Opening the roof at AT&T Stadium before Monday night's Texans-Cowboys game suddenly became dangerous.

Monday Night Football: Metal fragment falls from AT&T Stadium roof after opening before Texans-Cowboys game - Yahoo Sports

The Latest News and Updates in brought to you by the team at WTRF:

AT&T is providing extra safety support for families this Cybersecurity 
Awareness Month

IntelBroker claims the breach impacts Microsoft, SAP, AT&T, Verizon, T-Mobile US, and more

Cisco confirms ongoing probe into alleged data breach • The Register

According to Nvidia, AT&T is teaming with Quantiphi to create a conversational AI platform to assist employees with software development and financial services tasks.

Nvidia highlights AI software and services at D.C. AI Summit

Realty Income, AT&T, and Opera look like undervalued income plays.

3 Dividend Stocks That Are Screaming Buys in October

AMARILLO, Texas (KAMR/KCIT) — With the growing number of scams coming into our lives through our phones and the internet AT&T give some helpful tips to stay ahead of the scams. Go to their webs…

AT&T Tips on Cybersecurity

Arkansas' regular season success predicated on performance in AT&T Stadium, needs offensive line to generate push in rushing attack.

Early ESPN Analytics Prediction for Hogs-Aggies Matchup

(Bloomberg) -- AT&T Inc. agreed to pay $13 million to settle an investigation by the US Federal Communications Commission into whether the telecommunications company failed to protect customer data that was stolen when a cloud vendor was hacked last year. Most Read from BloombergCalifornia’s ...

AT&T Agrees to Pay $13 Million FCC Fine For Cloud Data Leak

Telecom operator AT&T and its joint-venture partner TPG are in early-stage talks to merge their DirecTV satellite TV service with EchoStar owned Dish, a person familiar with the matter told Reuters on Friday. The two companies first attempted to merge back in 2002 when the U.S. Justice Department ...

US satellite-TV providers DirecTV and Dish are in talks to merge again, source says

AT&T

TORTOLA, British Virgin Island, October 25th, 2024/Chainwire/--Aark Digital recently experienced a security incident in which an unauthorized party accessed and withdrew approximately 1,386,085 USDC and 24.143 ETH. In response, the project has launched 24/7 recovery efforts, while also implementing additional security measures to safeguard against any further unauthorized activity. As part of its recovery strategy, Aark Digital is offering a 15% bounty to the individual responsible, contingent upon the safe and complete return of the misappropriated assets. This bounty, amounting to 225,000 USDC, is intended to incentivize the return of the funds, and additional details are available on Arkham's bounty platform here. Should a full recovery not occur, Aark Digitial has indicated that it will announce a distribution plan to address the situation appropriately. The organization has set a deadline of 26th October 2024, 15:00 UTC, for the responsible party to respond, stating that it is prepared to pursue legal measures if the funds are not returned within this timeframe. Aark Digital is committed to keeping its community informed throughout this process, emphasizing its dedication to transparency and user security. Aark Digital will continue to release updates on the situation as they become available. About Aark Digital Aark Digital is a blockchain-focused project dedicated to providing secure, innovative solutions within the digital asset space. Committed to transparency and user security, Aark Digital leverages cutting-edge technology and industry best practices to safeguard assets and build trust within its community. Aark Digital aims to advance the growth and adoption of decentralized finance worldwide. Contact Manager Henry Aark Digital henry@aark.digital This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program here TORTOLA, British Virgin Island, October 25th, 2024/Chainwire/--Aark Digital recently experienced a security incident in which an unauthorized party accessed and withdrew approximately 1,386,085 USDC and 24.143 ETH. In response, the project has launched 24/7 recovery efforts, while also implementing additional security measures to safeguard against any further unauthorized activity. As part of its recovery strategy, Aark Digital is offering a 15% bounty to the individual responsible, contingent upon the safe and complete return of the misappropriated assets. This bounty, amounting to 225,000 USDC, is intended to incentivize the return of the funds, and additional details are available on Arkham's bounty platform here . here here Should a full recovery not occur, Aark Digitial has indicated that it will announce a distribution plan to address the situation appropriately. The organization has set a deadline of 26th October 2024, 15:00 UTC, for the responsible party to respond, stating that it is prepared to pursue legal measures if the funds are not returned within this timeframe. Aark Digital is committed to keeping its community informed throughout this process, emphasizing its dedication to transparency and user security. Aark Digital will continue to release updates on the situation as they become available. About Aark Digital Aark Digital is a blockchain-focused project dedicated to providing secure, innovative solutions within the digital asset space. Committed to transparency and user security, Aark Digital leverages cutting-edge technology and industry best practices to safeguard assets and build trust within its community. Aark Digital aims to advance the growth and adoption of decentralized finance worldwide. Aark Digital Aark Digital Contact Manager Henry Aark Digital henry@aark.digital This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program here This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program here here

Aark Digital Offers 15% Bounty to Hacker Responsible for $1.5M Attack on Vaults

Aark Digital tilbyder 15 % dusør til hacker ansvarlig for $1,5 mio. angreb på vaults

Aark Digital ofrece una recompensa del 15 % al pirata informático responsable de un ataque de 1,5 millones de dólares a bóvedas

Aark Digital מציעה פרס של 15% להאקר האחראי להתקפה של 1.5 מיליון דולר על כספות

आर्क डिजिटल ने वॉल्ट पर $1.5 मिलियन के हमले के लिए जिम्मेदार हैकर को 15% इनाम देने की पेशकश की

Aark Digital-ն առաջարկում է 15% պարգև հակերին, որը պատասխանատու է պահոցների վրա 1,5 միլիոն դոլար հարձակման համար

Aark Digital、金庫への150万ドル攻撃の犯人に15%の賞金を提供

Aark Digital қоймаларға жасалған 1,5 миллион долларлық шабуылға жауапты хакерлерге 15% сыйақы ұсынады.

Aark Digital Epesi 15% Bounty na Hacker Responsable ya Attaque ya $1.5M na ba Coffres-forts

Aark Digital нь агуулах руу 1.5 сая долларын дайралт хийсэн хакеруудад 15%-ийн шагнал санал болгож байна.

Aark Digital oferece 15% de recompensa para hacker responsável por ataque de US$ 1,5 milhão em cofres

Aark Digital ба ҳакерҳо, ки барои ҳамлаи 1,5 миллион доллар ба анборҳо масъуланд, 15% мукофот пешниҳод мекунад

Aark Digital 向对金库发动 150 万美元攻击的黑客提供 15% 的赏金

Last week, Gotbit's CEO was arrested in Portugal at the request of the U.S. The company was charged with market manipulation and wash trading. Wash trading occurs when one person (or company) acts as both the buyer and seller in the same transaction. At first glance, it makes no sense. However, the goal of wash trading is to either boost the price of an instrument (by constantly buying at higher prices) or to boost trading volumes in an illiquid asset. Many traders and algorithms use price and volume patterns to find attractive investments. If an instrument goes up in price at a high trading volume, it will attract real traders and investors, providing the illicit actor with an opportunity to cash out by unloading his initial position into new money at high prices. That’s why wash trading is prohibited. At first glance, the Gotbit case looks like a trivial fraud case that happened in crypto. However, several days after Gotbit CEO was arrested, the SEC filed a lawsuit against a major crypto market maker, Cumberland. The SEC charges Cumberland for “operating as an unregistered dealer in the crypto asset markets.” According to the SEC, Cumberland bought and sold crypto assets offered and sold as securities for its own accounts as part of its regular business. As usual, the SEC treats most cryptos as securities and denies the industry’s view, which implies that the sales of crypto assets are similar to the sales of commodities. The SEC has been long trying to increase its control over the crypto markets. This time, the SEC has decided to focus on market makers. The market maker function is extremely important in the markets. Market makers serve as liquidity providers of last resort and smoothen price fluctuations. The cases of Gotbit and Cumberland have just one similarity — both companies were market makers. While Gotbit was accused of outright criminal activity, the claims against Cumberland are centered around the fact that the company did not register as a securities dealer with the regulator. Given the role market makers play in financial markets, it’s safe to say that the SEC continues to boost its control over crypto and tries to limit the pace of crypto market development. Any restrictions on market makers’ activities will hurt crypto market liquidity. Should we worry about this scenario? In Gotbit’s case, we are talking about scam practices in illiquid coins. The elimination of such actors will improve the marketplace for all. In Cumberland’s case, we are looking at a legitimate market maker. At first glance, the company’s business is not at risk, since the SEC accuses the company of violating registration rules rather than fraud. Apparently, the case of SEC vs Cumberland could drag on for a long time, similar to the seemingly endless SEC vs Ripple. It should be noted that the market has barely noticed the SEC’s latest attempt to put pressure on crypto, which indicates that investors have already developed immunity to the SEC’s FUD. Last week, Gotbit's CEO was arrested in Portugal at the request of the U.S. The company was charged with market manipulation and wash trading. Wash trading occurs when one person (or company) acts as both the buyer and seller in the same transaction. At first glance, it makes no sense. However, the goal of wash trading is to either boost the price of an instrument (by constantly buying at higher prices) or to boost trading volumes in an illiquid asset. Many traders and algorithms use price and volume patterns to find attractive investments. If an instrument goes up in price at a high trading volume, it will attract real traders and investors, providing the illicit actor with an opportunity to cash out by unloading his initial position into new money at high prices. That’s why wash trading is prohibited. At first glance, the Gotbit case looks like a trivial fraud case that happened in crypto. However, several days after Gotbit CEO was arrested, the SEC filed a lawsuit against a major crypto market maker, Cumberland. lawsuit lawsuit The SEC charges Cumberland for “operating as an unregistered dealer in the crypto asset markets.” According to the SEC, Cumberland bought and sold crypto assets offered and sold as securities for its own accounts as part of its regular business. As usual, the SEC treats most cryptos as securities and denies the industry’s view, which implies that the sales of crypto assets are similar to the sales of commodities. The SEC has been long trying to increase its control over the crypto markets. This time, the SEC has decided to focus on market makers. The market maker function is extremely important in the markets. Market makers serve as liquidity providers of last resort and smoothen price fluctuations. The cases of Gotbit and Cumberland have just one similarity — both companies were market makers. While Gotbit was accused of outright criminal activity, the claims against Cumberland are centered around the fact that the company did not register as a securities dealer with the regulator. Given the role market makers play in financial markets, it’s safe to say that the SEC continues to boost its control over crypto and tries to limit the pace of crypto market development. Any restrictions on market makers’ activities will hurt crypto market liquidity. Should we worry about this scenario? In Gotbit’s case, we are talking about scam practices in illiquid coins. The elimination of such actors will improve the marketplace for all. In Cumberland’s case, we are looking at a legitimate market maker. At first glance, the company’s business is not at risk, since the SEC accuses the company of violating registration rules rather than fraud. Apparently, the case of SEC vs Cumberland could drag on for a long time, similar to the seemingly endless SEC vs Ripple. It should be noted that the market has barely noticed the SEC’s latest attempt to put pressure on crypto, which indicates that investors have already developed immunity to the SEC’s FUD.

Why Does the SEC Attack Crypto Market Makers?

Favor immutability by converting attributes to sets TL;DR: Using sets for attributes simplifies your code and makes state management easier Problems Addressed Mutability
Complexity
Attributes become polluted
Setters Related Code Smells https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-vii-8dk31x0 https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-vi-cmj31om Steps Identify attributes representing states
Replace the attributes with sets: one for each state
Adjust methods to move items between sets instead of mutating attributes Sample Code Before class Bill {
  amount: number;
  paid: boolean;

  constructor(amount: number) {
    this.amount = amount;
    this.paid = false;
  }

  pay() {
    if (!this.paid) {
      this.paid = true;
    }
  }
}

const bill = new Bill(100);
console.log(bill.paid); // false
bill.pay();
console.log(bill.paid); // true After // 1. Identify attributes representing states

class Accountant {  
   // 2. Replace the attributes with sets: one for each state
  unpaidBills: Set<Bill>;
  paidBills: Set<Bill>;

  constructor() {
    this.unpaidBills = new Set();
    this.paidBills = new Set();
  }

  addBill(bill: Bill) {
    this.unpaidBills.add(bill);
  }

  payBill(bill: Bill) {    
    // 3. Adjust methods to move items
    // between sets instead of mutating attributes
    if (this.unpaidBills.has(bill)) {
      this.unpaidBills.delete(bill);
      this.paidBills.add(bill);
    }
  }
}

class Bill {
  amount: number;

  constructor(amount: number) {
    this.amount = amount;
  }
}

const bill = new Bill(100);
const accountant = new Accountant();
manager.addBill(bill);
console.log(accountant.unpaidBills.has(bill)); // true
manager.payBill(bill);
console.log(accountant.paidBills.has(bill)); // true Type [x]Semi-Automatic Safety This refactoring is safe when your attributes don't rely on specific indexing behavior. Since sets don't maintain element order, check if your logic depends on order. Why is the code better? Entities are immutable in the essence. Using sets ensures uniqueness and simplifies logic. You no longer need to check for duplicates before adding elements. Operations like union, intersection, and difference become straightforward, making your code more maintainable and flexible. Limitations Sets don't preserve element order. If your logic depends on sequence, converting to a set may not be appropriate and you should use an Ordered Collection or Array AI Refactoring You can prompt your AI assistants to make this refactoring for you. Try Them! Without Proper Instructions

With Specific Instructions



ChatGPT

ChatGPT



Claude

Claude



Perplexity

Perplexity



Copilot

Copilot



Gemini

Gemini Tags Mutability Related Refactorings https://hackernoon.com/refactoring-remove-setters-codesmell?embedable=true See also https://hackernoon.com/is-it-crystal-clear-for-everybody-that-a-date-should-not-mutate-wuoy3z03?embedable=true Credits Image by Angelo Giordano in Pixabay This article is part of the Refactoring Series. https://maximilianocontieri.com/how-to-improve-your-code-with-easy-refactorings?embedable=true Favor immutability by converting attributes to sets Favor immutability by converting attributes to sets TL;DR: Using sets for attributes simplifies your code and makes state management easier TL;DR: Using sets for attributes simplifies your code and makes state management easier Problems Addressed Mutability Complexity Attributes become polluted Setters Mutability Complexity Attributes become polluted Setters Related Code Smells https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-vii-8dk31x0 https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-vii-8dk31x0 https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-vi-cmj31om https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-vi-cmj31om Steps Identify attributes representing states Replace the attributes with sets: one for each state Adjust methods to move items between sets instead of mutating attributes Identify attributes representing states Replace the attributes with sets: one for each state Adjust methods to move items between sets instead of mutating attributes Sample Code Before class Bill {
  amount: number;
  paid: boolean;

  constructor(amount: number) {
    this.amount = amount;
    this.paid = false;
  }

  pay() {
    if (!this.paid) {
      this.paid = true;
    }
  }
}

const bill = new Bill(100);
console.log(bill.paid); // false
bill.pay();
console.log(bill.paid); // true class Bill {
  amount: number;
  paid: boolean;

  constructor(amount: number) {
    this.amount = amount;
    this.paid = false;
  }

  pay() {
    if (!this.paid) {
      this.paid = true;
    }
  }
}

const bill = new Bill(100);
console.log(bill.paid); // false
bill.pay();
console.log(bill.paid); // true After // 1. Identify attributes representing states

class Accountant {  
   // 2. Replace the attributes with sets: one for each state
  unpaidBills: Set<Bill>;
  paidBills: Set<Bill>;

  constructor() {
    this.unpaidBills = new Set();
    this.paidBills = new Set();
  }

  addBill(bill: Bill) {
    this.unpaidBills.add(bill);
  }

  payBill(bill: Bill) {    
    // 3. Adjust methods to move items
    // between sets instead of mutating attributes
    if (this.unpaidBills.has(bill)) {
      this.unpaidBills.delete(bill);
      this.paidBills.add(bill);
    }
  }
}

class Bill {
  amount: number;

  constructor(amount: number) {
    this.amount = amount;
  }
}

const bill = new Bill(100);
const accountant = new Accountant();
manager.addBill(bill);
console.log(accountant.unpaidBills.has(bill)); // true
manager.payBill(bill);
console.log(accountant.paidBills.has(bill)); // true // 1. Identify attributes representing states

class Accountant {  
   // 2. Replace the attributes with sets: one for each state
  unpaidBills: Set<Bill>;
  paidBills: Set<Bill>;

  constructor() {
    this.unpaidBills = new Set();
    this.paidBills = new Set();
  }

  addBill(bill: Bill) {
    this.unpaidBills.add(bill);
  }

  payBill(bill: Bill) {    
    // 3. Adjust methods to move items
    // between sets instead of mutating attributes
    if (this.unpaidBills.has(bill)) {
      this.unpaidBills.delete(bill);
      this.paidBills.add(bill);
    }
  }
}

class Bill {
  amount: number;

  constructor(amount: number) {
    this.amount = amount;
  }
}

const bill = new Bill(100);
const accountant = new Accountant();
manager.addBill(bill);
console.log(accountant.unpaidBills.has(bill)); // true
manager.payBill(bill);
console.log(accountant.paidBills.has(bill)); // true Type [x]Semi-Automatic [x] Semi-Automatic [x] Safety This refactoring is safe when your attributes don't rely on specific indexing behavior. Since sets don't maintain element order, check if your logic depends on order. Why is the code better? Entities are immutable in the essence. Using sets ensures uniqueness and simplifies logic. You no longer need to check for duplicates before adding elements. Operations like union, intersection, and difference become straightforward, making your code more maintainable and flexible. Limitations Sets don't preserve element order. If your logic depends on sequence, converting to a set may not be appropriate and you should use an Ordered Collection or Array AI Refactoring You can prompt your AI assistants to make this refactoring for you. Try Them! Without Proper Instructions

With Specific Instructions



ChatGPT

ChatGPT



Claude

Claude



Perplexity

Perplexity



Copilot

Copilot



Gemini

Gemini Without Proper Instructions

With Specific Instructions ChatGPT

ChatGPT Claude

Claude Perplexity

Perplexity Copilot

Copilot 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 Copilot Copilot Copilot Copilot Copilot Copilot Copilot Copilot Gemini Gemini Gemini Gemini Gemini Gemini Gemini Gemini Tags Mutability Mutability Related Refactorings https://hackernoon.com/refactoring-remove-setters-codesmell?embedable=true https://hackernoon.com/refactoring-remove-setters-codesmell?embedable=true See also 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 Angelo Giordano in Pixabay Angelo Giordano Pixabay This article is part of the Refactoring Series. https://maximilianocontieri.com/how-to-improve-your-code-with-easy-refactorings?embedable=true https://maximilianocontieri.com/how-to-improve-your-code-with-easy-refactorings?embedable=true

 Refactoring 017 - Convert Attributes to Sets

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 3.2 Anchor-based Self-Attention Networks Anchor-based Attention Masks. To accomplish this, we devise anchor-based attention masks, as illustrated in Figure 2. Assuming that the current token in the sequence is a non-anchor token, we allow attention towards previous non-anchor tokens within the same sequence and anchor tokens from preceding sequences, while blocking attention towards non-anchor tokens from previous sequences. This approach ensures that non-anchor tokens can only access information from anchor tokens in previous sequences and the current sequence’s information. Conversely, when the current token is an anchor token, which is the last token in the sequence, we exclusively permit its attention towards previous non-anchor tokens within the same sequence, blocking all other attention. This constraint forces the anchor token to aggregate information solely from its current sequence. Consequently, we replace Eq. (3) with anchor-based attention masks in Eq. (4) to determine the mask of the i-th token in the input text concerning the j-th token (assuming that the i-th token belongs to the k-th sequence). Anchor Token Selection. By implementing the AnSAN mechanism for training LLMs, we can compel the model to compress sequence information into the anchor token and generate new tokens based on the anchor token information from previous sequences and non-anchor token information from the current sequence. The challenge now lies in selecting an appropriate anchor token. In our experiment, we propose two implementation methods: one using the endpoint as the anchor token, and the other appending a new token specifically as the anchor token. 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 3.2 Anchor-based Self-Attention Networks Anchor-based Attention Masks. To accomplish this, we devise anchor-based attention masks, as illustrated in Figure 2. Assuming that the current token in the sequence is a non-anchor token, we allow attention towards previous non-anchor tokens within the same sequence and anchor tokens from preceding sequences, while blocking attention towards non-anchor tokens from previous sequences. This approach ensures that non-anchor tokens can only access information from anchor tokens in previous sequences and the current sequence’s information. Conversely, when the current token is an anchor token, which is the last token in the sequence, we exclusively permit its attention towards previous non-anchor tokens within the same sequence, blocking all other attention. This constraint forces the anchor token to aggregate information solely from its current sequence. Consequently, we replace Eq. (3) with anchor-based attention masks in Eq. (4) to determine the mask of the i -th token in the input text concerning the j -th token (assuming that the i -th token belongs to the k -th sequence). Anchor-based Attention Masks. i j i k Anchor Token Selection. By implementing the AnSAN mechanism for training LLMs, we can compel the model to compress sequence information into the anchor token and generate new tokens based on the anchor token information from previous sequences and non-anchor token information from the current sequence. Anchor Token Selection. The challenge now lies in selecting an appropriate anchor token. In our experiment, we propose two implementation methods: one using the endpoint as the anchor token, and the other appending a new token specifically as the anchor token. 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

The Role of Anchor Tokens in Self-Attention Networks

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 3.1 Background Transformers. LLMs are primarily realized as decoder-only transformers (Vaswani et al., 2017; Touvron et al., 2023a,b), incorporating an input embedding layer and multiple decoder layers. Each layer contains a self-attention network and a feedforward network with normalization modules. Crucially, causal attention masks are employed, allowing tokens to attend only to preceding ones. Self-Attention Networks. Typically for decoderonly LLMs like Llama2 (Touvron et al., 2023b), self-attention networks (SANs) map queries Q, keys K, and values V into an output, as delineated in the following equations, where M denotes an L × L masking matrix, facilitating the current i-th token to attend to only preceding tokens whilst disregarding subsequent tokens during the training and inference phases. Keys/Values Caches. In the application of LLMs, the keys/values caches increase with lengthy prefix texts and continuously generated tokens during the inference phase, such as in question-answering (Saad-Falcon et al., 2023), text summarization (Basyal and Sanghvi, 2023), and machine translation (Pang et al., 2024). The key and value matrices associated with tokens of prefix inputs are cached to avoid recomputation and expedite subsequent token prediction (Radford et al., 2019). Additionally, the model generates the output token-by-token in the real-time inference process, which requires more cache memory to store the newly generated sequence. Therefore, addressing the challenges arising from the ever-expanding texts is crucial for enhancing the efficiency of LLM inference. 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 3.1 Background Transformers . LLMs are primarily realized as decoder-only transformers (Vaswani et al., 2017; Touvron et al., 2023a,b), incorporating an input embedding layer and multiple decoder layers. Each layer contains a self-attention network and a feedforward network with normalization modules. Crucially, causal attention masks are employed, allowing tokens to attend only to preceding ones. Transformers Self-Attention Networks. Typically for decoderonly LLMs like Llama2 (Touvron et al., 2023b), self-attention networks (SANs) map queries Q , keys K , and values V into an output, as delineated in the following equations, Self-Attention Networks. Q K V where M denotes an L × L masking matrix, facilitating the current i -th token to attend to only preceding tokens whilst disregarding subsequent tokens during the training and inference phases. M L L i Keys/Values Caches . In the application of LLMs, the keys/values caches increase with lengthy prefix texts and continuously generated tokens during the inference phase, such as in question-answering (Saad-Falcon et al., 2023), text summarization (Basyal and Sanghvi, 2023), and machine translation (Pang et al., 2024). The key and value matrices associated with tokens of prefix inputs are cached to avoid recomputation and expedite subsequent token prediction (Radford et al., 2019). Additionally, the model generates the output token-by-token in the real-time inference process, which requires more cache memory to store the newly generated sequence. Therefore, addressing the challenges arising from the ever-expanding texts is crucial for enhancing the efficiency of LLM inference. Keys/Values Caches 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

Unlocking the Mechanics of Decoder-Only Transformers and Self-Attention

Anatomy of a Multi-Block Attack

የብዝሃ-ብሎክ ጥቃት አናቶሚ

Ανατομία μιας επίθεσης πολλαπλών μπλοκ

Anatomía de un ataque de múltiples bloques

Anatomía dun ataque multibloque

A többblokkos támadás anatómiája

マルチブロック攻撃の解剖

მრავალბლოკიანი შეტევის ანატომია

Kelių blokų atakos anatomija

Олон блоктой дайралтын анатоми

Анатомия многоблочной атаки

I-Anatomy ye-Multi-Block Attack

The development of the Internet of Things has shaped our modern life, connected thousands of devices, and improved the intelligence and efficiency of everything we use, from houses to urban areas. However, this interconnection has brought some weaknesses, as criminals are hackers are increasingly using IoT devices for large-scale attacks. The emergency of botnets is among the top risks currently faced by IoT ecosystems. These networks of compromised devices can be used as tools to carry out cyber-attacks that can bring severe consequences. The need to understand how IoT botnets work, their effects, and the mechanism to reduce their threat is very important in the current interconnected world. The Development of IoT and Its Security Concerns IoT devices consist of security cameras, smart thermostats, sensors, home appliances, smart TVs, and medical devices. According to the report, it is estimated that over 75 billion IoT devices will be connected globally, a tremendous increase from the millions in use. While the devices make lives easy and convenient for human usage, they bring some set of security issues. The majority of IoT devices are built for functionality rather than security. Some come with weak default passwords, lack automatic security update installations, or operate under minimal processing power, which reduces advanced security features. The rise in these insecure devices has generated a breeding ground for manipulation. Cybercriminals take advantage of these loopholes by using malware to control vulnerable IoT devices and integrate them into botnets. Once in the hands of a hacker, these devices can be used to carry out cyber attacks without user knowledge. What is an IoT Botnet? The botnet is a group of internet-connected devices infected with malware and controlled by one or more bots called the “botmaster.” In IoT, this network comprises various internet-connected devices that have been compromised through vulnerabilities. These include weak passwords or unpatched software. The botmaster can control the compromised device remotely, making it a power tool for cyber-attacks. IoT botnets pose a significant threat because of the large quantity of IoT gadgets, while an individual IoT device may not have as much computing power as a regular computer; their effectiveness increases significantly when they are joined together due to their large numbers. Furthermore, numerous IoT devices run non-top and can go unnoticed when hacked, enabling the botnet to endure for extended periods undetected. How IoT Botnets are Weaponized in Cyber Attacks Cyber attacks of different forms have been carried out via the Internet of Things botnet, with Distributed Denial of Service attacks being the most prevalent. DDOS attacks overload a target’s network or server with heavy traffic, slowing down the service or crashing it so that legitimate users can not have access. IoT botnets can produce enormous volumes of bandwidth due to their large number of infiltrated devices, which makes these attacks more sophisticated, effective, and difficult to defend. For example, the famous Mirai botnet attack in 2016 infected hundreds of IoT devices, such as security cameras and home routers is a clear example of how dangerous this attack can be. One of the largest attacks in history, which targeted services like Netflix, Twitter, and Reddit, was caused by the Mirai botnet. The attack shows how a DDOS attack can disrupt services across different platforms. Aside from DDOS attacks, IoT botnets have been used for a variety of nefarious activities, such as email spam, data theft, click fraud, and cryptocurrency mining. Because these devices are frequently forgotten by the users, hackers then use them to build a massive infrastructure for unlawful activities, all while the device owners are ignorant of malicious activities. Vulnerabilities that Allow IoT Botnets There are different factors that contribute to the vulnerability of IoT devices, making them attractive to target for botnets: Weak Authentication: The majority of IoT devices come with a default username and password that users barely change after purchase. Hackers leverage human ignorance and use automated tools to search the internet for a gadget that still has default credentials, illegally gaining access to them without any restriction. Unpatched Firmware: Many IoT devices lack ways of obtaining automatic security upgrades. As vulnerabilities in devices and software are uncovered, a lot stay unpatched, making the devices vulnerable to attack. Limited Security Feature: Because of the economic constraints or the requirements for efficiency, many IoT devices offer few security measures. Firewalls, encryption, and advanced intrusion detection systems are often unavailable, allowing attackers to hack the devices. Inconsistent Security Standards: IoT manufacturers’ approaches to security vary and there are sometimes no industry standards in place. This gap leads to a variety of vulnerabilities across several devices. Impact of IoT Botnets The increase of IoT botnets poses serious threats to both users and business owners. An IoT botnet attack can have serious effects on an organization ranging from financial loss to reputation damage, and operational downtime. Massive DDOS attacks can shut down websites, online services, and critical infrastructure, with far-reaching consequences for global markets. National Security is under serious threat on a broader scale because more essential infrastructures such as power grids, transportation, systems, military equipment and healthcare networks, linked to IoT could be affected by IoT botnets. The possibility of botnet attacks on this critical infrastructure could cause disruption and risk lives. Defending Against IoT Botnets Fighting against the threat of IoT botnets requires a multi-dimension approach, involving all stakeholders that’s both users and manufacturers. For manufacturers, it’s imperative to design with security consciousness in mind.  This approach includes using strong usernames and passwords, ensuring that automatic firmware is enabled, and using secure coding practices. The usage of industry standards and regulations to checkmate the security features will help in reducing risks and threats. For users and enterprises, protecting IoT devices should be a top priority because everything we do depends on it. The security of the devices should be a must not a choice and the responsibility includes changing default credentials, enabling multifactor authentication (MFA), updating devices constantly, and placing IoT devices on separate networks from critical systems. The adoption of firewalls, intrusion detection systems, and monitoring of traffic for anomalous behavior helps to detect and prevent IoT botnet activity. Conclusion The increase of IoT botnets demonstrated the rise of complexity in cyber threats in the age of connected devices. As more gadgets become available, the probability of cyber-attacks targeting them will grow. Defending the weaponization of IoT cyber attacks needs a collaborative effort with manufacturers, regulators, and users. Without addressing the current weakness in IoT devices, the digital world would continue to be vulnerable to botnets capable of causing massive disruption to lives. We can reduce risks and fully realize the potential of linked devices by ensuring security improvement in IoT development and usage. The development of the Internet of Things has shaped our modern life, connected thousands of devices, and improved the intelligence and efficiency of everything we use, from houses to urban areas. However, this interconnection has brought some weaknesses, as criminals are hackers are increasingly using IoT devices for large-scale attacks. The emergency of botnets is among the top risks currently faced by IoT ecosystems. These networks of compromised devices can be used as tools to carry out cyber-attacks that can bring severe consequences. The need to understand how IoT botnets work, their effects, and the mechanism to reduce their threat is very important in the current interconnected world. The Development of IoT and Its Security Concerns The Development of IoT and Its Security Concerns IoT devices consist of security cameras, smart thermostats, sensors, home appliances, smart TVs, and medical devices. According to the report, it is estimated that over 75 billion IoT devices will be connected globally, a tremendous increase from the millions in use. While the devices make lives easy and convenient for human usage, they bring some set of security issues. The majority of IoT devices are built for functionality rather than security. Some come with weak default passwords, lack automatic security update installations, or operate under minimal processing power, which reduces advanced security features. The rise in these insecure devices has generated a breeding ground for manipulation. Cybercriminals take advantage of these loopholes by using malware to control vulnerable IoT devices and integrate them into botnets. Once in the hands of a hacker, these devices can be used to carry out cyber attacks without user knowledge. What is an IoT Botnet? What is an IoT Botnet? The botnet is a group of internet-connected devices infected with malware and controlled by one or more bots called the “botmaster.” In IoT, this network comprises various internet-connected devices that have been compromised through vulnerabilities. These include weak passwords or unpatched software. The botmaster can control the compromised device remotely, making it a power tool for cyber-attacks. IoT botnets pose a significant threat because of the large quantity of IoT gadgets, while an individual IoT device may not have as much computing power as a regular computer; their effectiveness increases significantly when they are joined together due to their large numbers. Furthermore, numerous IoT devices run non-top and can go unnoticed when hacked, enabling the botnet to endure for extended periods undetected. How IoT Botnets are Weaponized in Cyber Attacks How IoT Botnets are Weaponized in Cyber Attacks Cyber attacks of different forms have been carried out via the Internet of Things botnet, with Distributed Denial of Service attacks being the most prevalent. DDOS attacks overload a target’s network or server with heavy traffic, slowing down the service or crashing it so that legitimate users can not have access. IoT botnets can produce enormous volumes of bandwidth due to their large number of infiltrated devices, which makes these attacks more sophisticated, effective, and difficult to defend. For example, the famous Mirai botnet attack in 2016 infected hundreds of IoT devices, such as security cameras and home routers is a clear example of how dangerous this attack can be. One of the largest attacks in history, which targeted services like Netflix, Twitter, and Reddit, was caused by the Mirai botnet. The attack shows how a DDOS attack can disrupt services across different platforms. Aside from DDOS attacks, IoT botnets have been used for a variety of nefarious activities, such as email spam, data theft, click fraud, and cryptocurrency mining. Because these devices are frequently forgotten by the users, hackers then use them to build a massive infrastructure for unlawful activities, all while the device owners are ignorant of malicious activities. Vulnerabilities that Allow IoT Botnets Vulnerabilities that Allow IoT Botnets There are different factors that contribute to the vulnerability of IoT devices, making them attractive to target for botnets: Weak Authentication: The majority of IoT devices come with a default username and password that users barely change after purchase. Hackers leverage human ignorance and use automated tools to search the internet for a gadget that still has default credentials, illegally gaining access to them without any restriction. Weak Authentication: The majority of IoT devices come with a default username and password that users barely change after purchase. Hackers leverage human ignorance and use automated tools to search the internet for a gadget that still has default credentials, illegally gaining access to them without any restriction. Unpatched Firmware: Many IoT devices lack ways of obtaining automatic security upgrades. As vulnerabilities in devices and software are uncovered, a lot stay unpatched, making the devices vulnerable to attack. Unpatched Firmware: Many IoT devices lack ways of obtaining automatic security upgrades. As vulnerabilities in devices and software are uncovered, a lot stay unpatched, making the devices vulnerable to attack. Limited Security Feature: Because of the economic constraints or the requirements for efficiency, many IoT devices offer few security measures. Firewalls, encryption, and advanced intrusion detection systems are often unavailable, allowing attackers to hack the devices. Limited Security Feature: Because of the economic constraints or the requirements for efficiency, many IoT devices offer few security measures. Firewalls, encryption, and advanced intrusion detection systems are often unavailable, allowing attackers to hack the devices. Inconsistent Security Standards: IoT manufacturers’ approaches to security vary and there are sometimes no industry standards in place. This gap leads to a variety of vulnerabilities across several devices. Inconsistent Security Standards: IoT manufacturers’ approaches to security vary and there are sometimes no industry standards in place. This gap leads to a variety of vulnerabilities across several devices. Impact of IoT Botnets Impact of IoT Botnets The increase of IoT botnets poses serious threats to both users and business owners. An IoT botnet attack can have serious effects on an organization ranging from financial loss to reputation damage, and operational downtime. Massive DDOS attacks can shut down websites, online services, and critical infrastructure, with far-reaching consequences for global markets. National Security is under serious threat on a broader scale because more essential infrastructures such as power grids, transportation, systems, military equipment and healthcare networks, linked to IoT could be affected by IoT botnets. The possibility of botnet attacks on this critical infrastructure could cause disruption and risk lives. Defending Against IoT Botnets Fighting against the threat of IoT botnets requires a multi-dimension approach, involving all stakeholders that’s both users and manufacturers. For manufacturers, it’s imperative to design with security consciousness in mind.  This approach includes using strong usernames and passwords, ensuring that automatic firmware is enabled, and using secure coding practices. The usage of industry standards and regulations to checkmate the security features will help in reducing risks and threats. For users and enterprises, protecting IoT devices should be a top priority because everything we do depends on it. The security of the devices should be a must not a choice and the responsibility includes changing default credentials, enabling multifactor authentication (MFA), updating devices constantly, and placing IoT devices on separate networks from critical systems. The adoption of firewalls, intrusion detection systems, and monitoring of traffic for anomalous behavior helps to detect and prevent IoT botnet activity. Conclusion Conclusion The increase of IoT botnets demonstrated the rise of complexity in cyber threats in the age of connected devices. As more gadgets become available, the probability of cyber-attacks targeting them will grow. Defending the weaponization of IoT cyber attacks needs a collaborative effort with manufacturers, regulators, and users. Without addressing the current weakness in IoT devices, the digital world would continue to be vulnerable to botnets capable of causing massive disruption to lives. We can reduce risks and fully realize the potential of linked devices by ensuring security improvement in IoT development and usage.

Securing the Internet of Things: Addressing the Growing Threat of Botnets and Cyber Attacks

Sicherung des Internets der Dinge: Umgang mit der wachsenden Bedrohung durch Botnetze und Cyberangriffe

Proteger la Internet de las cosas: cómo afrontar la creciente amenaza de las redes de bots y los ciberataques

Sécuriser l'Internet des objets : répondre à la menace croissante des botnets et des cyberattaques

Proteggere l'Internet delle cose: affrontare la crescente minaccia delle botnet e degli attacchi informatici

Protegendo a Internet das Coisas: Abordando a crescente ameaça de botnets e ataques cibernéticos

保障物联网安全：应对日益严重的僵尸网络和网络攻击威胁

**TORRANCE, United States / California, October 3rd, 2024/CyberNewsWire/--**An exclusive live webinar will take place on October 4th at noon Eastern Time (ET), demonstrating how Criminal IP’s Attack Surface Management (ASM) can help organizations proactively detect and mitigate cyber threats. The webinar will feature a Criminal IP ASM Live Demo, providing a comprehensive view of attack surface visibility. As cyber threats continue to evolve, businesses must strengthen their defenses, and this session will outline essential steps for achieving that goal. ▶ Users can register now for the free webinar at 12PM ET on October 10 ◀ Key Takeaways: Criminal IP ASM Live Demo: Seeing ASM in action as it uncovers hidden threats and helps secure user's organization attack surface.
Comprehensive Attack Surface Visibility: Discovering how to gain real-time insights into a user's organization’s digital ecosystem, identifying vulnerabilities before they can be exploited.
Proactive Threat Detection and Mitigation: Learning how to detect risks early and implement mitigation strategies to minimize exposure.
Strengthening Cyber Defenses with ASM: Exploring how ASM can be integrated into cybersecurity frameworks to enhance protection and provide continuous defense against threats. This webinar is designed for IT professionals, security managers, and decision-makers looking to advance their cybersecurity strategies and stay ahead in today’s rapidly changing threat landscape. Users can register now to gain valuable insights on safeguarding their organization's attack surface and taking proactive steps toward a more secure future. Contact Michael Sena AI SPERA support@aispera.com This story was distributed as a release by CyberNewsWire under HackerNoon’s Business Blogging Program. Learn more about the program here **TORRANCE, United States / California, October 3rd, 2024/CyberNewsWire/--**An exclusive live webinar will take place on October 4th at noon Eastern Time (ET), demonstrating how Criminal IP’s Attack Surface Management (ASM) can help organizations proactively detect and mitigate cyber threats. The webinar will feature a Criminal IP ASM Live Demo, providing a comprehensive view of attack surface visibility. As cyber threats continue to evolve, businesses must strengthen their defenses, and this session will outline essential steps for achieving that goal. Criminal IP ASM Criminal IP ASM ▶ Users can register now for the free webinar at 12PM ET on October 10 ◀ ▶ Users can register now for the free webinar at 12PM ET on October 10 ◀ ▶ Users can register now for the free webinar at 12PM ET on October 10 ◀ Key Takeaways: Criminal IP ASM Live Demo: Seeing ASM in action as it uncovers hidden threats and helps secure user's organization attack surface. Comprehensive Attack Surface Visibility: Discovering how to gain real-time insights into a user's organization’s digital ecosystem, identifying vulnerabilities before they can be exploited. Proactive Threat Detection and Mitigation: Learning how to detect risks early and implement mitigation strategies to minimize exposure. Strengthening Cyber Defenses with ASM: Exploring how ASM can be integrated into cybersecurity frameworks to enhance protection and provide continuous defense against threats. Criminal IP ASM Live Demo: Seeing ASM in action as it uncovers hidden threats and helps secure user's organization attack surface. Comprehensive Attack Surface Visibility: Discovering how to gain real-time insights into a user's organization’s digital ecosystem, identifying vulnerabilities before they can be exploited. Proactive Threat Detection and Mitigation: Learning how to detect risks early and implement mitigation strategies to minimize exposure. Strengthening Cyber Defenses with ASM: Exploring how ASM can be integrated into cybersecurity frameworks to enhance protection and provide continuous defense against threats. This webinar is designed for IT professionals, security managers, and decision-makers looking to advance their cybersecurity strategies and stay ahead in today’s rapidly changing threat landscape. Users can register now to gain valuable insights on safeguarding their organization's attack surface and taking proactive steps toward a more secure future. Users can register now Users can register now Contact Michael Sena AI SPERA support@aispera.com This story was distributed as a release by CyberNewsWire under HackerNoon’s Business Blogging Program. Learn more about the program here This story was distributed as a release by CyberNewsWire under HackerNoon’s Business Blogging Program. Learn more about the program here here here

Webinar: Attack Surface Management To The Rescue – Find, Fix, Fortify Your ASM With Criminal IP

Webinar: Aanvalsoppervlakbestuur tot die redding – Vind, maak reg, versterk jou ASM met kriminele IP

ندوة عبر الإنترنت: إدارة سطح الهجوم للإنقاذ - العثور على سطح الهجوم وإصلاحه وتعزيزه باستخدام الملكية الفكرية الإجرامية

Вэбінар: Attack Surface Management на дапамогу – знайдзіце, выпраўце, умацуйце свой ASM злачынным IP

Seminari web: Gestió de superfícies d'atac al rescat: trobeu, arregleu i enfortigueu el vostre ASM amb IP criminal

Seminario web: La gestión de la superficie de ataque al rescate: encuentre, repare y fortalezca su ASM con propiedad intelectual delictiva

ウェビナー: 攻撃サーフェス管理による救済 - 犯罪 IP による ASM の検出、修正、強化

Webinarium: Zarządzanie powierzchnią ataku na ratunek – znajdź, napraw, wzmocnij swój ASM za pomocą przestępczego IP

Webinar: Gerenciamento de superfície de ataque para o resgate – Encontre, conserte e fortaleça seu ASM com IP criminoso

Urubuga: Gutera Ubuyobozi bwa Surface Kubatabazi - Shakisha, Gukosora, Komeza ASM yawe hamwe na IP mpanabyaha

Webbseminarium: Attack Surface Management to the Rescue – Hitta, fixa, förstärk din ASM med kriminell IP

Вебінар: Attack Surface Management To The Rescue – Знайдіть, виправте, зміцніть свій ASM за допомогою злочинної IP-адреси

I-Webinar: Uhlaselo lwe-Surface Management ku-Rescue-Fumana, Lungisa, Yomeleza i-ASM yakho nge-IP yoLwaphulo-mthetho

Authors:
(1) Seokil Ham, KAIST;
(2) Jungwuk Park, KAIST;
(3) Dong-Jun Han, Purdue University;
(4) Jaekyun Moon, KAIST. Table of Links Abstract and 1. Introduction 2. Related Works 3. Proposed NEO-KD Algorithm and 3.1 Problem Setup: Adversarial Training in Multi-Exit Networks 3.2 Algorithm Description 4. Experiments and 4.1 Experimental Setup 4.2. Main Experimental Results 4.3. Ablation Studies and Discussions 5. Conclusion, Acknowledgement and References A. Experiment Details B. Clean Test Accuracy and C. Adversarial Training via Average Attack D. Hyperparameter Tuning E. Discussions on Performance Degradation at Later Exits F. Comparison with Recent Defense Methods for Single-Exit Networks G. Comparison with SKD and ARD and H. Implementations of Stronger Attacker Algorithms G Comparison with SKD and ARD Existing self-distillation schemes [20, 24] for multi-exit networks improve the performance on clean samples by self-distilling the knowledge of the last exit, as the last exit has the best prediction quality. Therefore, following the original philosophy, we also used the last exit in implementing the SKD baseline. Regarding ARD [8], since it was proposed for single-exit networks, we also utilized the last exit with high performance when applying ARD to multi-exit networks. Nevertheless, we perform additional experiments to consider comprehensive baselines using various exits for distillation. Table A4 above shows the results of SKD and ARD using a specific exit or an ensemble of all exits for distillation. The results show that our scheme consistently outperforms all baselines. H Implementations of Stronger Attacker Algorithms In Section 4.3 of the main manuscript, during inference, we replaced the Projected Gradient Descent (PGD) attack with other attacker algorithms (PGD-100 attack, Carlini and Wagner (CW) attack [2], and AutoAttack [5]) to generate stronger attacks for multi-exit neural networks. This section provides explanation on how these stronger attacks are implemented tailored to multi-exit neural networks. H.1 Carlini and Wagner (CW) attack The Carlini and Wagner (CW) attack is a method of generating adversarial examples designed to reduce the difference between the logits of the correct label and the largest logits among incorrect labels. In alignment with this attack strategy, we modify the CW attack for multi-exit neural networks. In the process of minimizing this difference, our modification aims to minimize the average difference across all exits of the multi-exit neural network. Moreover, when deciding whether a sample has been successfully converted into an adversarial example, we consider a sample adversarial if it misleads all exits in the multi-exit neural network. H.2 AutoAttack AutoAttack produces adversarial attacks by ensembling various attacker algorithms. For our experiment, we sequentially use APGD [5], APGD-T [5], FAB [4], and Square [1] algorithms to generate adversarial attacks, as they are commonly used. H.2.1 APGD and APGD-T The APGD attack is a modified version of the PGD attack, which is limited by its fixed step size, a suboptimal choice. The APGD attack overcomes this limitation by introducing an adaptive step size and a momentum term. Similarly, the APGD-T attack is a variation of the APGD attack where the attack perturbs a sample to change to a specific class. In this process, we use the average loss of all exits in the multi-exit neural network as the loss for computing the gradient for adversarial example updates. Moreover, we define a sample as adversarial if it misleads all exits in the multi-exit neural network. H.2.2 FAB The FAB attack creates adversarial attacks through a process involving linear approximation of classifiers, projection to the classifier hyperplane, convex combinations, and extrapolation. The FAB attack first defines a hyperplane classifier separating two classes, then finds a new adversarial example through a convex combination of the extrapolation-projected current adversarial example and the extrapolation-projected original sample with the minimum perturbation norm. Here, we use the average gradient of all exits in the multi-exit neural network as the gradient for updating adversarial examples. Similar to above, we label a sample as adversarial if it can mislead all exits in the multi-exit neural network. H.2.3 Square The Square attack generates adversarial attacks via random searches of adversarial patches with variable degrees of perturbation and position. The Square attack algorithm iteratively samples the perturbation degree and the positions of patches while reducing the size of patches. The sampled adversarial patches are added to a sample, and the patches that maximizes the loss of the target model are selected. Here, we use the average loss of all exits in the multi-exit neural network as the loss for determining whether a sampled perturbation increases or decreases the loss of the target model. Additionally, we determine a sample as adversarial if it misleads all exits in the multi-exit neural network. H.3 Experiment details For all attacks, we commonly use ϵ = 0.03 as the perturbation degree and generate adversarial examples over 50 steps. All the attacks are based on L∞ norm. For the APGD attack, we employ cross entropy loss for computing the gradient to update adversarial examples. In both the APGD-T and FAB attacks, all class pairs are considered when generating adversarial attacks. For the Square attack, the random search operation is conducted 5000 times (the number of queries). Other settings follow [14]. In terms of performance comparison against stronger attacker algorithms, we adopt adversarial training via average attack in both Adv. w/o Distill [12] and NEO-KD (our approach). However, since the original CW attack algorithm and AutoAttack algorithm were designed for single-exit neural networks, this adapted versions targeting multi-exit neural networks are relatively weak. This paper is available on arxiv under CC 4.0 license. Authors: (1) Seokil Ham, KAIST; (2) Jungwuk Park, KAIST; (3) Dong-Jun Han, Purdue University; (4) Jaekyun Moon, KAIST. Authors: Authors: (1) Seokil Ham, KAIST; (2) Jungwuk Park, KAIST; (3) Dong-Jun Han, Purdue University; (4) Jaekyun Moon, KAIST. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Related Works 2. Related Works 3. Proposed NEO-KD Algorithm and 3.1 Problem Setup: Adversarial Training in Multi-Exit Networks 3. Proposed NEO-KD Algorithm and 3.1 Problem Setup: Adversarial Training in Multi-Exit Networks 3.2 Algorithm Description 3.2 Algorithm Description 4. Experiments and 4.1 Experimental Setup 4. Experiments and 4.1 Experimental Setup 4.2. Main Experimental Results 4.2. Main Experimental Results 4.3. Ablation Studies and Discussions 4.3. Ablation Studies and Discussions 5. Conclusion, Acknowledgement and References 5. Conclusion, Acknowledgement and References A. Experiment Details A. Experiment Details B. Clean Test Accuracy and C. Adversarial Training via Average Attack B. Clean Test Accuracy and C. Adversarial Training via Average Attack D. Hyperparameter Tuning D. Hyperparameter Tuning E. Discussions on Performance Degradation at Later Exits E. Discussions on Performance Degradation at Later Exits F. Comparison with Recent Defense Methods for Single-Exit Networks F. Comparison with Recent Defense Methods for Single-Exit Networks G. Comparison with SKD and ARD and H. Implementations of Stronger Attacker Algorithms G. Comparison with SKD and ARD and H. Implementations of Stronger Attacker Algorithms G Comparison with SKD and ARD Existing self-distillation schemes [20, 24] for multi-exit networks improve the performance on clean samples by self-distilling the knowledge of the last exit, as the last exit has the best prediction quality. Therefore, following the original philosophy, we also used the last exit in implementing the SKD baseline. Regarding ARD [8], since it was proposed for single-exit networks, we also utilized the last exit with high performance when applying ARD to multi-exit networks. Nevertheless, we perform additional experiments to consider comprehensive baselines using various exits for distillation. Table A4 above shows the results of SKD and ARD using a specific exit or an ensemble of all exits for distillation. The results show that our scheme consistently outperforms all baselines. H Implementations of Stronger Attacker Algorithms In Section 4.3 of the main manuscript, during inference, we replaced the Projected Gradient Descent (PGD) attack with other attacker algorithms (PGD-100 attack, Carlini and Wagner (CW) attack [2], and AutoAttack [5]) to generate stronger attacks for multi-exit neural networks. This section provides explanation on how these stronger attacks are implemented tailored to multi-exit neural networks. H.1 Carlini and Wagner (CW) attack The Carlini and Wagner (CW) attack is a method of generating adversarial examples designed to reduce the difference between the logits of the correct label and the largest logits among incorrect labels. In alignment with this attack strategy, we modify the CW attack for multi-exit neural networks. In the process of minimizing this difference, our modification aims to minimize the average difference across all exits of the multi-exit neural network. Moreover, when deciding whether a sample has been successfully converted into an adversarial example, we consider a sample adversarial if it misleads all exits in the multi-exit neural network. H.2 AutoAttack AutoAttack produces adversarial attacks by ensembling various attacker algorithms. For our experiment, we sequentially use APGD [5], APGD-T [5], FAB [4], and Square [1] algorithms to generate adversarial attacks, as they are commonly used. H.2.1 APGD and APGD-T H.2.1 APGD and APGD-T The APGD attack is a modified version of the PGD attack, which is limited by its fixed step size, a suboptimal choice. The APGD attack overcomes this limitation by introducing an adaptive step size and a momentum term. Similarly, the APGD-T attack is a variation of the APGD attack where the attack perturbs a sample to change to a specific class. In this process, we use the average loss of all exits in the multi-exit neural network as the loss for computing the gradient for adversarial example updates. Moreover, we define a sample as adversarial if it misleads all exits in the multi-exit neural network. H.2.2 FAB H.2.2 FAB The FAB attack creates adversarial attacks through a process involving linear approximation of classifiers, projection to the classifier hyperplane, convex combinations, and extrapolation. The FAB attack first defines a hyperplane classifier separating two classes, then finds a new adversarial example through a convex combination of the extrapolation-projected current adversarial example and the extrapolation-projected original sample with the minimum perturbation norm. Here, we use the average gradient of all exits in the multi-exit neural network as the gradient for updating adversarial examples. Similar to above, we label a sample as adversarial if it can mislead all exits in the multi-exit neural network. H.2.3 Square H.2.3 Square The Square attack generates adversarial attacks via random searches of adversarial patches with variable degrees of perturbation and position. The Square attack algorithm iteratively samples the perturbation degree and the positions of patches while reducing the size of patches. The sampled adversarial patches are added to a sample, and the patches that maximizes the loss of the target model are selected. Here, we use the average loss of all exits in the multi-exit neural network as the loss for determining whether a sampled perturbation increases or decreases the loss of the target model. Additionally, we determine a sample as adversarial if it misleads all exits in the multi-exit neural network. H.3 Experiment details H.3 Experiment details For all attacks, we commonly use ϵ = 0.03 as the perturbation degree and generate adversarial examples over 50 steps. All the attacks are based on L∞ norm. For the APGD attack, we employ cross entropy loss for computing the gradient to update adversarial examples. In both the APGD-T and FAB attacks, all class pairs are considered when generating adversarial attacks. For the Square attack, the random search operation is conducted 5000 times (the number of queries). Other settings follow [14]. In terms of performance comparison against stronger attacker algorithms, we adopt adversarial training via average attack in both Adv. w/o Distill [12] and NEO-KD (our approach). However, since the original CW attack algorithm and AutoAttack algorithm were designed for single-exit neural networks, this adapted versions targeting multi-exit neural networks are relatively weak. 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

Comparison with SKD and ARD and Implementations of Stronger Attacker Algorithms

Authors:
(1) Seokil Ham, KAIST;
(2) Jungwuk Park, KAIST;
(3) Dong-Jun Han, Purdue University;
(4) Jaekyun Moon, KAIST. Table of Links Abstract and 1. Introduction 2. Related Works 3. Proposed NEO-KD Algorithm and 3.1 Problem Setup: Adversarial Training in Multi-Exit Networks 3.2 Algorithm Description 4. Experiments and 4.1 Experimental Setup 4.2. Main Experimental Results 4.3. Ablation Studies and Discussions 5. Conclusion, Acknowledgement and References A. Experiment Details B. Clean Test Accuracy and C. Adversarial Training via Average Attack D. Hyperparameter Tuning E. Discussions on Performance Degradation at Later Exits F. Comparison with Recent Defense Methods for Single-Exit Networks G. Comparison with SKD and ARD and H. Implementations of Stronger Attacker Algorithms B Clean Test Accuracy Table A1 shows the clean accuracy results using the model built upon adversarial training via maxaverage attack. We observe that NEO-KD generally shows comparable clean test accuracy with Adv. w/o Distill [12], especially on the more complicated dataset Tiny-ImageNet [17] while achieving much better adversarial test accuracy as reported in the main manuscript. C Adversarial Training via Average Attack In the main manuscript, we presented experimental results using the model trained based on maxaverage attack. Here, we also adversarially train the model via average attack [12] and measure adversarial test accuracy on CIFAR-100 dataset. Table A2 compares adversarial test accuracies of NEO-KD and other baselines against max-average attack and average attack. The overall results are consistent with the ones in the main manuscript with adversarial training via max-average attack, further confirming the advantage of NEO-KD. This paper is available on arxiv under CC 4.0 license. Authors: (1) Seokil Ham, KAIST; (2) Jungwuk Park, KAIST; (3) Dong-Jun Han, Purdue University; (4) Jaekyun Moon, KAIST. Authors: Authors: (1) Seokil Ham, KAIST; (2) Jungwuk Park, KAIST; (3) Dong-Jun Han, Purdue University; (4) Jaekyun Moon, KAIST. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Related Works 2. Related Works 3. Proposed NEO-KD Algorithm and 3.1 Problem Setup: Adversarial Training in Multi-Exit Networks 3. Proposed NEO-KD Algorithm and 3.1 Problem Setup: Adversarial Training in Multi-Exit Networks 3.2 Algorithm Description 3.2 Algorithm Description 4. Experiments and 4.1 Experimental Setup 4. Experiments and 4.1 Experimental Setup 4.2. Main Experimental Results 4.2. Main Experimental Results 4.3. Ablation Studies and Discussions 4.3. Ablation Studies and Discussions 5. Conclusion, Acknowledgement and References 5. Conclusion, Acknowledgement and References A. Experiment Details A. Experiment Details B. Clean Test Accuracy and C. Adversarial Training via Average Attack B. Clean Test Accuracy and C. Adversarial Training via Average Attack D. Hyperparameter Tuning D. Hyperparameter Tuning E. Discussions on Performance Degradation at Later Exits E. Discussions on Performance Degradation at Later Exits F. Comparison with Recent Defense Methods for Single-Exit Networks F. Comparison with Recent Defense Methods for Single-Exit Networks G. Comparison with SKD and ARD and H. Implementations of Stronger Attacker Algorithms G. Comparison with SKD and ARD and H. Implementations of Stronger Attacker Algorithms B Clean Test Accuracy Table A1 shows the clean accuracy results using the model built upon adversarial training via maxaverage attack. We observe that NEO-KD generally shows comparable clean test accuracy with Adv. w/o Distill [12], especially on the more complicated dataset Tiny-ImageNet [17] while achieving much better adversarial test accuracy as reported in the main manuscript. C Adversarial Training via Average Attack In the main manuscript, we presented experimental results using the model trained based on maxaverage attack. Here, we also adversarially train the model via average attack [12] and measure adversarial test accuracy on CIFAR-100 dataset. Table A2 compares adversarial test accuracies of NEO-KD and other baselines against max-average attack and average attack. The overall results are consistent with the ones in the main manuscript with adversarial training via max-average attack, further confirming the advantage of NEO-KD. 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

Stories, News & Mentions about ATT

Att’s Stories on HackerNoon

Att's News & Mentions