NEXT Insurance, a digital insurtech specialising in small businesses, has announced it has entered into strategic partnerships with US primary insurer

NEXT Insurance enters partnerships with Allstate and Allianz X

Last valued at $4 billion, Next Insurance raised $265 million in funding from Allstate and Allianz X, it said Wednesday. Why it matters: Insurtech was the most-funded sector within fintech in the third quarter, per FT Partners data. Strategic investors are helping play a role in that.

SMB insurance company Next raises $265 million

Allstate and Allianz have made a $265 million strategic equity investment in small business-focused insuretech startup Next Insurance.

Allstate, Allianz invest $265 million in Next Insurance in a big bet on insuretech

NEXT Insurance, a leading technology-first small business insurer, today announced its strategic partnership with Allstate that will combine the companies' expertise in small business insurance to develop and offer innovative and unique new products to a large underserved market,

NEXT Insurance Enters Partnerships with Allstate and Allianz X, Supported by $265 Million Strategic Investment

The small businesses digital insurance company has entered a strategic partnership with Allstate and Allianz X. Israeli insurtech company Next Insurance today announced its strategic partnership with Allstate and Allianz X,

Israeli co Next Insurance raises $265m

An in-depth review of American Family car insurance, including coverage options, pricing, customer satisfaction and how AmFam compares to other insurance companies.

American Family car insurance review 2023

Allstate Corporation, a leading insurance company based in the United States, has made a significant impact on the global insurance market through its growth and expansion strategies. With a strong focus on customer satisfaction and innovative products ...

The Growth and Expansion of Allstate Corporation in the Global Insurance Market

Allstate Corporation is a well-known insurance company that has been providing coverage to individuals and businesses for over 90 years. Understanding the history and evolution of this company is crucial to understanding its unique approach to insurance.

Understanding Allstate Corporation’s Unique Approach to Insurance

That follows similar decisions by Allstate and State Farm, earlier this year. The most recent move is by smaller insurers. Merastar Insurance Company; Unitrin Auto and Home Insurance Company; Unitrin Direct Property and Casualty Company and Kemper ...

4 more insurers are leaving California after Allstate and State Farm flee wildfire risk

Moves follow a large withdrawal of companies from Florida, whose CFO called them “the Bud Light of the insurance industry."

Allstate

allstate

A pair of newly public documents filed with a California administrative law judge show experts accusing the company of systematically overcharging customers it believed to be the most loyal around $1 billion over the past decade. This practice of charging higher premiums to customers an insurance company suspects are unlikely to defect to a competitor is termed “price optimization” and was the subject of a 2020 Markup that found Allstate was attempting to use a new pricing algorithm for auto insurance in Maryland that would have unfairly targeted its highest-paying customers—and that the algorithm had been approved in several other states. investigation Nearly every state, including California, bars insurers from setting car insurance rates on factors apart from the actual risk the drivers pose. Insurance regulators in 18 states and Washington, D.C., have explicitly declared price optimization illegal. The new documents, which were initially filed in late October by the California Department of Insurance and Consumer Watchdog, a consumer advocacy group allowed by the state to intervene in the case and provide expertise, consist of written testimony from insurance industry experts who examined how Allstate set its prices. They allege that Allstate was engaging in price optimization by giving smaller than appropriate discounts to the least-price-sensitive among its customers with clean driving records who held multiple policies with the company or who had several decades of driving experience. Edward Cimini Jr., a senior casualty actuary with the California Department of Insurance, he reviewed internal Allstate documents, documents the company submitted to the state describing its auto insurance pricing plan, and depositions of company employees and found that Allstate gave smaller discounts to drivers with more than 39 years of experience, a group he said is unlikely to shop around. said “Since Allstate’s selections were not based on underlying costs, the final rates that Allstate charged these policyholders were actuarily unsound and unfairly discriminatory,” he said. Allan Schwartz, an actuarial consultant hired by Consumer Watchdog to review Allstate’s pricing practices, that Allstate overcharged California drivers who were owed discounts “about $1 billion.” estimated Since Allstate’s selections were not based on underlying costs, the final rates that Allstate charged … were actuarily unsound and unfairly discriminatory. Edward Cimini Jr., California Department of Insurance “Those policyholders were known by Allstate to have a lower elasticity of demand and were more likely to renew with Allstate even though they were charged premiums in excess of those based upon an actuarially sound estimate of the cost of risk transfer,” he said. The company denies the allegations. “Allstate does not employ, and has never employed, price optimization in determining premiums in California because Allstate does not take into account an individual’s or class’s willingness to pay a higher premium relative to other individuals or classes,” Allstate spokesperson Ben Corey wrote in an email to The Markup. In its court , Allstate points out that in 2011 the California Department of Insurance reviewed and approved the 2011 plan without highlighting the issues now raised in the long-running class action that triggered this hearing. filings The company was sued in 2015 over alleged price optimization practices in California. Allstate moved to have the case thrown out, arguing in part that it wasn’t a matter for civil courts but rather for the state department of insurance. In 2016, the United States District Court for the Northern District of California put the case on hold and referred it to the authority of the Commissioner of the California Department of Insurance for its opinion, which triggered the administrative law proceeding. Last week, the parties agreed to voluntary mediation. enter Two years ago, The Markup published an revealing that a new method of calculating rates for car insurance customers that Allstate was trying to implement across the country attempted to charge higher rates to customers who were already paying a lot for their car insurance—essentially creating a “suckers list” of big spenders and squeezing even more money out of them. investigation The investigation was based on details the insurer provided to regulators in Maryland as part of its 2014 filing there that revealed the current and proposed rate for policyholders in the state. Using statistical regressions, we found that the rating factor, called CGR, would charge more—and severely limit discounts—to big spenders. Maryland regulators rejected Allstate’s plan over price optimization concerns. We found Allstate was using similar algorithms in 10 other states, but we were not able to determine if they worked exactly the same way in those states because we lacked the data we had in Maryland. Allstate did not answer our questions about the algorithms. Consumer Watchdog founder Harvey Rosenfield, who has long been critical of Allstate’s pricing practices, said that the company found a different method of achieving the same goals in California, which only allows insurers to use a limited number of approved factors—like driving record, type of car, or the number of years behind the wheel—to determine how much customers have to pay in premiums. “What we’re contending is they took their knowledge of price elasticity and figured out a pretty straightforward way of doing it without saying so,” Rosenfield said. “Their actuaries selected relativities to set people’s premiums that punished people who Allstate knew were inelastic, who fit the profile of being less sensitive to price changes. They charged those people more.” California isn’t the only place where Allstate has faced litigation over its use of price optimization. Shortly after the publication of The Markup’s Maryland investigation, the company was hit with a that directly referenced The Markup’s reporting. class action lawsuit in Texas That suit that the company was “charging higher premiums to its more tenured policyholders than it charges otherwise identically-situated newer policyholders for the same or materially the same coverages.” alleges That case is still ongoing. Correction The story originally stated that the Texas class action lawsuit against Allstate had settled last year. It is still ongoing. The story has been updated to correct the error. Written by: Aaron Sankin Photo by on Vlad Deep Unsplash Also published here

Allstate Accused of Overcharging Loyal California Customers $1 Billion

A joint investigation by The Markup and Consumer Reports was cited in a lawsuit over discriminatory car insurance rates On April 28, a group of current and former Allstate Corp. insurance policyholders sued the company, alleging that it had engaged in discriminatory pricing practices, including using a that The Markup and Consumer Reports uncovered and reported about in February. The was filed in the Austin division of the United States District Court for the Western District of Texas. secret algorithm complaint The suit alleges that “Allstate essentially has to pick the rate that it will charge each policyholder” using the secret algorithm. It claims that Allstate viewed longtime customers as less sensitive to premium price changes and alleges that those customers were “generally charged higher premiums for the same policies and coverages than otherwise identically situated customers with less tenure.” carte blanche Allstate essentially has carte blanche to pick the rate that it will charge each policyholder. The suit against Allstate The suit cites the joint investigation by The Markup and Consumer Reports, whose of Allstate customer data from a Maryland filing (which was withdrawn before it was put into effect) found that customers who were already paying some of the highest rates would have also been charged rate hikes of up to 20 percent, compared to more limited increases of up to 5 percent for similarly situated customers who had been paying cheaper rates. The analysis also found that no customer would have received a substantial decrease, with rate decreases capped at half of one percent. analysis Allstate spokesperson Shaundra Turner Jones responded to the report by The Markup and Consumer Reports saying that Allstate’s rating plans comply with state laws and regulations and that the reporting on the Maryland filing was “inaccurate and misleading” because it was “based on a rating plan that was never used.” The Texas lawsuit also claims that Allstate charged customers differently depending on what “book” they were in. Allstate maintains several auto insurance companies (referred to as “books of business” in the insurance industry) that issue insurance policies to customers. While some of those books are “open,” meaning that they take new customers, several of the Allstate books operating in Texas are “closed,” meaning that they only work with pre-existing customers. The suit alleges that “the premiums that Allstate charges new policyholders in the Open Books are generally significantly lower (often, very significantly so) than the premiums Allstate charges existing policyholders in the Closed Books for the same or materially the same coverages.” In an email to The Markup, Allstate spokesperson Roberto DeLeon said, “There is no merit to this complaint. We pair quality, customized coverage with great savings and reward long-term customers for their loyalty.” Texas law prohibits insurers from treating policyholders differently “without a sound actuarial basis.” The suit alleges that Allstate knowingly violated this prohibition as early as 2015, when the company was sent by the Texas Office of Public Insurance Counsel (OPIC), an agency independent from the state’s insurance department, stating that the retention model it was proposing violated multiple Texas laws. a letter Deeia Beck, public counsel for OPIC, wrote that “allowing clearly non-risk-and-cost-based rating factors opens the door to further conditions where certain classes of consumers are unfairly priced out of the market or shunted into low quality products.” In explaining the differences between accepted insurance practices and what Allstate was doing, Beck noted that “discriminatory intent is introduced” when insurers introduce “for example, a 25% rate cap for older drivers and a 40% cap for youthful drivers.” The Markup’s and Consumer Reports’ of Allstate customer data in Maryland found that customers hit with a 20 percent increase were more likely to be middle-aged. analysis It seems Allstate continued its use of the retention model even after receiving the OPIC letter: The Markup and Consumer Reports found a from Allstate representatives to the Ohio Department of Insurance that claimed that the company was still using the retention model in Texas at that time but that it was under “active review.” July 2016 letter Credits Originally published as " Texas Drivers Sue Allstate over Secret “Suckers List” " with the Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) license.

What Is “Suckers List” And Why Texas Drivers Sued Its Creator

Allstate's Car Insurance Algorithm: How Insurance Algorithm Was Analyzed

Allstate's Car Insurance Algorithm: How Insurance Algorithm Squeezes Big Spenders

Building Immutable foundations from the ground TL;DR: Pass essential attributes during object creation to reduce mutability and eliminate getters and setters. Problems Addressed Mutability
Getters
Setters
Unclear object state
Violating the least surprise principle Steps Identify essential attributes required for object creation
Create a constructor that accepts all essential attributes
Remove setter and getter methods
Update object creation calls to pass all required attributes upfront Sample Code Before public class CreditCard {
    private String cardNumber;
    private String cardHolderName;
    private String expirationMonthYear;
    private int cvv;

    public CreditCard() {} // Empty Constructor

    public void setCardNumber(String cardNumber) { 
        this.cardNumber = cardNumber; 
    }
    public void setCardHolderName(String cardHolderName) { 
        this.cardHolderName = cardHolderName; 
    }
    public void setExpirationMonthYear(String expirationMonthYear) { 
        this.expirationMonthYear = expirationMonthYear;
    }
    public void setCvv(int cvv) { 
        this.cvv = cvv; 
    }

    public String getCardNumber() { 
        return cardNumber; 
    }
    public String getCardHolderName() {
        return cardHolderName;
    }
    public String getExpirationMonthYear() {
        return expirationMonthYear; 
    }
    public int getCvv() {
        return cvv; 
    }
}

CreditCard card = new CreditCard();
card.setCardNumber("1234-5678-9012-3456");
card.setCardHolderName("Lilywhite Lilith");
card.setExpirationMonthYear("12/25");
card.setCvv(123); After public class CreditCard {
    private final String cardNumber;
    private final String cardHolderName;
    private final String expirationMonthYear;
    private final int cvv;

    public CreditCard(String cardNumber,
                      String cardHolderName,
                      String expirationMonthYear,
                      int cvv) {
        // 1. Identify essential attributes for object creation
        // 2. Create a constructor that accepts all essential attributes
        this.cardNumber = cardNumber;
        this.cardHolderName = cardHolderName;
        this.expirationMonthYear = expirationMonthYear;
        this.cvv = cvv;
    }

    // 3. Remove setter and getter methods 
    
    // Find real behavior related to credit card usage
  
}

// 4. Update object creation calls to pass all required attributes upfront
CreditCard card = new CreditCard("1234-5678-9012-3456",
                                 "Lilywhite Lilith", 
                                 "12/25", 
                                 123); Type Semi-Automatic This is a step-by-step refactor. Safety This refactoring is generally safe if you ensure you pass all essential attributes during object creation. You must update all object creation sites, which may require refactoring tools and careful review in larger codebases. Why is the code better? The refactored code enforces object integrity by requiring all essential attributes at creation time. The objects become thread-safe reducing the risk of being in an inconsistent state. AI Correction Most AI tools can correct this code with explicit instructions. You can have clear pre-prompts requesting all your code samples to favor immutability. Try Them! Remember AI Assistants make lots of mistakes Without Proper Instructions

With Specific Instructions



ChatGPT

ChatGPT



Claude

Claude



Perplexity

Perplexity



Gemini

Gemini Tags Mutability Related Refactorings Refactoring 001 - Remove Setters See also https://hackernoon.com/nude-models-part-i-setters-5tm3u3y?embedable=true https://hackernoon.com/nude-models-part-ii-getters-sjo3ua2?embedable=true https://hackernoon.com/is-it-crystal-clear-for-everybody-that-a-date-should-not-mutate-wuoy3z03?embedable=true Credits Image by Denis on Pixabay This article is part of the Refactoring Series. How to Improve your Code With Easy Refactorings Building Immutable foundations from the ground Building Immutable foundations from the ground TL;DR: Pass essential attributes during object creation to reduce mutability and eliminate getters and setters. TL;DR: Pass essential attributes during object creation to reduce mutability and eliminate getters and setters. Problems Addressed Mutability Getters Setters Unclear object state Violating the least surprise principle Mutability Mutability Getters Getters Setters Setters Unclear object state Violating the least surprise principle Steps Identify essential attributes required for object creation Create a constructor that accepts all essential attributes Remove setter and getter methods Update object creation calls to pass all required attributes upfront Identify essential attributes required for object creation Create a constructor that accepts all essential attributes Remove setter and getter methods Update object creation calls to pass all required attributes upfront Sample Code Before public class CreditCard {
    private String cardNumber;
    private String cardHolderName;
    private String expirationMonthYear;
    private int cvv;

    public CreditCard() {} // Empty Constructor

    public void setCardNumber(String cardNumber) { 
        this.cardNumber = cardNumber; 
    }
    public void setCardHolderName(String cardHolderName) { 
        this.cardHolderName = cardHolderName; 
    }
    public void setExpirationMonthYear(String expirationMonthYear) { 
        this.expirationMonthYear = expirationMonthYear;
    }
    public void setCvv(int cvv) { 
        this.cvv = cvv; 
    }

    public String getCardNumber() { 
        return cardNumber; 
    }
    public String getCardHolderName() {
        return cardHolderName;
    }
    public String getExpirationMonthYear() {
        return expirationMonthYear; 
    }
    public int getCvv() {
        return cvv; 
    }
}

CreditCard card = new CreditCard();
card.setCardNumber("1234-5678-9012-3456");
card.setCardHolderName("Lilywhite Lilith");
card.setExpirationMonthYear("12/25");
card.setCvv(123); public class CreditCard {
    private String cardNumber;
    private String cardHolderName;
    private String expirationMonthYear;
    private int cvv;

    public CreditCard() {} // Empty Constructor

    public void setCardNumber(String cardNumber) { 
        this.cardNumber = cardNumber; 
    }
    public void setCardHolderName(String cardHolderName) { 
        this.cardHolderName = cardHolderName; 
    }
    public void setExpirationMonthYear(String expirationMonthYear) { 
        this.expirationMonthYear = expirationMonthYear;
    }
    public void setCvv(int cvv) { 
        this.cvv = cvv; 
    }

    public String getCardNumber() { 
        return cardNumber; 
    }
    public String getCardHolderName() {
        return cardHolderName;
    }
    public String getExpirationMonthYear() {
        return expirationMonthYear; 
    }
    public int getCvv() {
        return cvv; 
    }
}

CreditCard card = new CreditCard();
card.setCardNumber("1234-5678-9012-3456");
card.setCardHolderName("Lilywhite Lilith");
card.setExpirationMonthYear("12/25");
card.setCvv(123); After public class CreditCard {
    private final String cardNumber;
    private final String cardHolderName;
    private final String expirationMonthYear;
    private final int cvv;

    public CreditCard(String cardNumber,
                      String cardHolderName,
                      String expirationMonthYear,
                      int cvv) {
        // 1. Identify essential attributes for object creation
        // 2. Create a constructor that accepts all essential attributes
        this.cardNumber = cardNumber;
        this.cardHolderName = cardHolderName;
        this.expirationMonthYear = expirationMonthYear;
        this.cvv = cvv;
    }

    // 3. Remove setter and getter methods 
    
    // Find real behavior related to credit card usage
  
}

// 4. Update object creation calls to pass all required attributes upfront
CreditCard card = new CreditCard("1234-5678-9012-3456",
                                 "Lilywhite Lilith", 
                                 "12/25", 
                                 123); public class CreditCard {
    private final String cardNumber;
    private final String cardHolderName;
    private final String expirationMonthYear;
    private final int cvv;

    public CreditCard(String cardNumber,
                      String cardHolderName,
                      String expirationMonthYear,
                      int cvv) {
        // 1. Identify essential attributes for object creation
        // 2. Create a constructor that accepts all essential attributes
        this.cardNumber = cardNumber;
        this.cardHolderName = cardHolderName;
        this.expirationMonthYear = expirationMonthYear;
        this.cvv = cvv;
    }

    // 3. Remove setter and getter methods 
    
    // Find real behavior related to credit card usage
  
}

// 4. Update object creation calls to pass all required attributes upfront
CreditCard card = new CreditCard("1234-5678-9012-3456",
                                 "Lilywhite Lilith", 
                                 "12/25", 
                                 123); Type Semi-Automatic Semi-Automatic This is a step-by-step refactor. Safety This refactoring is generally safe if you ensure you pass all essential attributes during object creation. You must update all object creation sites, which may require refactoring tools and careful review in larger codebases. Why is the code better? The refactored code enforces object integrity by requiring all essential attributes at creation time. The objects become thread-safe reducing the risk of being in an inconsistent state. AI Correction Most AI tools can correct this code with explicit instructions. You can have clear pre-prompts requesting all your code samples to favor immutability. Try Them! Remember AI Assistants make lots of mistakes Remember AI Assistants make lots of mistakes Without Proper Instructions

With Specific Instructions



ChatGPT

ChatGPT



Claude

Claude



Perplexity

Perplexity



Gemini

Gemini Without Proper Instructions

With Specific Instructions ChatGPT

ChatGPT Claude

Claude Perplexity

Perplexity Gemini

Gemini Without Proper Instructions With Specific Instructions Without Proper Instructions Without Proper Instructions With Specific Instructions With Specific Instructions ChatGPT ChatGPT ChatGPT ChatGPT ChatGPT ChatGPT ChatGPT ChatGPT Claude Claude Claude Claude Claude Claude Claude Claude Perplexity Perplexity Perplexity Perplexity Perplexity Perplexity Perplexity Perplexity Gemini Gemini Gemini Gemini Gemini Gemini Gemini Gemini Tags Mutability Mutability Related Refactorings Refactoring 001 - Remove Setters Refactoring 001 - Remove Setters See also https://hackernoon.com/nude-models-part-i-setters-5tm3u3y?embedable=true https://hackernoon.com/nude-models-part-i-setters-5tm3u3y?embedable=true https://hackernoon.com/nude-models-part-ii-getters-sjo3ua2?embedable=true https://hackernoon.com/nude-models-part-ii-getters-sjo3ua2?embedable=true https://hackernoon.com/is-it-crystal-clear-for-everybody-that-a-date-should-not-mutate-wuoy3z03?embedable=true https://hackernoon.com/is-it-crystal-clear-for-everybody-that-a-date-should-not-mutate-wuoy3z03?embedable=true Credits Image by Denis on Pixabay Denis Pixabay This article is part of the Refactoring Series. How to Improve your Code With Easy Refactorings How to Improve your Code With Easy Refactorings

Refactoring 016 - Building With The Essence

Table of Links Abstract and 1 Introduction 2 Preliminaries 3 End-to-End Adaptive Local Learning 3.1 Loss-Driven Mixture-of-Experts 3.2 Synchronized Learning via Adaptive Weight 4 Debiasing Experiments and 4.1 Experimental Setup 4.2 Debiasing Performance 4.3 Ablation Study 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 5 Related Work 6 Conclusion, Acknowledgements, and References 4.2 Debiasing Performance First, we conduct a comparative analysis to show the effectiveness of the proposed TALL. In Table 2, we evaluate the overall NDCG@20 and average NDCG@20 for five user subgroups with varying mainstream levels for all methods and datasets. The best results of each metric and subgroup for all datasets are marked in bold, and the improvement rate of the proposed TALL over the best baseline MultVAE, LOCA, and LFT is exhibited as well. The user subgroups are categorized based on their mainstream scores. TALL vs. MultVAE & WL. First, we can observe that the utilities for users of all five subgroups are greatly promoted by our proposed TALL compared to the widely used model MultVAE. Moreover, we can see that although the global debiasing method WL can alleviate the mainstream bias to a certain degree compared to MultVAE, our proposed TALL can produce higher NDCG@20 for all five user subgroups than WL, depicting that the proposed TALL exhibits a more outstanding debiasing ability over WL. TALL vs. EnLFT & LOCA. Hence, we next have a fair comparison between models of the same complexity and compare directly across local learning methods. From Table 2, we observe that LOCA and EnLFT are more effective in mitigating the mainstream bias than WL, as they remarkably enhance utility across all five groups on all datasets. Meanwhile, our TALL significantly outperforms LOCA and EnLFT across all user groups and datasets. The improvement is especially prominent for niche users: TALL improves NDCG@20 of the ‘low’ user group by 6.07% on average over LOCA and 10% over EnLFT. This shows that with the same model complexity, the proposed end-to-end adaptive local learning model is more effective than heuristic-based local learning models. TALL vs. LFT. Last, we compare TALL with the state-of-the-art local learning baseline LFT, which is heavily computationally intense and time-consuming. But due to its special design that every user gets their own customized model trained by their local data, LFT can effectively address the discrepancy modeling problem. From Table 2, we observe that, for most of the time, LFT achieves the best performance for niche users among all baselines. And LFT can perform especially effectively for dense datasets (i.e., ML1M). In fact, given that the model complexity of LFT is much higher than TALL, it is unfair to compare them only based on recommendation accuracy (i.e., NDCG). For example, in the ML1M dataset, our TALL contains 100 expert models (MultVAE), while LFT trains 6,040 (#users) models separately, which is over 60 times larger than TALL. Although it is not a fair comparison, we can still observe in Table 2 that TALL can outperform LFT in most cases. Especially for the two sparse datasets Yelp and CDs&Vinyl, TALL produces significantly higher utilities for all types of users. This demonstrates the efficacy and necessity of an end-to-end local learning method compared to a heuristic-based one. In sum, from Table 2, we see that for all datasets, the proposed TALL produces the greatest NDCG@20 improvement for each subgroup of different mainstream levels and leads to the state-of-the-art overall model performance. TALL can outperform baselines with lower and the same model complexity, and it can even outperform the baseline model that is way more complex than it. Authors:
(1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA;
(2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA;
(3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA;
(4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA;
(5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. This paper is available on arxiv under CC BY 4.0 DEED license. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Preliminaries 2 Preliminaries 3 End-to-End Adaptive Local Learning 3 End-to-End Adaptive Local Learning 3.1 Loss-Driven Mixture-of-Experts 3.1 Loss-Driven Mixture-of-Experts 3.2 Synchronized Learning via Adaptive Weight 3.2 Synchronized Learning via Adaptive Weight 4 Debiasing Experiments and 4.1 Experimental Setup 4 Debiasing Experiments and 4.1 Experimental Setup 4.2 Debiasing Performance 4.2 Debiasing Performance 4.3 Ablation Study Ablation 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 5 Related Work 5 Related Work 6 Conclusion, Acknowledgements, and References 6 Conclusion, Acknowledgements, and References 4.2 Debiasing Performance First, we conduct a comparative analysis to show the effectiveness of the proposed TALL. In Table 2, we evaluate the overall NDCG@20 and average NDCG@20 for five user subgroups with varying mainstream levels for all methods and datasets. The best results of each metric and subgroup for all datasets are marked in bold, and the improvement rate of the proposed TALL over the best baseline MultVAE, LOCA, and LFT is exhibited as well. The user subgroups are categorized based on their mainstream scores. TALL vs. MultVAE & WL. First, we can observe that the utilities for users of all five subgroups are greatly promoted by our proposed TALL compared to the widely used model MultVAE. Moreover, we can see that although the global debiasing method WL can alleviate the mainstream bias to a certain degree compared to MultVAE, our proposed TALL can produce higher NDCG@20 for all five user subgroups than WL, depicting that the proposed TALL exhibits a more outstanding debiasing ability over WL. TALL vs. MultVAE & WL. TALL vs. EnLFT & LOCA. Hence, we next have a fair comparison between models of the same complexity and compare directly across local learning methods. From Table 2, we observe that LOCA and EnLFT are more effective in mitigating the mainstream bias than WL, as they remarkably enhance utility across all five groups on all datasets. Meanwhile, our TALL significantly outperforms LOCA and EnLFT across all user groups and datasets. The improvement is especially prominent for niche users: TALL improves NDCG@20 of the ‘low’ user group by 6.07% on average over LOCA and 10% over EnLFT. This shows that with the same model complexity, the proposed end-to-end adaptive local learning model is more effective than heuristic-based local learning models. TALL vs. EnLFT & LOCA. TALL vs. LFT. Last, we compare TALL with the state-of-the-art local learning baseline LFT, which is heavily computationally intense and time-consuming. But due to its special design that every user gets their own customized model trained by their local data, LFT can effectively address the discrepancy modeling problem. From Table 2, we observe that, for most of the time, LFT achieves the best performance for niche users among all baselines. And LFT can perform especially effectively for dense datasets (i.e., ML1M). In fact, given that the model complexity of LFT is much higher than TALL, it is unfair to compare them only based on recommendation accuracy (i.e., NDCG). For example, in the ML1M dataset, our TALL contains 100 expert models (MultVAE), while LFT trains 6,040 (#users) models separately, which is over 60 times larger than TALL. Although it is not a fair comparison, we can still observe in Table 2 that TALL can outperform LFT in most cases. Especially for the two sparse datasets Yelp and CDs&Vinyl, TALL produces significantly higher utilities for all types of users. This demonstrates the efficacy and necessity of an end-to-end local learning method compared to a heuristic-based one. TALL vs. LFT. In sum, from Table 2, we see that for all datasets, the proposed TALL produces the greatest NDCG@20 improvement for each subgroup of different mainstream levels and leads to the state-of-the-art overall model performance. TALL can outperform baselines with lower and the same model complexity, and it can even outperform the baseline model that is way more complex than it. Authors: (1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA; (2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA; (3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA; (4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA; (5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. Authors: Authors: (1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA; (2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA; (3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA; (4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA; (5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. 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 available on arxiv

Countering Mainstream Bias via End-to-End Adaptive Local Learning: Debiasing Performance

COPA v. Wright, Court Filing, retrieved on January 29, 2024, is part of HackerNoon’s Legal PDF Series. You can jump to any part of this filing here. This part is 28 of 42. 26. The SSRN Submission (Particulars of Claim at [30]-[35] {A/2/10}) The Fourth Pleaded Example 489. On or about 21 August 2019, Dr Wright uploaded to the Social Science Research Network (SSRN) a document which he asserts to be the ‘final’ version of the Bitcoin White Paper, in which he is named as the author. The post claims that the document was written on 21 August 2008. Dr Wright posted two versions of what he claims to be his original version of the Bitcoin White Paper within SSRN. **(a) COPA’s Reasons for Alleging Forgery** 490. The metadata of the actual Bitcoin White Paper has a creation date of 24 March 2009. The first version Dr Wright submitted to SSRN has discrepancies in the document properties. Examination of this first version reveals that its metadata was tampered with - supplemental metadata was added to this version to superficially indicate a creation date of 24 January 2008 and a “modified” date of 21 May 2008. This added metadata displays as the creation and modified date in the user interface of standard consumer software, i.e. the tampered dates are what would appear to a layperson upon cursory examination. However, a forensic examination of the metadata of this first version reveals another “creation date” entry embedded within the metadata, specifying a creation date that corresponds exactly to the creation date of the Bitcoin White Paper, specifically “/CreationDate (D:20090324113315-06:00)". 491. The metadata of this first version includes an entry indicating that the original creation date is 29 March 2009, which is at odds with the creation date of 24 January 2008 that shows up in other parts of the metadata. In the premises, it is averred that this first version posted to SSRN is actually a copied version of the Bitcoin White Paper with alterations to its metadata (at least that the dates of 24 January 2008 and 21 May 2008 were added to the genuine metadata of the Bitcoin White Paper) to make it appear that it was created in 2008. It is averred that the dates of 24 January 2008 and 21 May 2008 were added to the genuine metadata of the Bitcoin White Paper by Wright to support his attempt to claim that he had drafted the Bitcoin White Paper. 492. The second version that Dr Wright uploaded to SSRN, which is the version he has currently uploaded there, had further changes to the metadata compared to the first version. However, the original creation date of 24 March 2009 from the actual Bitcoin White Paper still exists within the second version’s metadata. COPA allege that Dr Wright again sought to manipulate the metadata a second time but failed to remove the original creation date. 493. The genuine metadata in the actual Bitcoin White Paper and metadata contained within the two versions posted on SSRN by Dr Wright all state the 24 March 2009 date as being the creation date, with the dates/times agreeing to the millisecond. (b) Dr Wright’s Explanations and COPA’s Rebuttal 494. In his Defence, Dr Wright admits that the two versions of the Bitcoin White Paper which Dr Wright uploaded to SSRN were not created in 2008 or 2009, but in 2019. He claims that his purpose in uploading these versions was not to prove that he was the author of the Bitcoin White Paper but “to assert [his] authorship of the White Paper”: see Defence, [55-56] {A/3/16}. In his Defence, Dr Wright gives no explanation for the uploaded versions having metadata creation dates of 2008 and 2009 respectively. 495. In Appendix B to Wright11, Dr Wright repeats the account given in his Defence. However, he now adds the assertion that he prepared each document in LATEX and that he set the metadata in LATEX to refer to creation dates of 2008 and 2009 respectively. See: {CSW/3/8}. 496. COPA submitted that Dr Wright’s explanation should be rejected for the following reasons: 496.1. As established by the evidence of Mr Rosendahl (which was agreed in the joint statement by Dr Wright’s expert, Mr Lynch), Dr Wright’s account of the Bitcoin White Paper being authored in LATEX is false. There is equally no basis for saying that the versions of the White Paper uploaded to the SSRN in 2019 were created in LATEX. 496.2. Accordingly, the only conclusion to be drawn from these versions having been created by Dr Wright in 2019 (as is admitted) but containing metadata referring to 2008 / 2009 creation dates is that the metadata information was deliberately backdated. 496.3. In summary, Dr Wright in 2019 posted publicly two versions of the Bitcoin White Paper which he presented as his work and which contained metadata timestamps backdated to 2008. In the context of Dr Wright claiming to be Satoshi Nakamoto and to have written the Bitcoin White Paper, the only plausible explanation is that he was presenting these to the world as early originals of the paper written by him. (c) Conclusion 497. In addition to the points set out above, I refer also to the analysis in the main Judgment at [489]-[495] of what Dr Wright did with the pdfcreationdate command and how those commands were added to the White Paper LaTeX files in November 2023. 498. Accordingly, Dr Wright’s latest explanation, based on the assertion that the Bitcoin White Paper was created in LATEX, is plainly false. The document was deliberately backdated by Dr Wright, in an attempt to present those two versions of the Bitcoin White Paper as early originals written by him. This was all lies. Continue Reading Here. About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings.

This court case retrieved on January 29, 2024, judiciary.uk is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. COPA v. Wright, Court Filing, retrieved on January 29, 2024, is part of HackerNoon’s Legal PDF Series. You can jump to any part of this filing here. This part is 28 of 42. COPA v. Wright, Court Filing, retrieved on January 29, 2024, is part of HackerNoon’s Legal PDF Series . You can jump to any part of this filing here . This part is 28 of 42. HackerNoon’s Legal PDF Series HackerNoon’s Legal PDF Series here here 26. The SSRN Submission (Particulars of Claim at [30]-[35] {A/2/10}) The Fourth Pleaded Example 489. On or about 21 August 2019, Dr Wright uploaded to the Social Science Research Network (SSRN) a document which he asserts to be the ‘final’ version of the Bitcoin White Paper, in which he is named as the author. The post claims that the document was written on 21 August 2008. Dr Wright posted two versions of what he claims to be his original version of the Bitcoin White Paper within SSRN. **(a) COPA’s Reasons for Alleging Forgery ** 490. The metadata of the actual Bitcoin White Paper has a creation date of 24 March 2009. The first version Dr Wright submitted to SSRN has discrepancies in the document properties. Examination of this first version reveals that its metadata was tampered with - supplemental metadata was added to this version to superficially indicate a creation date of 24 January 2008 and a “modified” date of 21 May 2008. This added metadata displays as the creation and modified date in the user interface of standard consumer software, i.e. the tampered dates are what would appear to a layperson upon cursory examination. However, a forensic examination of the metadata of this first version reveals another “creation date” entry embedded within the metadata, specifying a creation date that corresponds exactly to the creation date of the Bitcoin White Paper, specifically “/CreationDate (D:20090324113315-06:00)". 491. The metadata of this first version includes an entry indicating that the original creation date is 29 March 2009, which is at odds with the creation date of 24 January 2008 that shows up in other parts of the metadata. In the premises, it is averred that this first version posted to SSRN is actually a copied version of the Bitcoin White Paper with alterations to its metadata (at least that the dates of 24 January 2008 and 21 May 2008 were added to the genuine metadata of the Bitcoin White Paper) to make it appear that it was created in 2008. It is averred that the dates of 24 January 2008 and 21 May 2008 were added to the genuine metadata of the Bitcoin White Paper by Wright to support his attempt to claim that he had drafted the Bitcoin White Paper. 492. The second version that Dr Wright uploaded to SSRN, which is the version he has currently uploaded there, had further changes to the metadata compared to the first version. However, the original creation date of 24 March 2009 from the actual Bitcoin White Paper still exists within the second version’s metadata. COPA allege that Dr Wright again sought to manipulate the metadata a second time but failed to remove the original creation date. 493. The genuine metadata in the actual Bitcoin White Paper and metadata contained within the two versions posted on SSRN by Dr Wright all state the 24 March 2009 date as being the creation date, with the dates/times agreeing to the millisecond. (b) Dr Wright’s Explanations and COPA’s Rebuttal (b) Dr Wright’s Explanations and COPA’s Rebuttal 494. In his Defence, Dr Wright admits that the two versions of the Bitcoin White Paper which Dr Wright uploaded to SSRN were not created in 2008 or 2009, but in 2019. He claims that his purpose in uploading these versions was not to prove that he was the author of the Bitcoin White Paper but “to assert [his] authorship of the White Paper”: see Defence, [55-56] {A/3/16}. In his Defence, Dr Wright gives no explanation for the uploaded versions having metadata creation dates of 2008 and 2009 respectively. 495. In Appendix B to Wright11, Dr Wright repeats the account given in his Defence. However, he now adds the assertion that he prepared each document in LATEX and that he set the metadata in LATEX to refer to creation dates of 2008 and 2009 respectively. See: {CSW/3/8}. 496. COPA submitted that Dr Wright’s explanation should be rejected for the following reasons: 496.1. As established by the evidence of Mr Rosendahl (which was agreed in the joint statement by Dr Wright’s expert, Mr Lynch), Dr Wright’s account of the Bitcoin White Paper being authored in LATEX is false. There is equally no basis for saying that the versions of the White Paper uploaded to the SSRN in 2019 were created in LATEX. 496.2. Accordingly, the only conclusion to be drawn from these versions having been created by Dr Wright in 2019 (as is admitted) but containing metadata referring to 2008 / 2009 creation dates is that the metadata information was deliberately backdated. 496.3. In summary, Dr Wright in 2019 posted publicly two versions of the Bitcoin White Paper which he presented as his work and which contained metadata timestamps backdated to 2008. In the context of Dr Wright claiming to be Satoshi Nakamoto and to have written the Bitcoin White Paper, the only plausible explanation is that he was presenting these to the world as early originals of the paper written by him. (c) Conclusion (c) Conclusion 497. In addition to the points set out above, I refer also to the analysis in the main Judgment at [489]-[495] of what Dr Wright did with the pdfcreationdate command and how those commands were added to the White Paper LaTeX files in November 2023. 498. Accordingly, Dr Wright’s latest explanation, based on the assertion that the Bitcoin White Paper was created in LATEX, is plainly false. The document was deliberately backdated by Dr Wright, in an attempt to present those two versions of the Bitcoin White Paper as early originals written by him. This was all lies. Continue Reading Here. Continue Reading Here . Here About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings. This court case retrieved on January 29, 2024, judiciary.uk is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings. This court case retrieved on January 29, 2024, judiciary.uk is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. judiciary.uk judiciary.uk

COPA Sues Craig Wright for Forging Bitcoin White Paper Metadata

After Uncle Sam Segregated America, Here’s How Everyday People Desegregate It 

আঙ্কেল স্যাম আমেরিকাকে আলাদা করার পরে, প্রতিদিনের লোকেরা কীভাবে এটি আলাদা করে তা এখানে রয়েছে 

Nachdem Uncle Sam Amerika segregierte, so desegregieren die Menschen es

Después de que el Tío Sam segregó a Estados Unidos, así es como la gente común lo elimina 

Après la ségrégation de l’Amérique par l’Oncle Sam, voici comment les gens ordinaires la déségrégent 

अंकल सैम द्वारा अमेरिका को अलग-थलग करने के बाद, यहां बताया गया है कि कैसे रोज़मर्रा के लोग इसे अलग-थलग करते हैं

サムおじさんがアメリカを人種差別した後、一般の人々が人種差別を撤廃した方法

엉클 샘이 미국을 인종차별한 후, 일반 사람들이 미국을 인종차별을 철폐하는 방법은 다음과 같습니다. 

Depois que o Tio Sam segregou a América, veja como as pessoas comuns a desagregam 

После того как дядя Сэм разделил Америку, вот как ее десегрегируют обычные люди 

Sam Amca Amerika'yı Ayırdıktan Sonra Sıradan İnsanlar Amerikayı Nasıl Ayırıyor? 

Sau khi chú Sam tách biệt nước Mỹ, đây là cách mọi người tách biệt nó 

山姆大叔对美国实行种族隔离后，普通民众如何废除种族隔离

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Jiwan Chung, MIR Lab Yonsei University (https://jiwanchung.github.io/);
(2) Youngjae Yu, MIR Lab Yonsei University (https://jiwanchung.github.io/). Table of Links Abstract and Intro
Method
Experiments
Related Work
Conclusion
Limitations and References
A. Experiment Details
B. Prompt Samples 3. Experiments For all experiments, we use GPT-3 [1] (text-davinci-003) as the backbone language model. Unless stated otherwise, we use the ground truth clip boundary to segment the videos. All LSS variants do not use any training data and thus are zero-shot methods. 3.1. Evaluating Long Story Short MovieQA [27] is a large-scale QA dataset sourced from 408 movies. There are multiple sources of information in the dataset; subtitles, scripts, DVS, video clips, and plots. We report four state-of-the-art supervised baselines; A2A [20], PAMN [11], UniversalQA [10], and DHTCN [21]. Table 1 shows zero-shot LSS improves over previous supervised approaches. Also, Ours-search shows strong performance even without the ground-truth segment index label. CLIPCheck slightly improves the accuracy in the video split. However, the difference is marginal since MovieQA often requires character-based grounding rather than general visual matching. Finally, we experiment with the null hypothesis: No Context tests whether GPT-3 solves MovieQA by simply memorizing every fact. No Context performs worse than LSS, rejecting the null hypothesis. PororoQA [13] is a video story QA dataset built from a cartoon series. The supervised baseline takes the human-generated plot and the ground truth video segment index, while LSS +Plot+Search takes neither. Table 2 summarizes our result on the PororoQA dataset. When using both the ground-truth episode and plots, GPT-3 performs almost on par with the supervised baseline. Substituting a human-generated summary with a model-generated one results in only a marginal performance drop. Perhaps intriguingly, the search process works better when using model-generated plots. We attribute this result to the fact that the human annotations are not designed for episode discriminability. 3.2. Evaluating CLIPCheck DramaQA [3] is video QA dataset that focuses on story understanding. The dataset is arranged with four levels of hierarchical difficulty, which follow the human cognitivedevelopmental stages. We evaluate LSS on the two high levels of DramaQA to test plot understanding. We report two latest baselines in level-wise DramaQA; CharacterAttention and Kim et al. [14]. We compare the effect of CLIPCheck and Caption, a prompt-based method of incorporating image frame descriptions extracted from BLIP [18] as inputs to GPT-3. Table 3 shows that CLIPCheck offers greater improvement than image descriptions. Also, while adding image captions improves LSS, the gain disappears when used jointly with CLIPCheck. We suspect that this is because frame captions provide similar information to CLIPCheck while being much noisier. Note that the automatic Captions here are not an integral component of LSS. As DramaQA has visually grounded annotations already, adding automatic image Captions on top of that would not necessarily improve the model performance. Rather, we use the Captions to explicitly compare early vs. late visual alignment methods. Finally, we check whether CLIPCheck exploits the dataset bias rather than understanding the visual context. To this end, we devise a variant of CLIPCheck with random visual context (CLIPCheck-Shuffle). CLIPCheck-Shuffle does not improve over LSS with no CLIPCheck, denying the bias hypothesis. 3.3. Ablation Study Are both the summarization and search important for narrative understanding? Here, we evaluate LSS variants with full context without the narrative search (LSS-Full) or with the plot summary and random segment as inputs (LSS-Random). Table 4 shows that both LSS-Full and LSS-Random fall behind LSS-Search, indicating the importance of retrieval. Note that we could not employ the full context in LSS-Full due to the token length limitation. Instead, we use the longest prefix of the full context that GPT3 accepts (4000 tokens minus the length of the instruction). 3.4. Qualitative Results Figure 3 shows the automatic plot summary generated as an intermediate context of the long video QA using the language model in the LSS framework. As shown in the qualitative sample, the generated plots align well with the human-written plots from Wikipedia. For example, in the first scene of the movie "Harry Potter and the Deathly Hallows", the LSS summary correctly writes that Harry Potter is currently 17 years old and the main event in which the death eaters attack the protagonist. Figure 4 depicts the connection between the searched plot piece and the answer likelihood. In the example on the left, the retrieved summary tells that Trench committed a crime and thus is on the run, suggesting that another character interested in him would be chasing him. The language model understands this context to modify the answer likelihood in the correct way. In the right example, the LSS plot piece suggests that Edward is confident in his decision. While this context does not offer a direct cue to the question, the language model sees it as information strong enough to alter the answer. This paper is available on arxiv under CC 4.0 license. Authors: (1) Jiwan Chung, MIR Lab Yonsei University (https://jiwanchung.github.io/); (2) Youngjae Yu, MIR Lab Yonsei University (https://jiwanchung.github.io/). This paper is available on arxiv under CC 4.0 license. Authors: Authors: (1) Jiwan Chung, MIR Lab Yonsei University ( https://jiwanchung.github.io/ ); https://jiwanchung.github.io/ (2) Youngjae Yu, MIR Lab Yonsei University ( https://jiwanchung.github.io/ ). https://jiwanchung.github.io/ Table of Links Abstract and Intro Method Experiments Related Work Conclusion Limitations and References A. Experiment Details B. Prompt Samples Abstract and Intro Abstract and Intro Method Method Experiments Experiments Related Work Related Work Conclusion Conclusion Limitations and References Limitations and References A. Experiment Details A. Experiment Details B. Prompt Samples B. Prompt Samples 3. Experiments For all experiments, we use GPT-3 [1] (text-davinci-003) as the backbone language model. Unless stated otherwise, we use the ground truth clip boundary to segment the videos. All LSS variants do not use any training data and thus are zero-shot methods. 3.1. Evaluating Long Story Short MovieQA [27] is a large-scale QA dataset sourced from 408 movies. There are multiple sources of information in the dataset; subtitles, scripts, DVS, video clips, and plots. We report four state-of-the-art supervised baselines; A2A [20], PAMN [11], UniversalQA [10], and DHTCN [21]. Table 1 shows zero-shot LSS improves over previous supervised approaches. Also, Ours-search shows strong performance even without the ground-truth segment index label. CLIPCheck slightly improves the accuracy in the video split. However, the difference is marginal since MovieQA often requires character-based grounding rather than general visual matching. Finally, we experiment with the null hypothesis: No Context tests whether GPT-3 solves MovieQA by simply memorizing every fact. No Context performs worse than LSS, rejecting the null hypothesis. PororoQA [13] is a video story QA dataset built from a cartoon series. The supervised baseline takes the human-generated plot and the ground truth video segment index, while LSS +Plot+Search takes neither. PororoQA Table 2 summarizes our result on the PororoQA dataset. When using both the ground-truth episode and plots, GPT-3 performs almost on par with the supervised baseline. Substituting a human-generated summary with a model-generated one results in only a marginal performance drop. Perhaps intriguingly, the search process works better when using model-generated plots. We attribute this result to the fact that the human annotations are not designed for episode discriminability. 3.2. Evaluating CLIPCheck DramaQA [3] is video QA dataset that focuses on story understanding. The dataset is arranged with four levels of hierarchical difficulty, which follow the human cognitivedevelopmental stages. We evaluate LSS on the two high levels of DramaQA to test plot understanding. We report two latest baselines in level-wise DramaQA; CharacterAttention and Kim et al. [14]. DramaQA We compare the effect of CLIPCheck and Caption, a prompt-based method of incorporating image frame descriptions extracted from BLIP [18] as inputs to GPT-3. Table 3 shows that CLIPCheck offers greater improvement than image descriptions. Also, while adding image captions improves LSS, the gain disappears when used jointly with CLIPCheck. We suspect that this is because frame captions provide similar information to CLIPCheck while being much noisier. Note that the automatic Captions here are not an integral component of LSS. As DramaQA has visually grounded annotations already, adding automatic image Captions on top of that would not necessarily improve the model performance. Rather, we use the Captions to explicitly compare early vs. late visual alignment methods. Finally, we check whether CLIPCheck exploits the dataset bias rather than understanding the visual context. To this end, we devise a variant of CLIPCheck with random visual context (CLIPCheck-Shuffle). CLIPCheck-Shuffle does not improve over LSS with no CLIPCheck, denying the bias hypothesis. 3.3. Ablation Study Are both the summarization and search important for narrative understanding? Here, we evaluate LSS variants with full context without the narrative search (LSS-Full) or with the plot summary and random segment as inputs (LSS-Random). Table 4 shows that both LSS-Full and LSS-Random fall behind LSS-Search, indicating the importance of retrieval. Note that we could not employ the full context in LSS-Full due to the token length limitation. Instead, we use the longest prefix of the full context that GPT3 accepts (4000 tokens minus the length of the instruction). 3.4. Qualitative Results Figure 3 shows the automatic plot summary generated as an intermediate context of the long video QA using the language model in the LSS framework. As shown in the qualitative sample, the generated plots align well with the human-written plots from Wikipedia. For example, in the first scene of the movie "Harry Potter and the Deathly Hallows", the LSS summary correctly writes that Harry Potter is currently 17 years old and the main event in which the death eaters attack the protagonist. Figure 4 depicts the connection between the searched plot piece and the answer likelihood. In the example on the left, the retrieved summary tells that Trench committed a crime and thus is on the run, suggesting that another character interested in him would be chasing him. The language model understands this context to modify the answer likelihood in the correct way. In the right example, the LSS plot piece suggests that Edward is confident in his decision. While this context does not offer a direct cue to the question, the language model sees it as information strong enough to alter the answer.

A Summarize-then-Search Method for Long Video Question Answering: Experiments

দীর্ঘ ভিডিও প্রশ্নের উত্তরের জন্য একটি সংক্ষিপ্ত-পরে অনুসন্ধান পদ্ধতি: পরীক্ষা-নিরীক্ষা

Eine „Zusammenfassen-dann-Suchen“-Methode zum Beantworten langer Videofragen: Experimente

Un método de resumen y luego búsqueda para responder preguntas largas en video: experimentos

Une méthode de synthèse puis de recherche pour répondre à des questions vidéo longues : expériences

लंबे वीडियो प्रश्न उत्तर के लिए सारांश-फिर-खोज विधि: प्रयोग

長いビデオの質問応答のための要約してから検索する方法：実験

긴 비디오 질문 답변을 위한 요약 후 검색 방법: 실험

Um método de resumir e pesquisar para responder a perguntas longas em vídeo: experimentos

Метод суммирования и поиска для ответов на длинные видеовопросы: эксперименты

Uzun Video Soru Cevaplama için Özetle-Sonra-Arama Yöntemi: Deneyler

Phương pháp tóm tắt rồi tìm kiếm để trả lời câu hỏi bằng video dài: Thử nghiệm

长视频问答的“先总结后搜索”方法：实验

This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. Authors: (1) Minghao Yan, University of Wisconsin-Madison; (2) Hongyi Wang, Carnegie Mellon University; (3) Shivaram Venkataraman, myan@cs.wisc.edu. Table of Links Abstract & Introduction Motivation Opportunities Architecture Overview Proble Formulation: Two-Phase Tuning Modeling Workload Interference Experiments Conclusion & References A. Hardware Details B. Experimental Results C. Arithmetic Intensity D. Predictor Analysis 8 CONCLUSION In this work, we examine the unique characteristics of energy consumption in neural network inference, especially for edge devices. We identified unique tradeoffs and dimensions between energy consumption and inference latency SLOs and empirically demonstrated hidden components in optimizing energy consumption. We then propose an optimization framework that automatically and holistically tunes various hardware components to find a configuration aligned with the Pareto Frontier. We empirically verify the effectiveness and efficiency of PolyThrottle. PolyThrottle also adapts to the need for fine-tuning and proposes a simple performance prediction model to adaptively schedule finetuning requests while keeping the online inference workload under the inference latency SLO whenever possible. We hope our study sheds more light on the hidden dimension of NN energy optimization. REFERENCES Jetson partner solutions ebook. URL https://resources.nvidia.com/ en-us-jetson-success-stories/ jetson-partner-solutions-ebook?lx=XRDs_y. Twitter streaming traces, 2018. URL https://archive.org/details/ archiveteam-twitter-stream-201804. Alipourfard, O., Liu, H. H., Chen, J., Venkataraman, S., Yu, M., and Zhang, M. Cherrypick: Adaptively unearthing the best cloud configurations for big data analytics. In NSDI, volume 2, pp. 4–2, 2017. Anderson, T., Belay, A., Chowdhury, M., Cidon, A., and Zhang, I. Treehouse: A case for carbon-aware datacenter software. 2022. Anthony, L. F. W., Kanding, B., and Selvan, R. Carbontracker: Tracking and predicting the carbon footprint of training deep learning models. 2020. Arafa, Y., ElWazir, A., ElKanishy, A., Aly, Y., Elsayed, A., Badawy, A., Chennupati, G., Eidenbenz, S., and Santhi, N. Verified instruction-level energy consumption measurement for nvidia gpus. 2020. Bai, Z., Zhang, Z., Zhu, Y., and Jin, X. Pipeswitch: Fast pipelined context switching for deep learning applications. In Proceedings of the 14th USENIX Conference on Operating Systems Design and Implementation, pp. 499–514, 2020. Banner, R., Hubara, I., Hoffer, E., and Soudry, D. Scalable methods for 8-bit training of neural networks. Advances in neural information processing systems, 31, 2018. Brochu, E., Cora, V. M., and De Freitas, N. A tutorial on bayesian optimization of expensive cost functions, with application to active user modeling and hierarchical reinforcement learning. arXiv preprint arXiv:1012.2599, 2010. Cai, H., Wang, T., Wu, Z., Wang, K., Lin, J., and Han, S. On-device image classification with proxyless neural architecture search and quantization-aware fine-tuning. In Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops, pp. 0–0, 2019. Cai, H., Gan, C., Zhu, L., and Han, S. Tinytl: Reduce memory, not parameters for efficient on-device learning. Advances in Neural Information Processing Systems, 33: 11285–11297, 2020. Cao, Q., Balasubramanian, A., and Balasubramanian, N. Towards accurate and reliable energy measurement of nlp models. In Proceedings of SustaiNLP: Workshop on Simple and Efficient Natural Language Processing, 2020. Censor, Y. Pareto optimality in multiobjective problems. Applied Mathematics and Optimization, 4(1):41–59, 1977. Courbariaux, M., Bengio, Y., and David, J.-P. Training deep neural networks with low precision multiplications. arXiv preprint arXiv:1412.7024, 2014. Courbariaux, M., Bengio, Y., and David, J.-P. Binaryconnect: Training deep neural networks with binary weights during propagations. Advances in neural information processing systems, 28, 2015. Crankshaw, D., Wang, X., Zhou, G., Franklin, M. J., Gonzalez, J. E., and Stoica, I. Clipper: A low latency online prediction serving system. In NSDI, volume 17, pp. 613– 627, 2017. Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018. Gardner, J. R., Kusner, M. J., Xu, Z. E., Weinberger, K. Q., and Cunningham, J. P. Bayesian optimization with inequality constraints. In ICML, volume 2014, pp. 937–945, 2014. Gholami, A., Kim, S., Dong, Z., Yao, Z., Mahoney, M. W., and Keutzer, K. A survey of quantization methods for efficient neural network inference. arXiv preprint arXiv:2103.13630, 2021. Gog, I., Kalra, S., Schafhalter, P., Gonzalez, J. E., and Stoica, I. D3: a dynamic deadline-driven approach for building autonomous vehicles. In Proceedings of the Seventeenth European Conference on Computer Systems, pp. 453–471, 2022. Gu, D., Xie, X., Huang, G., Jin, X., and Liu, X. Energyefficient gpu clusters scheduling for deep learning. arXiv preprint arXiv:2304.06381, 2023. Gujarati, A., Karimi, R., Alzayat, S., Hao, W., Kaufmann, A., Vigfusson, Y., and Mace, J. Serving {DNNs} like clockwork: Performance predictability from the bottom up. In 14th USENIX Symposium on Operating Systems Design and Implementation (OSDI 20), pp. 443–462, 2020. Gupta, U., Kim, Y. G., Lee, S., Tse, J., Lee, H.-H. S., Wei, G.-Y., Brooks, D., and Wu, C.-J. Chasing carbon: The elusive environmental footprint of computing. IEEE Micro, 42(4):37–47, 2022. He, C., Li, S., So, J., Zeng, X., Zhang, M., Wang, H., Wang, X., Vepakomma, P., Singh, A., Qiu, H., et al. Fedml: A research library and benchmark for federated machine learning. arXiv preprint arXiv:2007.13518, 2020. Hodak, M., Gorkovenko, M., and Dholakia, A. Towards power efficiency in deep learning on data center hardware. In IEEE International Conference on Big Data, 2019. Hong, S. and Kim, H. An integrated gpu power and performance model. In ISCA, 2010. Howard, A., Sandler, M., Chu, G., Chen, L.-C., Chen, B., Tan, M., Wang, W., Zhu, Y., Pang, R., Vasudevan, V., et al. Searching for mobilenetv3. pp. 1314–1324, 2019. Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., Andreetto, M., and Adam, H. Mobilenets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861, 2017. Ivanov, A., Dryden, N., Ben-Nun, T., Li, S., and Hoefler, T. Data movement is all you need: A case study on optimizing transformers. Proceedings of Machine Learning and Systems, 3:711–732, 2021. Kandasamy, K., Vysyaraju, K. R., Neiswanger, W., Paria, B., Collins, C. R., Schneider, J., Poczos, B., and Xing, E. P. Tuning hyperparameters without grad students: Scalable and robust bayesian optimisation with dragonfly. The Journal of Machine Learning Research, 21(1):3098–3124, 2020. Kandiah, V., Peverelle, S., Khairy, M., Pan, J., Manjunath, A., Rogers, T. G., Aamodt, T. M., and Hardavellas, N. Accelwattch: A power modeling framework for modern gpus. In MICRO, 2021. Kang, D.-K., Lee, K.-B., and Kim, Y.-C. Cost efficient gpu cluster management for training and inference of deep learning. Energies, 15(2):474, 2022. Kim, S., Gholami, A., Yao, Z., Mahoney, M. W., and Keutzer, K. I-bert: Integer-only bert quantization. pp. 5506–5518. PMLR, 2021. Klein, A., Bartels, S., Falkner, S., Hennig, P., and Hutter, F. Towards efficient bayesian optimization for big data. In NIPS 2015 Bayesian Optimization Workshop, 2015. Komoda, T., Hayashi, S., Nakada, T., Miwa, S., and Nakamura, H. Power capping of cpu-gpu heterogeneous systems through coordinating dvfs and task mapping. In 2013 IEEE 31st International Conference on computer design (ICCD). IEEE, 2013. Lacoste, A., Luccioni, A., Schmidt, V., and Dandres, T. Quantifying the carbon emissions of machine learning. arXiv preprint arXiv:1910.09700, 2019. Lane, N. D. and Georgiev, P. Can deep learning revolutionize mobile sensing? In Proceedings of the 16th international workshop on mobile computing systems and applications, pp. 117–122, 2015. Lee, J., Chirkov, N., Ignasheva, E., Pisarchyk, Y., Shieh, M., Riccardi, F., Sarokin, R., Kulik, A., and Grundmann, M. On-device neural net inference with mobile gpus. arXiv preprint arXiv:1907.01989, 2019. Lowe-Power, J., Ahmad, A. M., Akram, A., Alian, M., Amslinger, R., Andreozzi, M., Armejach, A., Asmussen, N., Beckmann, B., Bharadwaj, S., et al. The gem5 simulator: Version 20.0+. arXiv preprint arXiv:2007.03152, 2020. Mei, X., Wang, Q., and Chu, X. A survey and measurement study of gpu dvfs on energy conservation. Digital Communications and Networks, 3(2):89–100, 2017. Nabavinejad, S. M., Reda, S., and Ebrahimi, M. Batchsizer: Power-performance tradeoff for dnn inference. In Proceedings of the 26th Asia and South Pacific Design Automation Conference, 2021. NVIDIA. Stream management, 2023a. URL https://docs.nvidia.com/cuda/ cuda-runtime-api. NVIDIA. Nvidia multi-instance gpu, 2023b. URL https://docs.nvidia.com/datacenter/ tesla/mig-user-guide/index.html. Peng, Y., Zhu, Y., Chen, Y., Bao, Y., Yi, B., Lan, C., Wu, C., and Guo, C. A generic communication scheduler for distributed dnn training acceleration. In SOSP, 2019. Qiao, A., Choe, S. K., Subramanya, S. J., Neiswanger, W., Ho, Q., Zhang, H., Ganger, G. R., and Xing, E. P. Pollux: Coadaptive cluster scheduling for goodput-optimized deep learning. In OSDI, 2021. Rajpurkar, P., Zhang, J., Lopyrev, K., and Liang, P. Squad: 100,000+ questions for machine comprehension of text. In EMNLP, 2016. Romero, F., Li, Q., Yadwadkar, N. J., and Kozyrakis, C. {INFaaS}: Automated model-less inference serving. In 2021 USENIX Annual Technical Conference (USENIX ATC 21), pp. 397–411, 2021. Sanh, V., Debut, L., Chaumond, J., and Wolf, T. Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108, 2019. Schwartz, R., Dodge, J., Smith, N. A., and Etzioni, O. Green ai. Commun. ACM, 63(12):54–63, 2020. Shen, H., Chen, L., Jin, Y., Zhao, L., Kong, B., Philipose, M., Krishnamurthy, A., and Sundaram, R. Nexus: A gpu cluster engine for accelerating dnn-based video analysis. In Proceedings of the 27th ACM Symposium on Operating Systems Principles, pp. 322–337, 2019. Snoek, J., Larochelle, H., and Adams, R. P. Practical bayesian optimization of machine learning algorithms. Advances in neural information processing systems, 25, 2012. Strubell, E., Ganesh, A., and McCallum, A. Energy and policy considerations for deep learning in nlp. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, 2019. Tambe, T., Hooper, C., Pentecost, L., Jia, T., Yang, E.- Y., Donato, M., Sanh, V., Whatmough, P., Rush, A. M., Brooks, D., et al. Edgebert: Sentence-level energy optimizations for latency-aware multi-task nlp inference. In MICRO, 2021. Tan, M. and Le, Q. Efficientnet: Rethinking model scaling for convolutional neural networks. pp.6105–6114. PMLR, 2019. Tang, Z., Wang, Y., Wang, Q., and Chu, X. The impact of gpu dvfs on the energy and performance of deep learning: An empirical study. In Proceedings of the Tenth ACM International Conference on Future Energy Systems, 2019. Touvron, H., Cord, M., Douze, M., Massa, F., Sablayrolles, A., and Jegou, H. Training data-efficient image transform- ´ ers & distillation through attention. pp. 10347–10357. PMLR, 2021. Venkataraman, S., Yang, Z., Franklin, M., Recht, B., and Stoica, I. Ernest: Efficient performance prediction for {Large-Scale} advanced analytics. In 13th USENIX Symposium on Networked Systems Design and Implementation (NSDI 16), pp. 363–378, 2016. Wan, C., Santriaji, M., Rogers, E., Hoffmann, H., Maire, M., and Lu, S. Alert: Accurate learning for energy and timeliness. In ATC, 2020. Wang, F., Zhang, W., Lai, S., Hao, M., and Wang, Z. Dynamic gpu energy optimization for machine learning training workloads. IEEE Transactions on Parallel and Distributed Systems, 2021. Wang, G., Venkataraman, S., Phanishayee, A., Devanur, N., Thelin, J., and Stoica, I. Blink: Fast and generic collectives for distributed ml. In Proceedings of Machine Learning and Systems, 2020a. Wang, Y., Wang, Q., Shi, S., He, X., Tang, Z., Zhao, K., and Chu, X. Benchmarking the performance and energy efficiency of ai accelerators for ai training. In 20th IEEE/ACM International Symposium on Cluster, Cloud and Internet Computing (CCGRID), 2020b. Wolf, T., Debut, L., Sanh, V., Chaumond, J., Delangue, C., Moi, A., Cistac, P., Rault, T., Louf, R., Funtowicz, M., Davison, J., Shleifer, S., von Platen, P., Ma, C., Jernite, Y., Plu, J., Xu, C., Scao, T. L., Gugger, S., Drame, M., Lhoest, Q., and Rush, A. Transformers: State-of-the-art natural language processing. In EMNLP, 2020. Wu, C.-J., Raghavendra, R., Gupta, U., Acun, B., Ardalani, N., Maeng, K., Chang, G., Aga, F., Huang, J., Bai, C., Gschwind, M., Gupta, A., Ott, M., Melnikov, A., Candido, S., Brooks, D., Chauhan, G., Lee, B.,Lee, H.-H., Akyildiz, B., Balandat, M., Spisak, J., Jain, R., Rabbat, M., and Hazelwood, K. Sustainable ai: Environmental implications, challenges and opportunities. In Proceedings of Machine Learning and Systems, 2022. Wu, X., Rao, J., Chen, W., Huang, H., Ding, C., and Huang, H. Switchflow: preemptive multitasking for deep learning. In Proceedings of the 22nd Inter Xu, M., Liu, J., Liu, Y., Lin, F. X., Liu, Y., and Liu, X. A first look at deep learning apps on smartphones. In The World Wide Web Conference, WWW ’19, pp. 2125–2136, 2019. You, J., Chung, J.-W., and Chowdhury, M. Zeus: Understanding and optimizing gpu energy consumption of dnn training. arXiv preprint arXiv:2208.06102, 2022. Yu, P. and Chowdhury, M. Fine-grained gpu sharing primitives for deep learning applications. Proceedings of Machine Learning and Systems, 2:98–111, 2020. Zhao, Y., Liu, X., Liu, S., Li, X., Zhu, Y., Huang, G., Liu, X., and Jin, X. Muxflow: Efficient and safe gpu sharing in large-scale production deep learning clusters. arXiv preprint arXiv:2303.13803, 2023.

Stories, News & Mentions about ALLSTATE

Allstate’s Stories on HackerNoon

Allstate's News & Mentions