials 4 Binary-outcome Experiments in Multi-phase trials and 4.1 Model of binary-outcome experiments in multi-phase trials 4.2 Determined versus sender-designed experiments 4.2 Determined versus sender-designed experiments 4.3 Multi-phase model and classical Bayesian persuasion and References 4.3 Multi-phase model and classical Bayesian persuasion and References 3 Binary-outcome Experiments in Two-phase Trials In this section the sender’s optimization problem presented in (2) Section 2.1, is solved starting with the simplest non-trivial case. There are only two phases in the trial studied here, and from this we will develop more insight into how different types of experiments (determined versus sender-designed) influence the optimal signaling strategy of the sender. To be more specific, we will analyze how two determined experiments (in phase II) and one sender-designed experiment (in phase I) will impact the sender’s optimal signaling strategy. Before we present the general case, we discuss a subset class of two-phase trials that are similar to single-phase trials. In this class of two-phase trials, in one of the phase-II experiments, called a trivial experiment, the outcome distribution is independent of the true state. Trivial experiments [2], also called (Blackwell) non-informative experiments in some literature, are frequently used as benchmarks to compare the agents’ expected utility change under different signaling schemes/mechanisms, e.g., [22,20,21]. This two-phase model with a trivial experiment tries to capture real-world problems with one actual (and costly) experiment, e.g., clinical trials, venture capital investments, or space missions. Since the experiment is costly, a screening procedure is provided to decide whether it is worth conducting the experiment. We will then analyze the optimal signaling strategy in the general scenario, where both experiments in phase II are non-trivial. 3.1 Experiments with screenings We start by analyzing the sender’s optimal strategy (signaling structure) in a simple scenario where there is one non-trivial experiment conducted in phase II. The sender’s authority on choosing the probability pair (p1, p2) controls the screening process. To avoid any ambiguity, we first define what a trivial experiment is. When a trivial experiment (in phase II) is conducted, the posterior belief of the state stays the same as the interim belief derived in (1). When there exists a trivial experiment in the two phase-II trial options, then Lemma 1 states that the sender and the receiver’s expected utility under the optimal signaling strategy is the same as in the (single-phase) classical Bayesian persuasion problem. Lemma 1. When the state space is binary, both sender and receiver’s expected utilities are the same in the following two Bayesian persuasion schemes under each scheme’s optimal signaling strategy: Lemma 1. When the state space is binary, both sender and receiver’s expected utilities are the same in the following two Bayesian persuasion schemes under each scheme’s optimal signaling strategy: Bayesian persuasion in a single-phase trial, Bayesian persuasion in a two-phase trial with a sender-designed phase-I experiment and a trivial experiment in phase II. Bayesian persuasion in a single-phase trial, Bayesian persuasion in a single-phase trial, Bayesian persuasion in a single-phase trial, Bayesian persuasion in a two-phase trial with a sender-designed phase-I experiment and a trivial experiment in phase II. Bayesian persuasion in a two-phase trial with a sender-designed phase-I experiment and a trivial experiment in phase II. Bayesian persuasion in a two-phase trial with a sender-designed phase-I experiment and a trivial experiment in phase II. In the single-trial classical Bayesian persuasion setting, the optimal signaling strategy only mixes the two possible states in one outcome (e.g., when the prosecutor claims the suspect is guilty). On the other outcome, the sender reveals the true state with probability one (e.g., when the prosecutor says the suspect is innocent). When there is a trivial experiment in phase II, the other experiment (supposing that it will be conducted at outcome ωB) will be rendered defunct by the sender’s choice of experiments in phase I. This phenomenon occurs because the sender can always choose to reveal the true state when the non-trivial experiment is to be conducted, i.e., by setting P(θ1|EB) = 1 or P(θ2|EB) = 1; and the classical Bayesian persuasion strategy can be replicated. In essence, having a trivial experiment in the phase-II trial does not constrain the sender. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

Analyzing Optimal Signaling with Binary-outcome Experiments in Two-phase Trials

Ontleed optimale sein met binêre uitkoms-eksperimente in twee-fase proewe

Análisis de la señalización óptima con experimentos de resultados binarios en ensayos de dos fases

تجزیه و تحلیل سیگنال دهی بهینه با آزمایش های باینری نتیجه در آزمایشات دو فازی

Analiza optimalnog signaliziranja s eksperimentima s binarnim ishodom u dvofaznim ispitivanjima

2段階試験におけるバイナリ結果実験による最適なシグナル伝達の分析

ოპტიმალური სიგნალიზაციის ანალიზი ორფაზიან ცდებში ორფაზიანი ექსპერიმენტებით

Екі фазалық сынақтардағы екілік нәтиже эксперименттерімен оңтайлы сигналды талдау

Go sekaseka Matshwao a Mabotse ka Diteko tša dipoelo tša Binary ka Ditekong tša Megato ye Mebedi

Анализ оптимальной сигнализации с помощью экспериментов с бинарным результатом в двухфазных испытаниях

Gusesengura Ibimenyetso Byiza hamwe na Binary-ibisubizo Ubushakashatsi Mubice bibiri

Анализа оптималне сигнализације са експериментима са бинарним исходом у двофазним испитивањима

دو فیز ٹرائلز میں بائنری نتائج کے تجربات کے ساتھ بہترین سگنلنگ کا تجزیہ کرنا

Is ByteDance’s Lumi an AI Experiment Disguised as an AI Creator Community?

ByteDance's Lumi የ AI ሙከራ እንደ AI ፈጣሪ ማህበረሰብ ተደብቋል?

¿Es Lumi de ByteDance un experimento de IA disfrazado de comunidad de creadores de IA?

Èske Lumi ByteDance a se yon eksperyans AI degize kòm yon kominote kreyatè AI?

Արդյո՞ք ByteDance-ի Lumi-ն AI փորձ է, որը քողարկված է որպես AI Ստեղծողների համայնք:

ByteDance の Lumi は AI クリエイター コミュニティを装った AI 実験なのか?

ByteDance의 Lumi는 AI 크리에이터 커뮤니티로 위장한 AI 실험일까요?

ByteDance’s Lumi ezali Expérience ya AI oyo e déguisé lokola Communauté ya Créateur ya AI?

Lumi ຂອງ ByteDance ເປັນການທົດລອງ AI ປອມຕົວເປັນຊຸມຊົນຜູ້ສ້າງ AI ບໍ?

Adakah Lumi ByteDance merupakan Eksperimen AI yang Menyamar sebagai Komuniti Pencipta AI?

Je, Lumi ya ByteDance ni Jaribio la AI Limefichwa kama Jumuiya ya Watayarishi wa AI?

Оё Lumi byteDance як озмоиши AI аст, ки ҳамчун ҷомеаи Офарандаи AI пинҳон карда шудааст?

Ngaba i-ByteDance's Lumi liLingelo le-AI eliguqulelwe njengoluntu lwabadali be-AI?

Google's Gemini AI: A Thought Experiment in Deception

Google ukan Gemini AI: ¿Mä amuyt’äw yant’äwix sallqjañataki?

La inteligencia artificial Gemini de Google: ¿un experimento mental sobre el engaño?

AI של Gemini של גוגל: ניסוי מחשבה בהונאה?

Google-ի Gemini AI. Մտքի փորձ խաբեության մեջ:

Google の Gemini AI: 欺瞞の思考実験?

구글의 제미니 AI: 기만에 대한 사고실험?

Google-ийн Gemini AI: Хууран мэхлэлтийн талаархи бодлын туршилт уу?

Gemini AI Google: Eksperimen Pemikiran dalam Penipuan?

د ګوګل جیمني AI: په فریب کې د فکر تجربه؟

Google Gemini AI: Tijaabo Fikirka ah ee Khiyaanada?

Gemini AI ya Google: Jaribio la Mawazo katika Udanganyifu?

Trí tuệ nhân tạo Gemini của Google: Một thí nghiệm tư duy về sự lừa dối?

Authors:
(1) Junwei Su, Department of Computer Science, the University of Hong Kong and jwsu@cs.hku.hk;
(2) Chuan Wu, Department of Computer Science, the University of Hong Kong and cwu@cs.hku.hk. Table of Links Abstract and 1 Introduction 2 Related Work 3 Framework 4 Main Results 5 A Case Study on Shortest-Path Distance 6 Conclusion and Discussion, and References 7 Proof of Theorem 1 8 Proof of Theorem 2 9 Procedure for Solving Eq. (6) 10 Additional Experiments Details and Results 11 Other Potential Applications 10 Additional Experiments Details and Results In this appendix, we provide additional experimental results and include a detailed set-up of the experiments for reproducibility. 10.1 Hardware and Software All the experiments of this paper are conducted on the following machine CPU: two Intel Xeon Gold 6230 2.1G, 20C/40T, 10.4GT/s, 27.5M Cache, Turbo, HT (125W) DDR4-2933 GPU: four NVIDIA Tesla V100 SXM2 32G GPU Accelerator for NV Link Memory: 256GB (8 x 32GB) RDIMM, 3200MT/s, Dual Rank OS: Ubuntu 18.04LTS 10.2 Training Hyper-parameters and Detailed Dataset Description Table 3 gives a more detailed description of the statistic of datasets used in the experiment as well as the hyper-parameters used for training. We keep the hyper-parameters the same across different models. To keep the experiment consistent, we do not employ randomized regularization techniques such as dropout. To ensure a fair comparison, we make sure each experiment is trained till convergence. 10.3 Additional Experiment Results In addition to the datasets we present in the main text, we also conduct experiments on Citeseer, Cora Full and OGBN-arxiv. The results of these experiments show that the theoretical structures proved in this paper are indeed consistent across different datasets. Next, we present additional experimental results on these datasets for validating the main theorems of our papers. Graph Distance and Performance We present additional results with Citeseer, CoraFull, and OGBN-Arxiv on the relation between the graph distance (shortest-path distance). For these experiments, we follow the same setting as the one presented in Fig. 4. We compute the average prediction accuracy in each group of vertices (corresponding to vertices at each hop of the neighbourhood of the training set). ’0 hop’ indicates the training set. We observe that GNNs tend to predict better on the vertices that are closer to the training set. The results are presented in Fig. 6, 7 and 8. Subgroup Embedding Distance and Accuracy We present additional results with Citeseer, CoraFull, and OGBN-Arxiv on that GNNs preserve distances (with a small distortion rate) locally, by evaluating the relation between the graph distances of vertices to the training set and the embedding distances (Euclidean distance) of vertex representations to the representations of the training set, i.e., the relation between d(v, D) and d(MθD (v),MθD (D)). In this experiment, we follow the same setting as the one in Fig. 3 and train the GNN models on the default training set provided in the dataset. We extract vertices within the 5-hop neighbourhood of the training set and group the vertices based on their distances to the training set. We then compute the average embedding distance of each group to the training set in the embedding space. Based on Def. 3, if an embedding has low distortion, we should expect to see linear relation with slop being the scaling factor. As shown in Fig. 9,Fig. 10, and Fig. 11 there exists a strong correlation between the graph distance and the embedding distance. The relation is near linear and the relative orders of distances are mostly preserved within the first few hops. This implies that the distortion α is indeed small with different GNNs and that the common aggregation mechanism of GNNs can well capture local graph structure information (i.e., graph distances). It also indicates that the premise on the distortion rate, which we used to prove our theoretical results, can be commonly satisfied in practice. 10.4 Graph Distance and Performance This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Junwei Su, Department of Computer Science, the University of Hong Kong and jwsu@cs.hku.hk; (2) Chuan Wu, Department of Computer Science, the University of Hong Kong and cwu@cs.hku.hk. Authors: Authors: (1) Junwei Su, Department of Computer Science, the University of Hong Kong and jwsu@cs.hku.hk; (2) Chuan Wu, Department of Computer Science, the University of Hong Kong and cwu@cs.hku.hk. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Related Work 2 Related Work 3 Framework 3 Framework 4 Main Results 4 Main Results 5 A Case Study on Shortest-Path Distance 5 A Case Study on Shortest-Path Distance 6 Conclusion and Discussion, and References 6 Conclusion and Discussion, and References 7 Proof of Theorem 1 7 Proof of Theorem 1 8 Proof of Theorem 2 8 Proof of Theorem 2 9 Procedure for Solving Eq. (6) 9 Procedure for Solving Eq. (6) 10 Additional Experiments Details and Results 10 Additional Experiments Details and Results 11 Other Potential Applications 11 Other Potential Applications 10 Additional Experiments Details and Results In this appendix, we provide additional experimental results and include a detailed set-up of the experiments for reproducibility. 10.1 Hardware and Software All the experiments of this paper are conducted on the following machine CPU: two Intel Xeon Gold 6230 2.1G, 20C/40T, 10.4GT/s, 27.5M Cache, Turbo, HT (125W) DDR4-2933 GPU: four NVIDIA Tesla V100 SXM2 32G GPU Accelerator for NV Link Memory: 256GB (8 x 32GB) RDIMM, 3200MT/s, Dual Rank OS: Ubuntu 18.04LTS 10.2 Training Hyper-parameters and Detailed Dataset Description Table 3 gives a more detailed description of the statistic of datasets used in the experiment as well as the hyper-parameters used for training. We keep the hyper-parameters the same across different models. To keep the experiment consistent, we do not employ randomized regularization techniques such as dropout. To ensure a fair comparison, we make sure each experiment is trained till convergence. 10.3 Additional Experiment Results In addition to the datasets we present in the main text, we also conduct experiments on Citeseer, Cora Full and OGBN-arxiv. The results of these experiments show that the theoretical structures proved in this paper are indeed consistent across different datasets. Next, we present additional experimental results on these datasets for validating the main theorems of our papers. Graph Distance and Performance We present additional results with Citeseer, CoraFull, and OGBN-Arxiv on the relation between the graph distance (shortest-path distance). For these experiments, we follow the same setting as the one presented in Fig. 4. We compute the average prediction accuracy in each group of vertices (corresponding to vertices at each hop of the neighbourhood of the training set). ’0 hop’ indicates the training set. We observe that GNNs tend to predict better on the vertices that are closer to the training set. The results are presented in Fig. 6, 7 and 8. Graph Distance and Performance Subgroup Embedding Distance and Accuracy We present additional results with Citeseer, CoraFull, and OGBN-Arxiv on that GNNs preserve distances (with a small distortion rate) locally, by evaluating the relation between the graph distances of vertices to the training set and the embedding distances (Euclidean distance) of vertex representations to the representations of the training set, i.e., the relation between d(v, D) and d(MθD (v),MθD (D)). In this experiment, we follow the same setting as the one in Fig. 3 and train the GNN models on the default training set provided in the dataset. We extract vertices within the 5-hop neighbourhood of the training set and group the vertices based on their distances to the training set. We then compute the average embedding distance of each group to the training set in the embedding space. Based on Def. 3, if an embedding has low distortion, we should expect to see linear relation with slop being the scaling factor. As shown in Fig. 9,Fig. 10, and Fig. 11 there exists a strong correlation between the graph distance and the embedding distance. The relation is near linear and the relative orders of distances are mostly preserved within the first few hops. This implies that the distortion α is indeed small with different GNNs and that the common aggregation mechanism of GNNs can well capture local graph structure information (i.e., graph distances). It also indicates that the premise on the distortion rate, which we used to prove our theoretical results, can be commonly satisfied in practice. Subgroup Embedding Distance and Accuracy 10.4 Graph Distance and Performance 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

Comprehensive Overview of GNN Experiments: Hardware, Hyperparameters, and Findings

Authors:
(1) Divyanshu Kumar, Enkrypt AI;
(2) Anurakt Kumar, Enkrypt AI;
(3) Sahil Agarwa, Enkrypt AI;
(4) Prashanth Harshangi, Enkrypt AI. Table of Links Abstract and 1 Introduction 2 Problem Formulation and Experiments 3 Experiment Set-up & Results 4 Conclusion and References A. Appendix 3 EXPERIMENT SET-UP & RESULTS The following section outlines the testing environments utilized for various LLMs. The LLMs are tested under three different downstream tasks: (1) fine-tuning, (2) quantization, and (3) guardrails. They are chosen to cover most of the LLM practical use cases and applications in the industry and academia. For TAP configuration, as mentioned before, we use GPT-3.5-turbo as the attack model, and GPT-4-turbo as the judge model. We employ Anyscale endpoints, the OpenAI API, and HuggingFace for our target model. Further information regarding the model and its sources is available in appendix A.1. The details about different settings are mentioned below : • Fine-tuning: Fine-tuning LLMs on different tasks increases their effectiveness in completing the tasks as it incorporates the specialized domain knowledge needed; these can include SQL code generation, chat, and more. We compare the jailbreaking vulnerability of foundational models compared to their fine tune versions. This helps us understand the role of fine-tuning in increasing or decreasing the vulnerability of LLMs and the strategies to mitigate this risk Weyssow et al. (2023). We use foundation models such as Llama2, Mistral, and MPT-7B and their fine-tuned versions such as CodeLlama, SQLCoder, Dolphin, and Intel Neural Chat. The details of the models and versions are specified in the appendix. From the table 1, we empirically conclude that fine-tuned models lose their safety alignment and are jailbroken quite easily compared to the foundational models. • Quantization: Many models require huge computational resources during training, finetuning, and even during inference Hu et al. (2021). Quantization is one of the most popular ways to reduce the computational burden, but it comes at the cost of the numerical precision of model parameters. The quantized models we evaluate below use GPT-Generated Unified Format (GGUF) for quantization, which involves scaling down model weights (stored in 16-bit floating point numbers) to save computational resources at the cost of numerical precision of the model parameters. Kashiwamura et al. (2024) Gorsline et al. (2021) Xiao et al.(2023) Hu et al. (2021) Dettmers et al. (2023). The table 2 demonstrates that quantization of the model renders it susceptible to vulnerabilities. • Guardrails: Guardrails act as a line of defence against LLM attacks. Their primary function is to meticulously filter out prompts that could potentially lead to harmful or malicious outcomes, preventing such prompts from reaching the LLM as an instruction Rebedea et al. (2023). This proactive approach not only promotes the safe utilization of LLMs but also facilitates their optimal performance, thereby maximizing their potential benefits in various domains Kumar et al. (2023) Wei et al. (2023) Zhou et al. (2024). We use our proprietary jailbreak attack detector derived from Deberta-V3 models, trained on harmful prompts generated to jailbreak LLMs. You can reach out to the authors to get more details on this model. From table 3, we observe that the introduction of guardrails as a pre-step has a significant effect and can mitigate jailbreaking attempts by a considerable margin. The results from tables 1, 2, and 3 conclusively show the vulnerability of LLMs post fine-tuning and quantization, and effectiveness of external guardrails in mitigating this vulnerability. The detailed experimental results can be found in appendix A.2. Additionally, this section provides a comprehensive analysis of the experiments conducted, offering insights into the various aspects of the research findings. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Divyanshu Kumar, Enkrypt AI; (2) Anurakt Kumar, Enkrypt AI; (3) Sahil Agarwa, Enkrypt AI; (4) Prashanth Harshangi, Enkrypt AI. Authors: Authors: (1) Divyanshu Kumar, Enkrypt AI; (2) Anurakt Kumar, Enkrypt AI; (3) Sahil Agarwa, Enkrypt AI; (4) Prashanth Harshangi, Enkrypt AI. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Problem Formulation and Experiments 2 Problem Formulation and Experiments 3 Experiment Set-up & Results 3 Experiment Set-up & Results 4 Conclusion and References 4 Conclusion and References A. Appendix A. Appendix 3 EXPERIMENT SET-UP & RESULTS The following section outlines the testing environments utilized for various LLMs. The LLMs are tested under three different downstream tasks: (1) fine-tuning, (2) quantization, and (3) guardrails. They are chosen to cover most of the LLM practical use cases and applications in the industry and academia. For TAP configuration, as mentioned before, we use GPT-3.5-turbo as the attack model, and GPT-4-turbo as the judge model. We employ Anyscale endpoints, the OpenAI API, and HuggingFace for our target model. Further information regarding the model and its sources is available in appendix A.1. The details about different settings are mentioned below : • Fine-tuning: Fine-tuning LLMs on different tasks increases their effectiveness in completing the tasks as it incorporates the specialized domain knowledge needed; these can include SQL code generation, chat, and more. We compare the jailbreaking vulnerability of foundational models compared to their fine tune versions. This helps us understand the role of fine-tuning in increasing or decreasing the vulnerability of LLMs and the strategies to mitigate this risk Weyssow et al. (2023). We use foundation models such as Llama2, Mistral, and MPT-7B and their fine-tuned versions such as CodeLlama, SQLCoder, Dolphin, and Intel Neural Chat. The details of the models and versions are specified in the appendix. From the table 1, we empirically conclude that fine-tuned models lose their safety alignment and are jailbroken quite easily compared to the foundational models. • Fine-tuning: • Quantization : Many models require huge computational resources during training, finetuning, and even during inference Hu et al. (2021). Quantization is one of the most popular ways to reduce the computational burden, but it comes at the cost of the numerical precision of model parameters. The quantized models we evaluate below use GPT-Generated Unified Format (GGUF) for quantization, which involves scaling down model weights (stored in 16-bit floating point numbers) to save computational resources at the cost of numerical precision of the model parameters. Kashiwamura et al. (2024) Gorsline et al. (2021) Xiao et al.(2023) Hu et al. (2021) Dettmers et al. (2023). The table 2 demonstrates that quantization of the model renders it susceptible to vulnerabilities. Quantization • Guardrails : Guardrails act as a line of defence against LLM attacks. Their primary function is to meticulously filter out prompts that could potentially lead to harmful or malicious outcomes, preventing such prompts from reaching the LLM as an instruction Rebedea et al. (2023). This proactive approach not only promotes the safe utilization of LLMs but also facilitates their optimal performance, thereby maximizing their potential benefits in various domains Kumar et al. (2023) Wei et al. (2023) Zhou et al. (2024). We use our proprietary jailbreak attack detector derived from Deberta-V3 models, trained on harmful prompts generated to jailbreak LLMs. You can reach out to the authors to get more details on this model. From table 3, we observe that the introduction of guardrails as a pre-step has a significant effect and can mitigate jailbreaking attempts by a considerable margin. Guardrails The results from tables 1, 2, and 3 conclusively show the vulnerability of LLMs post fine-tuning and quantization, and effectiveness of external guardrails in mitigating this vulnerability. The detailed experimental results can be found in appendix A.2. Additionally, this section provides a comprehensive analysis of the experiments conducted, offering insights into the various aspects of the research findings. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Increased LLM Vulnerabilities from Fine-tuning and Quantization: Experiment Set-up & Results

Authors:
(1) Divyanshu Kumar, Enkrypt AI;
(2) Anurakt Kumar, Enkrypt AI;
(3) Sahil Agarwa, Enkrypt AI;
(4) Prashanth Harshangi, Enkrypt AI. Table of Links Abstract and 1 Introduction 2 Problem Formulation and Experiments 3 Experiment Set-up & Results 4 Conclusion and References A. Appendix 2 PROBLEM FORMULATION AND EXPERIMENTS We want to understand the role played by fine-tuning, quantization, and guardrails on LLM’s vulnerability towards jailbreaking attacks. We create a pipeline to test for jailbreaking of LLMs which undergo these further processes before deployment. As mentioned in the introduction, we attack the LLM via the TAP algorithm using the AdvBench subset. We use a subset of AdvBenchAndy Zou (2023). It contains 50 prompts asking for harmful information across 32 categories. The evaluation results, along with the complete system information, are then logged. The overall flow is shown in the figure 2. This process continues for multiple iterations, taking into account the stochastic nature associated with LLMs. The complete experiment pipeline is shown in Figure 2. TAP Mehrotra et al. (2023) is used as the jailbreaking method, as it is currently the state-of-the-art, black-box, and automatic method which generates prompts with semantic meaning to jailbreak LLMs. TAP algorithm uses an attacker LLM A, which sends a prompt P to the target LLM T. The response of the target LLM R along with the prompt P are fed into the evaluator LLM JUDGE, which determines if the prompt is on-topic or off-topic. If the prompt is off-topic, it is removed, thereby eliminating the tree of bad attack prompts it was going to generate. Otherwise, if the prompt is on-topic, it receives a score between 0-10 by the JUDGE LLM. This prompt is used to generate a new attack prompt using breadth-first search. This process continues till the LLM is jailbroken or a specified number of iterations are exhausted. Now describing our guardrail against jailbreaking prompts, we use our in-house Deberta-V3 model, which has been trained to detect jailbreaking prompts. It acts as an input filter to ensure that only sanitized prompts are received by the LLM. If the input prompt is filtered out by the guardrail or fails to jailbreak the LLM, then the TAP algorithm generates a new prompt considering the initial prompt and response. The new attacking prompt is then again passed through the guardrail. This process is repeated till a jailbreaking prompt is found or a pre-specified number of iterations are exhausted. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Divyanshu Kumar, Enkrypt AI; (2) Anurakt Kumar, Enkrypt AI; (3) Sahil Agarwa, Enkrypt AI; (4) Prashanth Harshangi, Enkrypt AI. Authors: Authors: (1) Divyanshu Kumar, Enkrypt AI; (2) Anurakt Kumar, Enkrypt AI; (3) Sahil Agarwa, Enkrypt AI; (4) Prashanth Harshangi, Enkrypt AI. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Problem Formulation and Experiments 2 Problem Formulation and Experiments 3 Experiment Set-up & Results 3 Experiment Set-up & Results 4 Conclusion and References 4 Conclusion and References A. Appendix A. Appendix 2 PROBLEM FORMULATION AND EXPERIMENTS We want to understand the role played by fine-tuning, quantization, and guardrails on LLM’s vulnerability towards jailbreaking attacks. We create a pipeline to test for jailbreaking of LLMs which undergo these further processes before deployment. As mentioned in the introduction, we attack the LLM via the TAP algorithm using the AdvBench subset. We use a subset of AdvBenchAndy Zou (2023). It contains 50 prompts asking for harmful information across 32 categories. The evaluation results, along with the complete system information, are then logged. The overall flow is shown in the figure 2. This process continues for multiple iterations, taking into account the stochastic nature associated with LLMs. The complete experiment pipeline is shown in Figure 2. TAP Mehrotra et al. (2023) is used as the jailbreaking method, as it is currently the state-of-the-art, black-box, and automatic method which generates prompts with semantic meaning to jailbreak LLMs. TAP algorithm uses an attacker LLM A, which sends a prompt P to the target LLM T. The response of the target LLM R along with the prompt P are fed into the evaluator LLM JUDGE, which determines if the prompt is on-topic or off-topic. If the prompt is off-topic, it is removed, thereby eliminating the tree of bad attack prompts it was going to generate. Otherwise, if the prompt is on-topic, it receives a score between 0-10 by the JUDGE LLM. This prompt is used to generate a new attack prompt using breadth-first search. This process continues till the LLM is jailbroken or a specified number of iterations are exhausted. Now describing our guardrail against jailbreaking prompts, we use our in-house Deberta-V3 model, which has been trained to detect jailbreaking prompts. It acts as an input filter to ensure that only sanitized prompts are received by the LLM. If the input prompt is filtered out by the guardrail or fails to jailbreak the LLM, then the TAP algorithm generates a new prompt considering the initial prompt and response. The new attacking prompt is then again passed through the guardrail. This process is repeated till a jailbreaking prompt is found or a pre-specified number of iterations are exhausted. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Increased LLM Vulnerabilities from Fine-tuning and Quantization: Problem Formulation and Experiments

Authors:
(1) Jianhui Pang, from the University of Macau, and work was done when Jianhui Pang and Fanghua Ye were interning at Tencent AI Lab (nlp2ct.pangjh3@gmail.com);
(2) Fanghua Ye, University College London, and work was done when Jianhui Pang and Fanghua Ye were interning at Tencent AI Lab (fanghua.ye.19@ucl.ac.uk);
(3) Derek F. Wong, University of Macau;
(4) Longyue Wang, Tencent AI Lab, and corresponding author. Table of Links Abstract and 1 Introduction 2 Related Work 3 Anchor-based Large Language Models 3.1 Background 3.2 Anchor-based Self-Attention Networks 3.3 Anchor-based Inference 4 Experiments and 4.1 Our Implementation 4.2 Data and Training Procedure 4.3 Evaluation 5 Results 6 Analysis 7 Conclusion, Limitations, Ethics Statement, and References A More Experimental Results B Data Settings A More Experimental Results A.1 Testing Acceleration Ratio to Full-Caching Method In Section 5, we report the testing acceleration ratio following the setting of Wang et al. (2023), comparing the time difference between caching and non-caching inference. Although our method reduces the keys/values caches, enabling smaller space for prefix information storage and improving testing time up to ×3.5, we are still curious about whether it would enhance inference efficiency if conventional methods use full caches that save all keys/values of prefix tokens. As a supplement to Table 1, we present the testing acceleration ratio between anchor-caching and full-caching inference in Table 3. The acceleration ratios for AnLLMEP-AnSAN and AnLLM-AC-AnSAN achieve the highest improvements observed in tasks such as HS, SCIQ, and BoolQ. The average acceleration ratios for AnLLM-EP-AnSAN and AnLLM-ACAnSAN are 1.03. These results demonstrate that our anchor-based caching method can enhance inference efficiency even when compared to conventional methods that save all keys/values of prefix tokens. These results suggest that our anchor-based caching approach, which saves only the keys/values caches of anchor tokens, can effectively accelerate the inference process for the lengthy prefix texts. A.2 Model Scalability Assessment To examine the scalability of our approach, we extend the AnLLM-AC model to 13B and assess its performance on eight question-answering benchmarks using the same evaluation strategy as previously mentioned. In comparison to the 7B AnLLM models in Table 1, Results in Table 4 indicate that as the model size expands, the AnLLM-AC model achieves accuracies of 67.5% and 70.0% for 0-shot and 5-shot testing, respectively, resulting in up to a 2.4% improvement. Moreover, by incorporating anchor-based attention, the AnLLM-AC-AnSAN model achieves an average accuracy of 69.5%, signifying a 2.0% increase. The performance enhancement underscores the effectiveness of our methods in accommodating larger model capacities. The consistent improvements observed in the AnLLMAC model across various scenarios highlight its robustness and adaptability. Furthermore, the increased performance of the AnLLM-AC-AnSAN model, facilitated by anchor-based attention, emphasizes the potential of our approaches in optimizing LLMs. Collectively, these findings point to promising avenues for future research aimed at maximizing the utility and efficiency of AnLLM. A.3 Case Study in Real-Time Inference To elaborate on the optimization of keys/values caches by AnLLM-EP and AnLLM-AC during real-time inference, we reference examples from the translation task in Section 6.2. As per Table 5, AnLLM-EP and AnLLM-AC use "endpoints" (".") and "" tokens as anchor tokens, respectively. During inference, both models employ auto-regressive generation, creating outputs token-by-token. Upon generating an anchor token (as per Line 16, Algorithm 1), the Reduction function (defined in Line 1) is activated, preserving relevant caches and eliminating others. As a result, the Keys/Values Cache lengths are reduced to roughly the sequence length, averaging around 50 for AnLLM-EP and 35 for AnLLM-AC, as shown in Table 2. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Jianhui Pang, from the University of Macau, and work was done when Jianhui Pang and Fanghua Ye were interning at Tencent AI Lab (nlp2ct.pangjh3@gmail.com); (2) Fanghua Ye, University College London, and work was done when Jianhui Pang and Fanghua Ye were interning at Tencent AI Lab (fanghua.ye.19@ucl.ac.uk); (3) Derek F. Wong, University of Macau; (4) Longyue Wang, Tencent AI Lab, and corresponding author. Authors: Authors: (1) Jianhui Pang, from the University of Macau, and work was done when Jianhui Pang and Fanghua Ye were interning at Tencent AI Lab (nlp2ct.pangjh3@gmail.com); (2) Fanghua Ye, University College London, and work was done when Jianhui Pang and Fanghua Ye were interning at Tencent AI Lab (fanghua.ye.19@ucl.ac.uk); (3) Derek F. Wong, University of Macau; (4) Longyue Wang, Tencent AI Lab, and corresponding author. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Related Work 2 Related Work 3 Anchor-based Large Language Models 3.1 Background 3.1 Background 3.2 Anchor-based Self-Attention Networks 3.2 Anchor-based Self-Attention Networks 3.3 Anchor-based Inference 3.3 Anchor-based Inference 4 Experiments and 4.1 Our Implementation 4 Experiments and 4.1 Our Implementation 4.2 Data and Training Procedure 4.2 Data and Training Procedure 4.3 Evaluation 4.3 Evaluation 5 Results 5 Results 6 Analysis 6 Analysis 7 Conclusion, Limitations, Ethics Statement, and References 7 Conclusion, Limitations, Ethics Statement, and References A More Experimental Results A More Experimental Results B Data Settings B Data Settings A More Experimental Results A.1 Testing Acceleration Ratio to Full-Caching Method In Section 5, we report the testing acceleration ratio following the setting of Wang et al. (2023), comparing the time difference between caching and non-caching inference. Although our method reduces the keys/values caches, enabling smaller space for prefix information storage and improving testing time up to ×3.5, we are still curious about whether it would enhance inference efficiency if conventional methods use full caches that save all keys/values of prefix tokens. As a supplement to Table 1, we present the testing acceleration ratio between anchor-caching and full-caching inference in Table 3. The acceleration ratios for AnLLMEP-AnSAN and AnLLM-AC-AnSAN achieve the highest improvements observed in tasks such as HS, SCIQ, and BoolQ. The average acceleration ratios for AnLLM-EP-AnSAN and AnLLM-ACAnSAN are 1.03. These results demonstrate that our anchor-based caching method can enhance inference efficiency even when compared to conventional methods that save all keys/values of prefix tokens. These results suggest that our anchor-based caching approach, which saves only the keys/values caches of anchor tokens, can effectively accelerate the inference process for the lengthy prefix texts. A.2 Model Scalability Assessment To examine the scalability of our approach, we extend the AnLLM-AC model to 13B and assess its performance on eight question-answering benchmarks using the same evaluation strategy as previously mentioned. In comparison to the 7B AnLLM models in Table 1, Results in Table 4 indicate that as the model size expands, the AnLLM-AC model achieves accuracies of 67.5% and 70.0% for 0-shot and 5-shot testing, respectively, resulting in up to a 2.4% improvement. Moreover, by incorporating anchor-based attention, the AnLLM-AC-AnSAN model achieves an average accuracy of 69.5%, signifying a 2.0% increase. The performance enhancement underscores the effectiveness of our methods in accommodating larger model capacities. The consistent improvements observed in the AnLLMAC model across various scenarios highlight its robustness and adaptability. Furthermore, the increased performance of the AnLLM-AC-AnSAN model, facilitated by anchor-based attention, emphasizes the potential of our approaches in optimizing LLMs. Collectively, these findings point to promising avenues for future research aimed at maximizing the utility and efficiency of AnLLM. A.3 Case Study in Real-Time Inference To elaborate on the optimization of keys/values caches by AnLLM-EP and AnLLM-AC during real-time inference, we reference examples from the translation task in Section 6.2. As per Table 5, AnLLM-EP and AnLLM-AC use "endpoints" (".") and "" tokens as anchor tokens, respectively. During inference, both models employ auto-regressive generation, creating outputs token-by-token. Upon generating an anchor token (as per Line 16, Algorithm 1), the Reduction function (defined in Line 1) is activated, preserving relevant caches and eliminating others. As a result, the Keys/Values Cache lengths are reduced to roughly the sequence length, averaging around 50 for AnLLM-EP and 35 for AnLLM-AC, as shown in Table 2. 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

Anchor-based Large Language Models: More Experimental Results

Authors:
(1) Jianhui Pang, from the University of Macau, and work was done when Jianhui Pang and Fanghua Ye were interning at Tencent AI Lab (nlp2ct.pangjh3@gmail.com);
(2) Fanghua Ye, University College London, and work was done when Jianhui Pang and Fanghua Ye were interning at Tencent AI Lab (fanghua.ye.19@ucl.ac.uk);
(3) Derek F. Wong, University of Macau;
(4) Longyue Wang, Tencent AI Lab, and corresponding author. Table of Links Abstract and 1 Introduction 2 Related Work 3 Anchor-based Large Language Models 3.1 Background 3.2 Anchor-based Self-Attention Networks 3.3 Anchor-based Inference 4 Experiments and 4.1 Our Implementation 4.2 Data and Training Procedure 4.3 Evaluation 5 Results 6 Analysis 7 Conclusion, Limitations, Ethics Statement, and References A More Experimental Results B Data Settings 4 Experiments In this section, we first detail AnLLM’s implementation, then outline the training procedure and model perplexity. Finally, we introduce the evaluation datasets and metrics. 4.1 Our Implementation Llama2-7b (Touvron et al., 2023b) is adopted as the base model in our experiments, which is an open-source and English-centric LLM. In accordance with the principles outlined in Section 3, we present our implementations here. The crux is to identify which tokens in a sequence can be considered anchor tokens. In light of this, we describe two implementation strategies: one employs the endpoints directly, and the other involves appending a new token at the end of a sequence to serve as the anchor token. The details are as follows: • AnLLM-EP. This approach uses punctuation marks within the sequence as anchor tokens. Punctuation marks, such as commas, periods, and question marks, are viewed as semantic boundaries within a sequence. As such, they can serve as anchor tokens in AnLLM. In our experiments of AnLLM-EP, we use the endpoint in English as the anchor tokens. • AnLLM-AC. This strategy entails the introduction of a new token to act as the sequence anchor. In our implementation, we designate as the new token and initialize its embedding using the mean value of the embedding matrix. For training data, we use the sentence tokenizer from the NLTK package to split texts into sentences, appending at the end of each sentence as the anchor token.[1] During inference, tokens can be flexibly added to the text based on user requirements, such as adding one anchor for each demonstration, allowing for flexible and controllable sequence compression. This paper is available on arxiv under CC BY 4.0 DEED license. [1] https://www.nltk.org/api/nltk.tokenize.punkt. html Authors: (1) Jianhui Pang, from the University of Macau, and work was done when Jianhui Pang and Fanghua Ye were interning at Tencent AI Lab (nlp2ct.pangjh3@gmail.com); (2) Fanghua Ye, University College London, and work was done when Jianhui Pang and Fanghua Ye were interning at Tencent AI Lab (fanghua.ye.19@ucl.ac.uk); (3) Derek F. Wong, University of Macau; (4) Longyue Wang, Tencent AI Lab, and corresponding author. Authors: Authors: (1) Jianhui Pang, from the University of Macau, and work was done when Jianhui Pang and Fanghua Ye were interning at Tencent AI Lab (nlp2ct.pangjh3@gmail.com); (2) Fanghua Ye, University College London, and work was done when Jianhui Pang and Fanghua Ye were interning at Tencent AI Lab (fanghua.ye.19@ucl.ac.uk); (3) Derek F. Wong, University of Macau; (4) Longyue Wang, Tencent AI Lab, and corresponding author. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Related Work 2 Related Work 3 Anchor-based Large Language Models 3.1 Background 3.1 Background 3.2 Anchor-based Self-Attention Networks 3.2 Anchor-based Self-Attention Networks 3.3 Anchor-based Inference 3.3 Anchor-based Inference 4 Experiments and 4.1 Our Implementation 4 Experiments and 4.1 Our Implementation 4.2 Data and Training Procedure 4.2 Data and Training Procedure 4.3 Evaluation 4.3 Evaluation 5 Results 5 Results 6 Analysis 6 Analysis 7 Conclusion, Limitations, Ethics Statement, and References 7 Conclusion, Limitations, Ethics Statement, and References A More Experimental Results A More Experimental Results B Data Settings B Data Settings 4 Experiments In this section, we first detail AnLLM’s implementation, then outline the training procedure and model perplexity. Finally, we introduce the evaluation datasets and metrics. 4.1 Our Implementation Llama2-7b (Touvron et al., 2023b) is adopted as the base model in our experiments, which is an open-source and English-centric LLM. In accordance with the principles outlined in Section 3, we present our implementations here. The crux is to identify which tokens in a sequence can be considered anchor tokens. In light of this, we describe two implementation strategies: one employs the endpoints directly, and the other involves appending a new token at the end of a sequence to serve as the anchor token. The details are as follows: • AnLLM-EP. This approach uses punctuation marks within the sequence as anchor tokens. Punctuation marks, such as commas, periods, and question marks, are viewed as semantic boundaries within a sequence. As such, they can serve as anchor tokens in AnLLM. In our experiments of AnLLM-EP, we use the endpoint in English as the anchor tokens. • AnLLM-EP. • AnLLM-AC. This strategy entails the introduction of a new token to act as the sequence anchor. In our implementation, we designate as the new token and initialize its embedding using the mean value of the embedding matrix. For training data, we use the sentence tokenizer from the NLTK package to split texts into sentences, appending at the end of each sentence as the anchor token.[1] During inference, tokens can be flexibly added to the text based on user requirements, such as adding one anchor for each demonstration, allowing for flexible and controllable sequence compression. • AnLLM-AC. 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 [1] https://www.nltk.org/api/nltk.tokenize.punkt. html

Anchor-based Large Language Models: Experiments and Implementation

Stories, News & Mentions about EXPERIMENT27

Experiment27’s Stories on HackerNoon