Hollywood, Health & Society (HH&S), a program of the USC Annenberg Norman Lear Center, has unveiled the winners of the 2024 Sentinel Awards, which recognize some of the past year’s best TV entertainment that responsibly informs audiences on critical topics.

"Abbott Elementary," "Expats" and "Hacks" Are Among Shows Set To Receive Sentinel Awards - SHOOTonline

USC Annenberg Norman Lear Center's Hollywood, Health & Society program will honor advocacy stories on 'Grey's Anatomy' and more with Sentinel Awards.

‘Grey’s Anatomy,’ ‘Hacks,’ ‘The Simpsons’ Among Series to Be Honored by Hollywood, Health & Society’s Sentinel Awards

The 2024 Sentinel Awards, organized by the Hollywood, Health & Society program at the USC Annenberg Norman Lear Center, will honor shows like 'Hacks.'

Sentinel Awards 2024 Winners Include 'The Morning Show,' 'Hacks,' More

Laurens County will see nearly 500 new jobs come to the area after an international automotive parts company opens their new manufacturing facility.

More than 450 jobs are coming to Laurens County as auto parts company opens facility

Real-time index price for S&P 500 Growth (IGX), along with buy or sell indicators, analysis, charts, historical performance, news and more

Consumer Discretionary Stocks Are at It Again! Don’t Miss Out on These Potential Winners!

The United Auto Workers union says it reached a tentative contract with General Motors, the last of the Detroit Three automakers to agree to a deal. Under the deal reached

UAW reaches deal with General Motors that ends strikes against Detroit automakers pending votes

The supplier of seats for the Ford Bronco has given pay raises to workers at its plant in Detroit in a bid to keep them from leaving for other jobs and to protect its business as it ramps up for the post-strike production rush.

UAW strike spurs pay raise for workers at Ford supplier

Fintel reports that on October 30, 2023, Barclays maintained coverage of Lear (NYSE:LEA) with a Overweight recommendation. Analyst Price Forecast Suggests 36.07% Upside As of October 5, 2023, the average one-year price target for Lear is 174.

Barclays Maintains Lear (LEA) Overweight Recommendation

Fintel reports that on October 30, 2023, Benchmark reiterated coverage of Lear (NYSE:LEA) with a Buy recommendation. Analyst Price Forecast Suggests 36.07% Upside As of October 5, 2023, the average one-year price target for Lear is 174.

Benchmark Reiterates Lear (LEA) Buy Recommendation

Automotive tier-1 supplier Lear Corporation has reported its results for the third quarter of 2023, with a sales increase of 10% year-on-year to US$5.8 billion.

Fifth consecutive quarter of growth for Lear Corp in Q3

Lear

lear

In the competitive world of sports betting, some use complex statistics and confusing systems to influence others' picks and rig the game to their advantage. However, a growing affiliate marketing network within the sports betting industry makes the game's rules accessible to everyone. Vault Sports employs AI-driven betting tools to empower and educate betters. Furthermore, their advanced sports betting affiliate program and dashboard make success in the sports industry an achievable dream for more entrants than ever before. With rapid user growth and a network that has reached over 7 million sports fans, Vault Sports is poised to redefine the game of sports betting for good. An Idea to Transform Sports With Machine Learning The story of Vault Sports began at a 2019 Christmas party attended by childhood friends Caleb Dykema and Zach Nichols. The two had always shared an interest in sports, but this party would transform passion into a business venture. Zach proposed an idea in which he would apply his AI expertise to create a tool that would empower the average sports better with data-driven insights, leveling the playing field and giving everyone a chance to succeed. Caleb and Zach dedicated themselves to this new business venture. Zach turned down a coaching analytics position with the Miami Dolphins to give themselves time to develop it. Caleb left his corporate job, picking up a nightly bartending gig to keep the lights on. The pair cultivated a dedicated following through engaging content on betting strategies, bankroll management, and line analysis. Responding to their community's needs, they developed the user-friendly Vault Sports app, which soared to the App Store Top 100 and amassed 40,000 downloads in just one NFL season. From Popular App to a Growing Affiliate Network Caleb and Zach continued to develop their idea of incorporating AI in sports betting. By 2023, they had achieved partnerships with popular influencers and betting sites. This was when they decided to pivot from a singular app to building their sports betting affiliate network. Vault Sports has become one of the fastest-growing affiliate networks in the business, reaching over 7 million sports fans in North America. They have also obtained nationwide betting affiliate licensing to operate as a US sportsbook affiliate. For each affiliate, Vault Sports provides a dedicated account manager for immediate assistance and an intuitive sports betting affiliate dashboard to provide clear insights into clicks, conversions, commissions, and campaign management. This next-generation affiliate marketing support allows affiliates to succeed in a small-margin industry. A Unique Approach in the Sports Betting Industry One factor that has allowed Vault Sports to stand out from the crowd and draw in so many fans is its unique approach to sports betting education. In contrast to the controversial sports touting practice in the industry, Vault Sports acts as a responsible betting affiliate. Their suite of sports betting technology empowers users to understand the dynamics of different bets rather than pushing specific picks. Another key to their success has been their focus on sports betting community building and retention marketing in modern marketing channels. Caleb and Zach saw the unleveraged potential of the massive audiences on social media. Seeing this is where sports betters live, they sought to reach audiences by utilizing sports betting influencer marketing, connecting through creators and the communities that follow them. Positioning for Future Growth and Opportunities As Vault Sports continues to garner affiliate marketing sports fans, Zach and Caleb set their sights on what lies ahead for their growing company. Their systemized approach to operations emphasizes scalability, allowing for rapid growth while maintaining quality betting affiliate support and tools. Prepared for future sports betting market growth, Zach and Caleb and their team at Vault Sports will continue their work in sports betting industry innovation that AI and a foundational love for the game power. In the competitive world of sports betting, some use complex statistics and confusing systems to influence others' picks and rig the game to their advantage. However, a growing affiliate marketing network within the sports betting industry makes the game's rules accessible to everyone. Vault Sports employs AI-driven betting tools to empower and educate betters. Furthermore, their advanced sports betting affiliate program and dashboard make success in the sports industry an achievable dream for more entrants than ever before. With rapid user growth and a network that has reached over 7 million sports fans, Vault Sports is poised to redefine the game of sports betting for good. An Idea to Transform Sports With Machine Learning The story of Vault Sports began at a 2019 Christmas party attended by childhood friends Caleb Dykema and Zach Nichols. The two had always shared an interest in sports, but this party would transform passion into a business venture. Zach proposed an idea in which he would apply his AI expertise to create a tool that would empower the average sports better with data-driven insights, leveling the playing field and giving everyone a chance to succeed. Caleb and Zach dedicated themselves to this new business venture. Zach turned down a coaching analytics position with the Miami Dolphins to give themselves time to develop it. Caleb left his corporate job, picking up a nightly bartending gig to keep the lights on. The pair cultivated a dedicated following through engaging content on betting strategies, bankroll management, and line analysis. Responding to their community's needs, they developed the user-friendly Vault Sports app, which soared to the App Store Top 100 and amassed 40,000 downloads in just one NFL season. From Popular App to a Growing Affiliate Network Caleb and Zach continued to develop their idea of incorporating AI in sports betting. By 2023, they had achieved partnerships with popular influencers and betting sites. This was when they decided to pivot from a singular app to building their sports betting affiliate network. Vault Sports has become one of the fastest-growing affiliate networks in the business, reaching over 7 million sports fans in North America. They have also obtained nationwide betting affiliate licensing to operate as a US sportsbook affiliate. For each affiliate, Vault Sports provides a dedicated account manager for immediate assistance and an intuitive sports betting affiliate dashboard to provide clear insights into clicks, conversions, commissions, and campaign management. This next-generation affiliate marketing support allows affiliates to succeed in a small-margin industry. A Unique Approach in the Sports Betting Industry One factor that has allowed Vault Sports to stand out from the crowd and draw in so many fans is its unique approach to sports betting education. In contrast to the controversial sports touting practice in the industry, Vault Sports acts as a responsible betting affiliate. Their suite of sports betting technology empowers users to understand the dynamics of different bets rather than pushing specific picks. Another key to their success has been their focus on sports betting community building and retention marketing in modern marketing channels. Caleb and Zach saw the unleveraged potential of the massive audiences on social media. Seeing this is where sports betters live, they sought to reach audiences by utilizing sports betting influencer marketing, connecting through creators and the communities that follow them. Positioning for Future Growth and Opportunities Positioning for Future Growth and Opportunities As Vault Sports continues to garner affiliate marketing sports fans, Zach and Caleb set their sights on what lies ahead for their growing company. Their systemized approach to operations emphasizes scalability, allowing for rapid growth while maintaining quality betting affiliate support and tools. Prepared for future sports betting market growth, Zach and Caleb and their team at Vault Sports will continue their work in sports betting industry innovation that AI and a foundational love for the game power.

From Machine Learning to a Sports Network of Millions

ከማሽን መማር ወደሚሊዮን የሚቆጠር የስፖርት መረብ

Ад машыннага навучання да шматмільённай спартыўнай сеткі

Del aprendizaje automático a una red deportiva de millones de personas

Մեքենայական ուսուցումից մինչև միլիոնավոր սպորտային ցանց

Dall'apprendimento automatico a una rete sportiva di milioni di persone

機械学習から数百万のスポーツネットワークへ

Машиналық оқытудан миллиондаған спорттық желіге дейін

د ماشین زده کړې څخه د ملیونونو سپورت شبکې ته

இயந்திர கற்றல் முதல் மில்லியன் கணக்கான விளையாட்டு நெட்வொர்க் வரை

ካብ ማሽን ሌርኒንግ ናብ ስፖርታዊ መርበብ ሚልዮናት

Від машинного навчання до багатомільйонної спортивної мережі

مشین لرننگ سے لے کر لاکھوں کے اسپورٹس نیٹ ورک تک

The concept of serverless ML is a new direction, leveraging loosely decoupled serverless technology for delivering core operational requirements like processing and storage in a view to enable AI-powered programs and solutions. Serverless computation platforms drive the execution and coordination of key workflows, such as feature engineering pipelines, training runs, and batch inference processes. These workflows generate deliverables such as feature sets, training sets, machine learning models, and prediction logs processed through a specific serverless feature and model store system. To wrap it off, serverless model deployment yields real-time model access via networked endpoints, with flawless online prediction capabilities. Here is How Serverless Machine Learning Pipelines Operate Machine learning pipelines function through a series of operations for processing and storing data, beginning with raw input data. That raw data is feature-engineered, utilized for training model, and culminates in driving a prediction service. A prediction service employs trained model and inference data, and a monitor system stores prediction logs for supporting debugging and observability. For a serverless ML environment, pipeline phases become stand-alone processes for feature engineering, model training, and inference. Individual pipelines coordinate with each other in terms of loading output and storing input out of a shared feature store or model registry. Prediction logs, such as feature and predicted values, can even be stored back into a feature store. That allows a model to monitor for accuracy, performance, and overall observability. What Does A Serverless Machine Learning Encompass? A machine learning workflow is built from a sequence of connected steps that build its core. The artifacts include those of data and model in nature, ranging from data collection and preprocessing to model evaluation and deployment. In other words, all these steps describe the lifecycle of an ML pipeline. In recent years, with the increased adoption of AI in commercial applications, there has emerged an entire field of MLOps with an aim to automate and streamline most of the processes concerned with the ML pipeline. The general key stages of a typical ML pipeline usually include: Data Retrieval: This is the process of locating and extracting specific data from a database based on queries provided by users or applications.
Data Preparation: The process of gathering, combining, structuring, and organizing data to prepare for any analysis or modeling.
Model Training: In machine learning, model training means providing an algorithm with features that work continuously to minimize an error in order to develop generalized representations from input data.
Model Evaluation: The model developed will be, quite appropriately, pitted against specific criteria to check their performance. This effectiveness can be measured as a performance metric, which assigns a numerical value to quantify the success of that model.
Hyperparameter Tuning: The selection of the best combination of hyperparameters is an important task in any machine learning algorithm. Hyperparameters are settings that guide the learning process and influence how the model is trained.
Model Deployment: Deploying the machine learning model in a production environment allows it to draw upon data to support real-world business decisions effectively.
Model Monitoring: Closely monitoring machine learning models in production is important for their reliability and to catch any potential issues that may arise over time. Integrating AI and ML Automates the Data Pipeline The diagram below depicts a general pipeline and warehousing system incorporating a machine-learning infrastructure for efficient information management. The pipeline begins with an intake stage, at which raw information is received from databases, external APIS, IoT sensors, and web platforms. Sources pass through a collector, in which information is compiled for processing. Information received is validated through a validator, discarding errors and discrepancies to produce high-quality, clean information for use in successive processing stages. Following intake, the pipeline then progresses toward transformation and information orchestration, in which information is cleaned and processed for warehousing and analysis. The data orchestrator ensures proper control over data motion in such a way that it will arrive at its target store at an appropriate speed. Automated processing workflows for dealing with information with an orchestrator are particularly important in high-data scenarios, as they eliminate unnecessary work and curtail the role of human errors. Information processing entails warehousing processed information in a data lake or data warehouse. Information is arranged and organized in a data catalog, and access is eased for real-time and retrospective analysis of stored information. The stored information forms the basis for training machine learning models in the ML portion. After training, saving, and referencing for inference, trained models enable the system to make estimates and draw intelligence from new streaming information. By continued performance of inference, actionable intelligence, and ever-growing complex data-intensive workloads are processed ever increasingly autonomously, with ever-growing efficiency. For access and use of information, humans query and inspect information through an information query engine. Business intelligence (BI) software, reporting platforms, and dashboards convert such information into graphical forms for insightfulness. Business intelligence tools, reporting tools, and dashboards present such information in real time for immediate consumption, and stakeholders can make educated, fact-based, and wise decisions. The model identifies an overall mechanism for information simplification, with a demonstration of how machine learning maximizes each portion—from information collection to creating meaningful and actionable insights. Industry Case Study of Serverless Machine Learning Pipeline in Healthcare McKesson, a seasoned care administration player, is leveraging its operations with effective utilization of machine learning and significant operational and care delivery improvements. With Snowflake’s AI Data Cloud, McKesson demolished walls between its information silos, and secure and unencumbered collaboration between its disparate care spheres is a reality. Real-time access for clinicians, claims, and social and economic information, key in enhancing care at a lowered expense, is a reality in its operations. Faster Data Analysis: With 75% shorter information processing times, McKesson completed pipeline run times of between one and three minutes and achieved sustained SLA. Better Patient Outcomes: With machine learning, providers have seen a 360-degree view of a patient and have delivered personalized care for individualized patient ecosystem requirements. Lower Operating Costs: With predictive analysis, operations and care management workflows have been optimized, and considerable cost savings have been achieved in overall health system operations. The concept of serverless ML is a new direction, leveraging loosely decoupled serverless technology for delivering core operational requirements like processing and storage in a view to enable AI-powered programs and solutions. Serverless computation platforms drive the execution and coordination of key workflows, such as feature engineering pipelines, training runs, and batch inference processes. These workflows generate deliverables such as feature sets, training sets, machine learning models, and prediction logs processed through a specific serverless feature and model store system. To wrap it off, serverless model deployment yields real-time model access via networked endpoints, with flawless online prediction capabilities. serverless serverless Serverless computation platforms drive the execution and coordination Serverless computation platforms drive the execution and coordination Here is How Serverless Machine Learning Pipelines Operate Here is How Serverless Machine Learning Pipelines Operate Machine learning pipelines function through a series of operations for processing and storing data, beginning with raw input data. That raw data is feature-engineered, utilized for training model, and culminates in driving a prediction service. A prediction service employs trained model and inference data, and a monitor system stores prediction logs for supporting debugging and observability. For a serverless ML environment, pipeline phases become stand-alone processes for feature engineering , model training, and inference. Individual pipelines coordinate with each other in terms of loading output and storing input out of a shared feature store or model registry. Prediction logs, such as feature and predicted values, can even be stored back into a feature store. That allows a model to monitor for accuracy, performance, and overall observability. feature engineering feature engineering What Does A Serverless Machine Learning Encompass? What Does A Serverless Machine Learning Encompass? A machine learning workflow is built from a sequence of connected steps that build its core. The artifacts include those of data and model in nature, ranging from data collection and preprocessing to model evaluation and deployment. In other words, all these steps describe the lifecycle of an ML pipeline. In recent years, with the increased adoption of AI in commercial applications, there has emerged an entire field of MLOps with an aim to automate and streamline most of the processes concerned with the ML pipeline. The general key stages of a typical ML pipeline usually include : MLOps MLOps general key stages of a typical ML pipeline usually include general key stages of a typical ML pipeline usually include Data Retrieval: This is the process of locating and extracting specific data from a database based on queries provided by users or applications. Data Preparation: The process of gathering, combining, structuring, and organizing data to prepare for any analysis or modeling. Model Training: In machine learning, model training means providing an algorithm with features that work continuously to minimize an error in order to develop generalized representations from input data. Model Evaluation: The model developed will be, quite appropriately, pitted against specific criteria to check their performance. This effectiveness can be measured as a performance metric, which assigns a numerical value to quantify the success of that model. Hyperparameter Tuning: The selection of the best combination of hyperparameters is an important task in any machine learning algorithm. Hyperparameters are settings that guide the learning process and influence how the model is trained. Model Deployment: Deploying the machine learning model in a production environment allows it to draw upon data to support real-world business decisions effectively. Model Monitoring: Closely monitoring machine learning models in production is important for their reliability and to catch any potential issues that may arise over time. Data Retrieval: This is the process of locating and extracting specific data from a database based on queries provided by users or applications. Data Retrieval: Data Preparation: The process of gathering, combining, structuring, and organizing data to prepare for any analysis or modeling. Data Preparation: Model Training: In machine learning, model training means providing an algorithm with features that work continuously to minimize an error in order to develop generalized representations from input data. Model Training: Model Evaluation: The model developed will be, quite appropriately, pitted against specific criteria to check their performance. This effectiveness can be measured as a performance metric, which assigns a numerical value to quantify the success of that model. Model Evaluation: Hyperparameter Tuning: The selection of the best combination of hyperparameters is an important task in any machine learning algorithm. Hyperparameters are settings that guide the learning process and influence how the model is trained. Hyperparameter Tuning : Hyperparameter Tuning Model Deployment: Deploying the machine learning model in a production environment allows it to draw upon data to support real-world business decisions effectively. Model Deployment: Model Monitoring: Closely monitoring machine learning models in production is important for their reliability and to catch any potential issues that may arise over time. Model Monitoring: Integrating AI and ML Automates the Data Pipeline Integrating AI and ML Automates the Data Pipeline The diagram below depicts a general pipeline and warehousing system incorporating a machine-learning infrastructure for efficient information management. The pipeline begins with an intake stage, at which raw information is received from databases, external APIS, IoT sensors, and web platforms. Sources pass through a collector, in which information is compiled for processing. Information received is validated through a validator, discarding errors and discrepancies to produce high-quality, clean information for use in successive processing stages. Following intake, the pipeline then progresses toward transformation and information orchestration, in which information is cleaned and processed for warehousing and analysis. general pipeline and warehousing system incorporating a machine-learning infrastructure general pipeline and warehousing system incorporating a machine-learning infrastructure The data orchestrator ensures proper control over data motion in such a way that it will arrive at its target store at an appropriate speed. Automated processing workflows for dealing with information with an orchestrator are particularly important in high-data scenarios, as they eliminate unnecessary work and curtail the role of human errors. Information processing entails warehousing processed information in a data lake or data warehouse. Information is arranged and organized in a data catalog, and access is eased for real-time and retrospective analysis of stored information. The stored information forms the basis for training machine learning models in the ML portion. After training, saving, and referencing for inference, trained models enable the system to make estimates and draw intelligence from new streaming information. By continued performance of inference, actionable intelligence, and ever-growing complex data-intensive workloads are processed ever increasingly autonomously, with ever-growing efficiency. For access and use of information, humans query and inspect information through an information query engine. Business intelligence (BI) software, reporting platforms, and dashboards convert such information into graphical forms for insightfulness. Business intelligence tools, reporting tools, and dashboards present such information in real time for immediate consumption, and stakeholders can make educated, fact-based, and wise decisions. The model identifies an overall mechanism for information simplification, with a demonstration of how machine learning maximizes each portion—from information collection to creating meaningful and actionable insights. Industry Case Study of Serverless Machine Learning Pipeline in Healthcare Industry Case Study of Serverless Machine Learning Pipeline in Healthcare McKesson, a seasoned care administration player, is leveraging its operations with effective utilization of machine learning and significant operational and care delivery improvements. With Snowflake’s AI Data Cloud, McKesson demolished walls between its information silos, and secure and unencumbered collaboration between its disparate care spheres is a reality. Real-time access for clinicians, claims, and social and economic information, key in enhancing care at a lowered expense, is a reality in its operations. Faster Data Analysis: With 75% shorter information processing times, McKesson completed pipeline run times of between one and three minutes and achieved sustained SLA. Faster Data Analysis: Better Patient Outcomes: With machine learning, providers have seen a 360-degree view of a patient and have delivered personalized care for individualized patient ecosystem requirements. Better Patient Outcomes: Lower Operating Costs: With predictive analysis, operations and care management workflows have been optimized, and considerable cost savings have been achieved in overall health system operations. Lower Operating Costs:

Your Machine Learning Model Doesn’t Need a Server Anymore

As a lifelong self-learner, I use all sorts of methods to learn new things, and AI is what I’m currently into. Although I’ve been in the game since 2022, my background wasn’t focused on AI. So, like everyone else, I had to do some “AI For Dummies” level study in order to get more involved. Below is a list of learning materials that I find very helpful for myself to get started with and might also be helpful for someone else in the same situation. Quick Start 1. Large Language Models Explained Briefly Type: Video


Note: Created by the popular channel 3Blue1Brown, this video uses intuitive visuals and simple language to explain how large language models (LLMs) work. Perfect for beginners who want a quick, accessible understanding of LLMs without diving into technical details. 2. Large Language Models: A Survey Type: Research paper


Note: This survey provides a high-level summary of LLM developments, making it ideal for curious learners interested in understanding the evolution and variety of these models without getting overwhelmed by book. A Comprehensive Overview of Large Language Models Type: Research paper


Note: Provides an in-depth examination of LLM architecture, training strategies, and multimodal applications. Essential for researchers and advanced practitioners seeking detailed insights into the state-of-the-art in LLMs. 4. 2024 AI Index Report Type: Comprehensive report


Note: This annual report from Stanford offers insights into the latest AI trends, including research, ethics, and global adoption. A must-read for policymakers, educators, and tech enthusiasts seeking a broad overview of AI’s impact. 5. Generative AI Overview Type: Web article
Note: Gartner’s accessible yet detailed coverage is perfect for business professionals and anyone interested in how generative AI can be applied across industries. From Basic to Advanced 1. Hands-On Large Language Models Type: Book


Note: This practical guide teaches you how to build, fine-tune, and deploy LLMs. Ideal for developers, data scientists, and AI practitioners ready to get their hands dirty. Also available on shadow libraries. 2. A Survey of Large Language Models Type: Bookish paper


Note: Dubbed more of a book than a paper, this resource dives deep into the technical architecture and methodologies of LLMs. Best suited for researchers and advanced learners. 3. The 2025 AI Engineer Reading List Type: Reading list
Note: By the popular podcast show Latent Space. A goldmine for aspiring AI engineers, this list highlights the most influential papers, books, and resources to stay ahead in AI development for this year. 4. Generative AI Handbook Type: Reading list
Note: Offers a roadmap to mastering generative AI, from fundamentals to advanced topics. Great for learners who value organized study paths. 5. Ahead of AI Magazine Type: Digital magazine
Note: Authored by AI expert Sebastian Raschka, this magazine covers technical and philosophical discussions about AI. Ideal for anyone interested in both the practical and theoretical aspects of the field. 6. Awesome List of Cybersecurity and AI Type: Reading list
Note: This curated list focuses on the intersection of AI and cybersecurity, making it invaluable for security professionals and researchers. Blogs 1. Simon Willison’s Weblog Note: Simon’s posts trending news and technical details with accessible narratives, catering to open-source developers and AI enthusiasts. 2. AI, Software, Tech, and People by X Note: Xavier Amatriain’s blog offers insightful commentary on AI’s impact on society, making it a thoughtful read for non-technical audiences. 3. The Kaitchup – AI on a Budget Note: Benjamin Marie’s posts The Weekly Kaitchup news and running open LLMs locally, ideal for small businesses, startups, and hobbyists. Podcasts 1. Around the Prompt Note: Delving into AI innovation with leading experts, uncovers untapped potential and provides practical insights for enthusiasts and newcomers alike. 2. The Gradient Note: Provides diverse perspectives on AI from researchers, builders, and users across academia, engineering, art, and entrepreneurship, making it ideal for a broad audience seeking multifaceted insights. 3. Practical AI Note: By Changelog Media, makes artificial intelligence practical, productive, and accessible to everyone. 4. The TWIML AI Podcast Note: Sam Charrington’s discussions with top industry experts on cutting-edge machine learning and AI technologies, providing valuable knowledge for researchers, engineers, and tech leaders. 5. Unsupervised Learning Note: By AI / Security Researcher and Entrepreneur, Daniel Miessler. Unsupervised Learning is a website, newsletter, and podcast about how to survive and thrive as humans in a post-AI world. 6. The Cognitive Revolution Note: A weekly podcast where host Nathan Labenz interviews AI innovators, diving into the transformative impact AI will have in the near future. 7. Latent Space Podcast Note: Top 10 US Tech podcast. Exploring AI UX, Agents, Devtools, Infra, Open Source Models for AI Engineers. Afterwords This list may get updated after time. But keep practicing while learning theory is the best way to go—using AI via self-hosting local LLMs, like my Homebrew AI Lab. It was done in 2023 and there are better tools nowadays, such as Ollama, jan and llamafile, which provide user friendly, efficiency and privacy. I found harbor as a great testbed for experimenting with different LLM stack. So there might be new post on that in the near future. As a lifelong self-learner, I use all sorts of methods to learn new things, and AI is what I’m currently into. Although I’ve been in the game since 2022, my background wasn’t focused on AI. So, like everyone else, I had to do some “AI For Dummies” level study in order to get more involved. Below is a list of learning materials that I find very helpful for myself to get started with and might also be helpful for someone else in the same situation. Quick Start Quick Start 1. Large Language Models Explained Briefly 1. Large Language Models Explained Briefly Large Language Models Explained Briefly Type: Video Note: Created by the popular channel 3Blue1Brown, this video uses intuitive visuals and simple language to explain how large language models (LLMs) work. Perfect for beginners who want a quick, accessible understanding of LLMs without diving into technical details. Type: Video Type : Video Type Note: Created by the popular channel 3Blue1Brown, this video uses intuitive visuals and simple language to explain how large language models (LLMs) work. Perfect for beginners who want a quick, accessible understanding of LLMs without diving into technical details. Note : Created by the popular channel 3Blue1Brown, this video uses intuitive visuals and simple language to explain how large language models (LLMs) work. Perfect for beginners who want a quick, accessible understanding of LLMs without diving into technical details. Note 2. Large Language Models: A Survey 2. Large Language Models: A Survey Large Language Models: A Survey Type: Research paper Note: This survey provides a high-level summary of LLM developments, making it ideal for curious learners interested in understanding the evolution and variety of these models without getting overwhelmed by book. Type: Research paper Type : Research paper Type Note: This survey provides a high-level summary of LLM developments, making it ideal for curious learners interested in understanding the evolution and variety of these models without getting overwhelmed by book. Note : This survey provides a high-level summary of LLM developments, making it ideal for curious learners interested in understanding the evolution and variety of these models without getting overwhelmed by book. Note A Comprehensive Overview of Large Language Models A Comprehensive Overview of Large Language Models A Comprehensive Overview of Large Language Models A Comprehensive Overview of Large Language Models Type: Research paper Note: Provides an in-depth examination of LLM architecture, training strategies, and multimodal applications. Essential for researchers and advanced practitioners seeking detailed insights into the state-of-the-art in LLMs. Type: Research paper Type : Research paper Type Note: Provides an in-depth examination of LLM architecture, training strategies, and multimodal applications. Essential for researchers and advanced practitioners seeking detailed insights into the state-of-the-art in LLMs. Note : Provides an in-depth examination of LLM architecture, training strategies, and multimodal applications. Essential for researchers and advanced practitioners seeking detailed insights into the state-of-the-art in LLMs. Note 4. 2024 AI Index Report 4. 2024 AI Index Report 2024 AI Index Report Type: Comprehensive report Note: This annual report from Stanford offers insights into the latest AI trends, including research, ethics, and global adoption. A must-read for policymakers, educators, and tech enthusiasts seeking a broad overview of AI’s impact. Type: Comprehensive report Type : Comprehensive report Type Note: This annual report from Stanford offers insights into the latest AI trends, including research, ethics, and global adoption. A must-read for policymakers, educators, and tech enthusiasts seeking a broad overview of AI’s impact. Note : This annual report from Stanford offers insights into the latest AI trends, including research, ethics, and global adoption. A must-read for policymakers, educators, and tech enthusiasts seeking a broad overview of AI’s impact. Note 5. Generative AI Overview 5. Generative AI Overview Generative AI Overview Type: Web article Note: Gartner’s accessible yet detailed coverage is perfect for business professionals and anyone interested in how generative AI can be applied across industries. Type : Web article Type Note : Gartner’s accessible yet detailed coverage is perfect for business professionals and anyone interested in how generative AI can be applied across industries. Note From Basic to Advanced From Basic to Advanced 1. Hands-On Large Language Models 1. Hands-On Large Language Models Hands-On Large Language Models Type: Book Note: This practical guide teaches you how to build, fine-tune, and deploy LLMs. Ideal for developers, data scientists, and AI practitioners ready to get their hands dirty. Also available on shadow libraries. Type: Book Type : Book Type Note: This practical guide teaches you how to build, fine-tune, and deploy LLMs. Ideal for developers, data scientists, and AI practitioners ready to get their hands dirty. Also available on shadow libraries. Note : This practical guide teaches you how to build, fine-tune, and deploy LLMs. Ideal for developers, data scientists, and AI practitioners ready to get their hands dirty. Also available on shadow libraries . Note shadow libraries 2. A Survey of Large Language Models 2. A Survey of Large Language Models A Survey of Large Language Models Type: Bookish paper Note: Dubbed more of a book than a paper, this resource dives deep into the technical architecture and methodologies of LLMs. Best suited for researchers and advanced learners. Type: Bookish paper Type : Bookish paper Type Note: Dubbed more of a book than a paper, this resource dives deep into the technical architecture and methodologies of LLMs. Best suited for researchers and advanced learners. Note : Dubbed more of a book than a paper, this resource dives deep into the technical architecture and methodologies of LLMs. Best suited for researchers and advanced learners. Note 3. The 2025 AI Engineer Reading List 3. The 2025 AI Engineer Reading List The 2025 AI Engineer Reading List Type: Reading list Note: By the popular podcast show Latent Space. A goldmine for aspiring AI engineers, this list highlights the most influential papers, books, and resources to stay ahead in AI development for this year. Type : Reading list Type Note : By the popular podcast show Latent Space. A goldmine for aspiring AI engineers, this list highlights the most influential papers, books, and resources to stay ahead in AI development for this year. Note 4. Generative AI Handbook 4. Generative AI Handbook Generative AI Handbook Type: Reading list Note: Offers a roadmap to mastering generative AI, from fundamentals to advanced topics. Great for learners who value organized study paths. Type : Reading list Type Note : Offers a roadmap to mastering generative AI, from fundamentals to advanced topics. Great for learners who value organized study paths. Note 5. Ahead of AI Magazine 5. Ahead of AI Magazine Ahead of AI Magazine Type: Digital magazine Note: Authored by AI expert Sebastian Raschka, this magazine covers technical and philosophical discussions about AI. Ideal for anyone interested in both the practical and theoretical aspects of the field. Type : Digital magazine Type Note : Authored by AI expert Sebastian Raschka, this magazine covers technical and philosophical discussions about AI. Ideal for anyone interested in both the practical and theoretical aspects of the field. Note 6. Awesome List of Cybersecurity and AI 6. Awesome List of Cybersecurity and AI Awesome List of Cybersecurity and AI Type: Reading list Note: This curated list focuses on the intersection of AI and cybersecurity, making it invaluable for security professionals and researchers. Type : Reading list Type Note : This curated list focuses on the intersection of AI and cybersecurity, making it invaluable for security professionals and researchers. Note Blogs Blogs 1. Simon Willison’s Weblog 1. Simon Willison’s Weblog Simon Willison’s Weblog Note: Simon’s posts trending news and technical details with accessible narratives, catering to open-source developers and AI enthusiasts. Note : Simon’s posts trending news and technical details with accessible narratives, catering to open-source developers and AI enthusiasts. Note 2. AI, Software, Tech, and People by X 2. AI, Software, Tech, and People by X AI, Software, Tech, and People by X Note: Xavier Amatriain’s blog offers insightful commentary on AI’s impact on society, making it a thoughtful read for non-technical audiences. Note : Xavier Amatriain’s blog offers insightful commentary on AI’s impact on society, making it a thoughtful read for non-technical audiences. Note 3. The Kaitchup – AI on a Budget 3. The Kaitchup – AI on a Budget The Kaitchup – AI on a Budget Note: Benjamin Marie’s posts The Weekly Kaitchup news and running open LLMs locally, ideal for small businesses, startups, and hobbyists. Note : Benjamin Marie’s posts The Weekly Kaitchup news and running open LLMs locally, ideal for small businesses, startups, and hobbyists. Note Podcasts Podcasts 1. Around the Prompt 1. Around the Prompt Around the Prompt Note: Delving into AI innovation with leading experts, uncovers untapped potential and provides practical insights for enthusiasts and newcomers alike. Note : Delving into AI innovation with leading experts, uncovers untapped potential and provides practical insights for enthusiasts and newcomers alike. Note 2. The Gradient 2. The Gradient The Gradient Note: Provides diverse perspectives on AI from researchers, builders, and users across academia, engineering, art, and entrepreneurship, making it ideal for a broad audience seeking multifaceted insights. Note : Provides diverse perspectives on AI from researchers, builders, and users across academia, engineering, art, and entrepreneurship, making it ideal for a broad audience seeking multifaceted insights. Note 3. Practical AI 3. Practical AI Practical AI Note: By Changelog Media, makes artificial intelligence practical, productive, and accessible to everyone. Note : By Changelog Media, makes artificial intelligence practical, productive, and accessible to everyone. Note 4. The TWIML AI Podcast 4. The TWIML AI Podcast The TWIML AI Podcast Note: Sam Charrington’s discussions with top industry experts on cutting-edge machine learning and AI technologies, providing valuable knowledge for researchers, engineers, and tech leaders. Note : Sam Charrington’s discussions with top industry experts on cutting-edge machine learning and AI technologies, providing valuable knowledge for researchers, engineers, and tech leaders. Note 5. Unsupervised Learning 5. Unsupervised Learning Unsupervised Learning Note: By AI / Security Researcher and Entrepreneur, Daniel Miessler. Unsupervised Learning is a website, newsletter, and podcast about how to survive and thrive as humans in a post-AI world. Note : By AI / Security Researcher and Entrepreneur, Daniel Miessler. Unsupervised Learning is a website, newsletter, and podcast about how to survive and thrive as humans in a post-AI world. Note 6. The Cognitive Revolution 6. The Cognitive Revolution The Cognitive Revolution Note: A weekly podcast where host Nathan Labenz interviews AI innovators, diving into the transformative impact AI will have in the near future. Note : A weekly podcast where host Nathan Labenz interviews AI innovators, diving into the transformative impact AI will have in the near future. Note 7. Latent Space Podcast 7. Latent Space Podcast Latent Space Podcast Note: Top 10 US Tech podcast. Exploring AI UX, Agents, Devtools, Infra, Open Source Models for AI Engineers. Note : Top 10 US Tech podcast. Exploring AI UX, Agents, Devtools, Infra, Open Source Models for AI Engineers. Note Afterwords Afterwords This list may get updated after time. But keep practicing while learning theory is the best way to go—using AI via self-hosting local LLMs, like my Homebrew AI Lab . It was done in 2023 and there are better tools nowadays, such as Ollama , jan and llamafile , which provide user friendly, efficiency and privacy. I found harbor as a great testbed for experimenting with different LLM stack. So there might be new post on that in the near future. my Homebrew AI Lab Ollama jan llamafile harbor

This Learning Web Helped Me 'Understand' What AI Was All About

Run DeepSeek R1 Locally to Learn How It 'Thinks'—Here's How

"Necə Düşündüyünü" öyrənmək üçün DeepSeek R1-i yerli olaraq işə salın - Budur

Spusťte DeepSeek R1 lokálně a zjistěte, jak to „myslí“ – zde je návod

Ejecute DeepSeek R1 localmente para aprender cómo "piensa": así es como se hace

Executa DeepSeek R1 localmente para saber como "pensa": aquí tes como

Esegui DeepSeek R1 in locale per scoprire come "pensa": ecco come

DeepSeek R1 をローカルで実行して、その「思考」の仕組みを学ぶ - 方法は次のとおりです

Стартувај DeepSeek R1 локално за да дознаеш како „размислува“ - еве како

Matha DeepSeek R1 Selegae go Ithuta Kamoo e ‘Naganago’—Kamoo ke Ye

Rulați DeepSeek R1 local pentru a afla cum „gândește” — Iată cum

එය 'හිතන' ආකාරය ඉගෙන ගැනීමට DeepSeek R1 දේශීයව ධාවනය කරන්න—මෙන්න ආකාරය

Drejtoni DeepSeek R1 në nivel lokal për të mësuar se si 'mendon'—ja se si

“Pikir edişini” öwrenmek üçin “DeepSeek R1” -ni ýerli derejede işlediň - Ine

The Surprisingly Weird Tech Making Drones Fly Smarter and Students Learn Faster

A Basic Knowledge of Python Can Help You Build Your Own Machine Learning Model

Python haqqında əsas biliklər sizə öz maşın öyrənmə modelinizi yaratmağa kömək edə bilər

Основните познания за Python могат да ви помогнат да изградите свой собствен модел за машинно обучение

Un conocimiento básico de Python puede ayudarle a crear su propio modelo de aprendizaje automático

Python-ի հիմնական գիտելիքները կարող են օգնել ձեզ կառուցել ձեր սեփական մեքենայական ուսուցման մոդելը

Una conoscenza di base di Python può aiutarti a costruire il tuo modello di apprendimento automatico

Pythonの基礎知識があれば、独自の機械学習モデルを構築できます

Um conhecimento básico de Python pode ajudar você a construir seu próprio modelo de aprendizado de máquina

Cunoștințele de bază despre Python vă pot ajuta să vă construiți propriul model de învățare automată

Ruzivo Rwekutanga rwePython Inogona Kukubatsira Iwe Kuvaka Wako Wega Machine Kudzidza Model

Njohuritë themelore të Python mund t'ju ndihmojnë të ndërtoni modelin tuaj të mësimit të makinerive

Базові знання Python можуть допомогти вам створити власну модель машинного навчання

Ulwazi oluSisiseko lwePython lunokukunceda ukuba uzenzele iModeli yakho yokuFunda ngomatshini

How to Learn AWS With Localstack and Reactive Kotlin - A Stamps and Coins Implementation

Table of Links Abstract and 1 Introduction Methodology
Experiments
Conclusion and References 2. METHODOLOGY In this section, we first introduce the problem of multi-objective ranking for recommender systems. Second, we describe the general design of DML. Finally, we elaborate on the introduced components. 2.1 Multi-Objective Ranking for Recsys 2.2 Overall Design of DML Please refer to figure 1 for the overall network architecture of DML. With the existing MTL algorithms[25], the equation (1) can be further decomposed as below. For simplicity, we omit the subscript 𝑛 in this section. In this research, rather than 𝐺, we focus on the enhancement of upper-level networks for improved prediction performance. First, the shared component of CTFM is introduced, which leverages the attention mechanism to extract relevant information from the inputs of other task towers (the results of Equation (3)) as a complement to the target task. Please note that this attention is welldesigned to solve the task-awareness missing issue, for which the excessive encouragement of knowledge sharing is not conducive to the extraction of task-specific knowledge. With our design, the gradients computed from the target task’s loss will not impact the inputs of other task towers. 2.3 Cross Task Feature Mining Authors:
(1) Yi Ren, Tencent, Beijing, China (yiren_bj@outlook.com);
(2) Ying Du, Tencent, Beijing, China (yingdu@tencent.com);
(3) Bin Wang, Tencent, Beijing, China (hillmwang@tencent.com);
(4) Shenzheng Zhang, Tencent, Beijing, China (qjzcyzhang@tencent.com). This paper is available on arxiv under CC BY 4.0 DEED license. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction Methodology Experiments Conclusion and References Methodology Methodology Experiments Experiments Conclusion and References Conclusion and References 2. METHODOLOGY In this section, we first introduce the problem of multi-objective ranking for recommender systems. Second, we describe the general design of DML. Finally, we elaborate on the introduced components. 2.1 Multi-Objective Ranking for Recsys 2.2 Overall Design of DML Please refer to figure 1 for the overall network architecture of DML. With the existing MTL algorithms[25], the equation (1) can be further decomposed as below. For simplicity, we omit the subscript 𝑛 in this section. In this research, rather than 𝐺, we focus on the enhancement of upper-level networks for improved prediction performance. First, the shared component of CTFM is introduced, which leverages the attention mechanism to extract relevant information from the inputs of other task towers (the results of Equation (3)) as a complement to the target task. Please note that this attention is welldesigned to solve the task-awareness missing issue, for which the excessive encouragement of knowledge sharing is not conducive to the extraction of task-specific knowledge. With our design, the gradients computed from the target task’s loss will not impact the inputs of other task towers. 2.3 Cross Task Feature Mining Authors: (1) Yi Ren, Tencent, Beijing, China (yiren_bj@outlook.com); (2) Ying Du, Tencent, Beijing, China (yingdu@tencent.com); (3) Bin Wang, Tencent, Beijing, China (hillmwang@tencent.com); (4) Shenzheng Zhang, Tencent, Beijing, China (qjzcyzhang@tencent.com). Authors: Authors: (1) Yi Ren, Tencent, Beijing, China (yiren_bj@outlook.com); (2) Ying Du, Tencent, Beijing, China (yingdu@tencent.com); (3) Bin Wang, Tencent, Beijing, China (hillmwang@tencent.com); (4) Shenzheng Zhang, Tencent, Beijing, China (qjzcyzhang@tencent.com). 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

Deep Mutual Learning Optimizes Multi-Task Recommender Systems with Cross Task Feature Mining

Table of Links Abstract and 1 Introduction Methodology
Experiments
Conclusion and References 4. CONCLUSION In this paper, we propose the framework of Deep Mutual Learning across task towers(DML), which is compatible with various backbone multi-task networks. Extensive offline experiments help to verify DML’s effectiveness on multiple real-world datasets and across various base models. Moreover, thorough ablation studies are carried out to verify and understand the value of each newly introduced module. Finally, DML achieves significant online gains and has already been deployed to the online platform. REFERENCES [1] Martín Abadi, Paul Barham, Jianmin Chen, Zhifeng Chen, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Geoffrey Irving, Michael Isard, et al. 2016. Tensorflow: a system for large-scale machine learning.. In Osdi, Vol. 16. Savannah, GA, USA, 265–283. [2] Raman Arora, Amitabh Basu, Poorya Mianjy, and Anirbit Mukherjee. 2018. Understanding Deep Neural Networks with Rectified Linear Units. In 6th International Conference on Learning Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Conference Track Proceedings. OpenReview.net. https://openreview.net/forum?id=B1J_rgWRW [3] Rich Caruana. 1997. Multitask learning. Machine learning 28, 1 (1997), 41–75. [4] Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep neural networks for youtube recommendations. In Proceedings of the 10th ACM conference on recommender systems. 191–198. [5] Ke Ding, Xin Dong, Yong He, Lei Cheng, Chilin Fu, Zhaoxin Huan, Hai Li, Tan Yan, Liang Zhang, Xiaolu Zhang, et al. 2021. MSSM: a multiple-level sparse sharing model for efficient multi-task learning. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2237–2241. [6] Long Duong, Trevor Cohn, Steven Bird, and Paul Cook. 2015. Low resource dependency parsing: Cross-lingual parameter sharing in a neural network parser. In Proceedings of the 53rd annual meeting of the Association for Computational Linguistics and the 7th international joint conference on natural language processing (volume 2: short papers). 845–850. [7] Peter A Flach, José Hernández-Orallo, and Cèsar Ferri Ramirez. 2011. A coherent interpretation of AUC as a measure of aggregated classification performance. In ICML. [8] F Maxwell Harper and Joseph A Konstan. 2015. The movielens datasets: History and context. Acm transactions on interactive intelligent systems (tiis) 5, 4 (2015), 1–19. [9] Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. 1991. Adaptive mixtures of local experts. Neural computation 3, 1 (1991), 79–87. [10] Diederik P. Kingma and Jimmy Ba. 2015. Adam: A Method for Stochastic Optimization. In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, Yoshua Bengio and Yann LeCun (Eds.). http://arxiv.org/abs/1412.6980 [11] Jiaqi Ma, Zhe Zhao, Jilin Chen, Ang Li, Lichan Hong, and Ed H Chi. 2019. Snr: Sub-network routing for flexible parameter sharing in multi-task learning. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 33. 216–223. [12] Jiaqi Ma, Zhe Zhao, Xinyang Yi, Jilin Chen, Lichan Hong, and Ed H Chi. 2018. Modeling task relationships in multi-task learning with multi-gate mixture-ofexperts. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 1930–1939. [13] Xiao Ma, Liqin Zhao, Guan Huang, Zhi Wang, Zelin Hu, Xiaoqiang Zhu, and Kun Gai. 2018. Entire space multi-task model: An effective approach for estimating post-click conversion rate. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval. 1137–1140. [14] Julian McAuley, Christopher Targett, Qinfeng Shi, and Anton Van Den Hengel. 2015. Image-based recommendations on styles and substitutes. In Proceedings of the 38th international ACM SIGIR conference on research and development in information retrieval. 43–52. [15] Ishan Misra, Abhinav Shrivastava, Abhinav Gupta, and Martial Hebert. 2016. Cross-stitch networks for multi-task learning. In Proceedings of the IEEE conference on computer vision and pattern recognition. 3994–4003. [16] Sebastian Ruder, Joachim Bingel, Isabelle Augenstein, and Anders Søgaard. 2019. Latent multi-task architecture learning. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 33. 4822–4829. [17] Hongyan Tang, Junning Liu, Ming Zhao, and Xudong Gong. 2020. Progressive layered extraction (ple): A novel multi-task learning (mtl) model for personalized recommendations. In Fourteenth ACM Conference on Recommender Systems. 269– 278. [18] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. Advances in neural information processing systems 30 (2017). [19] Yining Wang, Liwei Wang, Yuanzhi Li, Di He, Wei Chen, and Tie-Yan Liu. 2013. A theoretical analysis of NDCG ranking measures. In Proceedings of the 26th annual conference on learning theory (COLT 2013), Vol. 8. 6. [20] Yuyan Wang, Zhe Zhao, Bo Dai, Christopher Fifty, Dong Lin, Lichan Hong, Li Wei, and Ed H Chi. 2022. Can Small Heads Help? Understanding and Improving Multi-Task Generalization. In Proceedings of the ACM Web Conference 2022. 3009– 3019. [21] Wikipedia. 2023. Pearson correlation coefficient — Wikipedia, The Free Encyclopedia. http://en.wikipedia.org/w/index.php?title=Pearson%20correlation% 20coefficient&oldid=1146097966. [Online; accessed 15-April-2023]. [22] Wikipedia contributors. 2022. Evaluation measures (information retrieval) — Wikipedia, The Free Encyclopedia. https://en.wikipedia.org/w/index.php?title= Evaluation_measures_(information_retrieval)&oldid=1095286224. [Online; accessed 9-January-2023]. [23] Wikipedia contributors. 2022. Mean squared error — Wikipedia, The Free Encyclopedia. https://en.wikipedia.org/w/index.php?title=Mean_squared_error& oldid=1127519968. [Online; accessed 31-January-2023]. [24] Dongbo Xi, Zhen Chen, Peng Yan, Yinger Zhang, Yongchun Zhu, Fuzhen Zhuang, and Yu Chen. 2021. Modeling the sequential dependence among audience multistep conversions with multi-task learning in targeted display advertising. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. 3745–3755. [25] Yu Zhang and Qiang Yang. 2021. A survey on multi-task learning. IEEE Transactions on Knowledge and Data Engineering (2021). [26] Zhe Zhao, Lichan Hong, Li Wei, Jilin Chen, Aniruddh Nath, Shawn Andrews, Aditee Kumthekar, Maheswaran Sathiamoorthy, Xinyang Yi, and Ed Chi. 2019. Recommending what video to watch next: a multitask ranking system. In Proceedings of the 13th ACM Conference on Recommender Systems. 43–51. Authors:
(1) Yi Ren, Tencent, Beijing, China (yiren_bj@outlook.com);
(2) Ying Du, Tencent, Beijing, China (yingdu@tencent.com);
(3) Bin Wang, Tencent, Beijing, China (hillmwang@tencent.com);
(4) Shenzheng Zhang, Tencent, Beijing, China (qjzcyzhang@tencent.com). This paper is available on arxiv under CC BY 4.0 DEED license. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction Methodology Experiments Conclusion and References Methodology Methodology Experiments Experiments Conclusion and References Conclusion and References 4. CONCLUSION In this paper, we propose the framework of D eep M utual L earning across task towers( DML ), which is compatible with various backbone multi-task networks. Extensive offline experiments help to verify DML’s effectiveness on multiple real-world datasets and across various base models. Moreover, thorough ablation studies are carried out to verify and understand the value of each newly introduced module. Finally, DML achieves significant online gains and has already been deployed to the online platform. D M L DML REFERENCES [1] Martín Abadi, Paul Barham, Jianmin Chen, Zhifeng Chen, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Geoffrey Irving, Michael Isard, et al. 2016. Tensorflow: a system for large-scale machine learning.. In Osdi, Vol. 16. Savannah, GA, USA, 265–283. [2] Raman Arora, Amitabh Basu, Poorya Mianjy, and Anirbit Mukherjee. 2018. Understanding Deep Neural Networks with Rectified Linear Units. In 6th International Conference on Learning Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Conference Track Proceedings. OpenReview.net. https://openreview.net/forum?id=B1J_rgWRW [3] Rich Caruana. 1997. Multitask learning. Machine learning 28, 1 (1997), 41–75. [4] Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep neural networks for youtube recommendations. In Proceedings of the 10th ACM conference on recommender systems. 191–198. [5] Ke Ding, Xin Dong, Yong He, Lei Cheng, Chilin Fu, Zhaoxin Huan, Hai Li, Tan Yan, Liang Zhang, Xiaolu Zhang, et al. 2021. MSSM: a multiple-level sparse sharing model for efficient multi-task learning. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2237–2241. [6] Long Duong, Trevor Cohn, Steven Bird, and Paul Cook. 2015. Low resource dependency parsing: Cross-lingual parameter sharing in a neural network parser. In Proceedings of the 53rd annual meeting of the Association for Computational Linguistics and the 7th international joint conference on natural language processing (volume 2: short papers). 845–850. [7] Peter A Flach, José Hernández-Orallo, and Cèsar Ferri Ramirez. 2011. A coherent interpretation of AUC as a measure of aggregated classification performance. In ICML. [8] F Maxwell Harper and Joseph A Konstan. 2015. The movielens datasets: History and context. Acm transactions on interactive intelligent systems (tiis) 5, 4 (2015), 1–19. [9] Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. 1991. Adaptive mixtures of local experts. Neural computation 3, 1 (1991), 79–87. [10] Diederik P. Kingma and Jimmy Ba. 2015. Adam: A Method for Stochastic Optimization. In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, Yoshua Bengio and Yann LeCun (Eds.). http://arxiv.org/abs/1412.6980 [11] Jiaqi Ma, Zhe Zhao, Jilin Chen, Ang Li, Lichan Hong, and Ed H Chi. 2019. Snr: Sub-network routing for flexible parameter sharing in multi-task learning. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 33. 216–223. [12] Jiaqi Ma, Zhe Zhao, Xinyang Yi, Jilin Chen, Lichan Hong, and Ed H Chi. 2018. Modeling task relationships in multi-task learning with multi-gate mixture-ofexperts. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 1930–1939. [13] Xiao Ma, Liqin Zhao, Guan Huang, Zhi Wang, Zelin Hu, Xiaoqiang Zhu, and Kun Gai. 2018. Entire space multi-task model: An effective approach for estimating post-click conversion rate. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval. 1137–1140. [14] Julian McAuley, Christopher Targett, Qinfeng Shi, and Anton Van Den Hengel. 2015. Image-based recommendations on styles and substitutes. In Proceedings of the 38th international ACM SIGIR conference on research and development in information retrieval. 43–52. [15] Ishan Misra, Abhinav Shrivastava, Abhinav Gupta, and Martial Hebert. 2016. Cross-stitch networks for multi-task learning. In Proceedings of the IEEE conference on computer vision and pattern recognition. 3994–4003. [16] Sebastian Ruder, Joachim Bingel, Isabelle Augenstein, and Anders Søgaard. 2019. Latent multi-task architecture learning. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 33. 4822–4829. [17] Hongyan Tang, Junning Liu, Ming Zhao, and Xudong Gong. 2020. Progressive layered extraction (ple): A novel multi-task learning (mtl) model for personalized recommendations. In Fourteenth ACM Conference on Recommender Systems. 269– 278. [18] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. Advances in neural information processing systems 30 (2017). [19] Yining Wang, Liwei Wang, Yuanzhi Li, Di He, Wei Chen, and Tie-Yan Liu. 2013. A theoretical analysis of NDCG ranking measures. In Proceedings of the 26th annual conference on learning theory (COLT 2013), Vol. 8. 6. [20] Yuyan Wang, Zhe Zhao, Bo Dai, Christopher Fifty, Dong Lin, Lichan Hong, Li Wei, and Ed H Chi. 2022. Can Small Heads Help? Understanding and Improving Multi-Task Generalization. In Proceedings of the ACM Web Conference 2022. 3009– 3019. [21] Wikipedia. 2023. Pearson correlation coefficient — Wikipedia, The Free Encyclopedia. http://en.wikipedia.org/w/index.php?title=Pearson%20correlation% 20coefficient&oldid=1146097966. [Online; accessed 15-April-2023]. [22] Wikipedia contributors. 2022. Evaluation measures (information retrieval) — Wikipedia, The Free Encyclopedia. https://en.wikipedia.org/w/index.php?title= Evaluation_measures_(information_retrieval)&oldid=1095286224. [Online; accessed 9-January-2023]. [23] Wikipedia contributors. 2022. Mean squared error — Wikipedia, The Free Encyclopedia. https://en.wikipedia.org/w/index.php?title=Mean_squared_error& oldid=1127519968. [Online; accessed 31-January-2023]. [24] Dongbo Xi, Zhen Chen, Peng Yan, Yinger Zhang, Yongchun Zhu, Fuzhen Zhuang, and Yu Chen. 2021. Modeling the sequential dependence among audience multistep conversions with multi-task learning in targeted display advertising. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. 3745–3755. [25] Yu Zhang and Qiang Yang. 2021. A survey on multi-task learning. IEEE Transactions on Knowledge and Data Engineering (2021). [26] Zhe Zhao, Lichan Hong, Li Wei, Jilin Chen, Aniruddh Nath, Shawn Andrews, Aditee Kumthekar, Maheswaran Sathiamoorthy, Xinyang Yi, and Ed Chi. 2019. Recommending what video to watch next: a multitask ranking system. In Proceedings of the 13th ACM Conference on Recommender Systems. 43–51. Authors: (1) Yi Ren, Tencent, Beijing, China (yiren_bj@outlook.com); (2) Ying Du, Tencent, Beijing, China (yingdu@tencent.com); (3) Bin Wang, Tencent, Beijing, China (hillmwang@tencent.com); (4) Shenzheng Zhang, Tencent, Beijing, China (qjzcyzhang@tencent.com). Authors: Authors: (1) Yi Ren, Tencent, Beijing, China (yiren_bj@outlook.com); (2) Ying Du, Tencent, Beijing, China (yingdu@tencent.com); (3) Bin Wang, Tencent, Beijing, China (hillmwang@tencent.com); (4) Shenzheng Zhang, Tencent, Beijing, China (qjzcyzhang@tencent.com). 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

DML Revolutionizes Multi-Task Learning with Proven Effectiveness and Real-World Deployment

Authors:
(1) Yi Ren, Tencent, Beijing, China (yiren_bj@outlook.com);
(2) Ying Du, Tencent, Beijing, China (yingdu@tencent.com);
(3) Bin Wang, Tencent, Beijing, China (hillmwang@tencent.com);
(4) Shenzheng Zhang, Tencent, Beijing, China (qjzcyzhang@tencent.com). Table of Links Abstract and 1 Introduction Methodology
Experiments
Conclusion and References ABSTRACT Recommender systems usually leverage multi-task learning methods to simultaneously optimize several objectives because of the multi-faceted user behavior data. The typical way of conducting multi-task learning(MTL) is to establish appropriate parameter sharing across multiple tasks at lower layers while reserving a separate task tower for each task at upper layers. With such design, the lower layers intend to explore the structure of task relationships and mine valuable information to be used by the task towers for accurate prediction. Since the task towers exert direct impact on the prediction results, we argue that the architecture of standalone task towers is suboptimal for promoting positive knowledge sharing. First, for each task, attending to the input information of other task towers is beneficial. For instance, the information useful for predicting the "like" task is also valuable for the "buy" task. Furthermore, because different tasks are inter-related, the training labels of multiple tasks should obey a joint distribution. It is undesirable for the prediction results for these tasks to fall into the low density areas. Accordingly, we propose the framework of Deep Mutual Learning across task towers(DML), which is compatible with various backbone multitask networks. At the entry layer of the task towers, the shared component of Cross Task Feature Mining(CTFM) is introduced to transfer input information across the task towers while still ensuring one task’s loss will not impact the inputs of other task towers. Moreover, for each task, dedicated network component called Global Knowledge Distillation(GKD) are utilized to distill valuable knowledge from the global results of the upper layer task towers to enhance the prediction consistency. Extensive offline experiments and online A/B tests are conducted to evaluate and verify the proposed approach’s effectiveness. 1. INTRODUCTION Recently, we have seen the widespread application of recommender systems, which involve different types of user feedback signals, such as clicking, rating, commenting, etc. Moreover, no single feedback signal can accurately reflect user satisfaction. For example, over-concentrating on clicking may aggravate the click-bait issue. Therefore, it is highly desirable to be able to effectively learn and estimate multiple types of user behaviors at the same time. And Multi-Task Learning is a promising technique to address this challenge. Given several related learning tasks, the goal of multi-task learning is to enhance the overall performance of different tasks by leveraging knowledge transfer among tasks. With the Multi-Task Learning(MTL) paradigm[3, 25], multiple tasks are learned simultaneously in a single model. Compared with single-task solutions, MTL costs much fewer machine resources and improves learning efficiency since we just need to train and deploy a single model. Moreover, MTL usually can warrant enhanced recommendation performance through appropriate parameter sharing. Most existing methods of Multi-Task Learning(MTL) appropriately share parameters across multiple tasks at lower layers while keeping separate task towers at upper layers. These methods can be roughly classified into four categories. The first category comprises the methods of hard parameter sharing. Among them, embedding sharing is the most intuitive structure to share information. For instance, the ESSM model [13] shares embedding parameters between the tasks of CTR (Click-Through Rate) and CVR (Conversion Rate) for improving the prediction performance of the sparse CVR task. In addition to the embedding parameters, the Shared-Bottom structure[3] is introduced to share the parameters of lower-layer MLPs among tasks. But these methods are severely plagued by the task conflicts and negative transfer issue. Second, for the methods of soft parameter sharing, each task owns separate parameters, which are regularized during training to minimize the differences between the shared parameters. L2-constrained[6] is a typical algorithm belonging to this category. Third, for the methods of customized routing, they learn customized routing weights for each task to combine and fuse information from lower-layer networks to counteract the negative transfer issue. Cross-stitch network[15] and sluice network[16] learn separate linear weights for each task to selectively merge representations from different lower-level branches. SNR[11] modularizes the shared low-level layers into parallel sub-networks and uses a transformation matrix multiplied by a scalar coding variable to learn their connections to upper-level layers to alleviate the task conflict and negative transfer issue. MSSM[5] learns differing combinations of feature fields for each expert and designs finer-grained sharing patterns among tasks through a set of coding variables that selectively choose which cells to route for a given task. But the learned routing parameters of these methods are static for all the samples, which can hardly warrant optimal performance. Finally, the methods of dynamic gating learn optimized weights for each task based on the input sample to effectively combine the outputs of lower-level networks and achieve success in industrial applications. The MMoE [12] model adapts the Mixture-of-Experts (MoE)[9] structure to multi-task learning by sharing the expert sub-networks across all tasks, while also maintaining separate gating network optimized for each task. And Zhao et al. [26] extend the MMoE model [12] and apply it to learn multiple ranking objectives in Youtube video recommender systems. Moreover, PLE [17] achieves superior performance for news recommendation by assigning both shared expert sub-networks among tasks and dedicated expert sub-networks for each task. AITM [24] is the most similar method, which also augments the architecture of task towers. Nevertheless, as a concrete implementation, it is not validated to enhance the performance of various multi-task models. Moreover, it can only work for the tasks with sequential dependence relations. Admittedly, the aforementioned methods achieve impressive performance. However, as the task towers exert a direct effect on the prediction results, the standalone task towers tend not to be the most effective design for promoting positive knowledge transfer by exploiting the task relationships. First, for each task, the information selected by the relevant tasks is extremely valuable. Accordingly, we introduce the shared component of Cross Task Feature Mining(CTFM), which utilizes delicate attention mechanisms to extract relevant information from other tasks at the entry layer of the task tower. With the common attention mechanisms, the explicit task-specific information distilled by lower-level networks are mingled together and one task’s loss will undesirably affect the inputs of other task towers, which is the task awareness missing problem and can hinder the learning of lower-level networks. In contrast to the usual attention mechanisms, our design can ensure appropriate information separation. We argue that reserving explicit task-specific knowledge has a positive effect on performance, which is validated in the experimental section. Second, because the tasks for recommender systems are related, the training labels of multiple tasks should obey a joint distribution. The prediction results for these tasks should not densely fall into the low-density areas. Therefore, a dedicated network named Global Knowledge Distillation(GKD) is introduced for each task to distill valuable global knowledge from the results of the upper layer task towers. For each task, the distilled global information helps to ensure consistent predictions with other tasks. We summarize our main contributions below. • We propose the framework of Deep Mutual Learning across task towers(DML), which is compatible with various backbone multi-task models • The proposed novel sharing structure helps to enhance effective knowledge transfer across different tasks. • We conduct offline experiments and online A/B testing to evaluate and understand the effectiveness of our method. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Yi Ren, Tencent, Beijing, China (yiren_bj@outlook.com); (2) Ying Du, Tencent, Beijing, China (yingdu@tencent.com); (3) Bin Wang, Tencent, Beijing, China (hillmwang@tencent.com); (4) Shenzheng Zhang, Tencent, Beijing, China (qjzcyzhang@tencent.com). Authors: Authors: (1) Yi Ren, Tencent, Beijing, China (yiren_bj@outlook.com); (2) Ying Du, Tencent, Beijing, China (yingdu@tencent.com); (3) Bin Wang, Tencent, Beijing, China (hillmwang@tencent.com); (4) Shenzheng Zhang, Tencent, Beijing, China (qjzcyzhang@tencent.com). Table of Links Abstract and 1 Introduction Abstract and 1 Introduction Methodology Experiments Conclusion and References Methodology Methodology Experiments Experiments Conclusion and References Conclusion and References ABSTRACT Recommender systems usually leverage multi-task learning methods to simultaneously optimize several objectives because of the multi-faceted user behavior data. The typical way of conducting multi-task learning(MTL) is to establish appropriate parameter sharing across multiple tasks at lower layers while reserving a separate task tower for each task at upper layers. With such design, the lower layers intend to explore the structure of task relationships and mine valuable information to be used by the task towers for accurate prediction. Since the task towers exert direct impact on the prediction results, we argue that the architecture of standalone task towers is suboptimal for promoting positive knowledge sharing. First, for each task, attending to the input information of other task towers is beneficial. For instance, the information useful for predicting the "like" task is also valuable for the "buy" task. Furthermore, because different tasks are inter-related, the training labels of multiple tasks should obey a joint distribution. It is undesirable for the prediction results for these tasks to fall into the low density areas. Accordingly, we propose the framework of Deep Mutual Learning across task towers(DML), which is compatible with various backbone multitask networks. At the entry layer of the task towers, the shared component of Cross Task Feature Mining(CTFM) is introduced to transfer input information across the task towers while still ensuring one task’s loss will not impact the inputs of other task towers. Moreover, for each task, dedicated network component called Global Knowledge Distillation(GKD) are utilized to distill valuable knowledge from the global results of the upper layer task towers to enhance the prediction consistency. Extensive offline experiments and online A/B tests are conducted to evaluate and verify the proposed approach’s effectiveness. 1. INTRODUCTION Recently, we have seen the widespread application of recommender systems, which involve different types of user feedback signals, such as clicking, rating, commenting, etc. Moreover, no single feedback signal can accurately reflect user satisfaction. For example, over-concentrating on clicking may aggravate the click-bait issue. Therefore, it is highly desirable to be able to effectively learn and estimate multiple types of user behaviors at the same time. And Multi-Task Learning is a promising technique to address this challenge. Given several related learning tasks, the goal of multi-task learning is to enhance the overall performance of different tasks by leveraging knowledge transfer among tasks. With the Multi-Task Learning(MTL) paradigm[3, 25], multiple tasks are learned simultaneously in a single model. Compared with single-task solutions, MTL costs much fewer machine resources and improves learning efficiency since we just need to train and deploy a single model. Moreover, MTL usually can warrant enhanced recommendation performance through appropriate parameter sharing. Most existing methods of Multi-Task Learning(MTL) appropriately share parameters across multiple tasks at lower layers while keeping separate task towers at upper layers. These methods can be roughly classified into four categories. The first category comprises the methods of hard parameter sharing . Among them, embedding sharing is the most intuitive structure to share information. For instance, the ESSM model [13] shares embedding parameters between the tasks of CTR (Click-Through Rate) and CVR (Conversion Rate) for improving the prediction performance of the sparse CVR task. In addition to the embedding parameters, the Shared-Bottom structure[3] is introduced to share the parameters of lower-layer MLPs among tasks. But these methods are severely plagued by the task conflicts and negative transfer issue. Second, for the methods of soft parameter sharing , each task owns separate parameters, which are regularized during training to minimize the differences between the shared parameters. L2-constrained[6] is a typical algorithm belonging to this category. Third, for the methods of customized routing , they learn customized routing weights for each task to combine and fuse information from lower-layer networks to counteract the negative transfer issue. Cross-stitch network[15] and sluice network[16] learn separate linear weights for each task to selectively merge representations from different lower-level branches. SNR[11] modularizes the shared low-level layers into parallel sub-networks and uses a transformation matrix multiplied by a scalar coding variable to learn their connections to upper-level layers to alleviate the task conflict and negative transfer issue. MSSM[5] learns differing combinations of feature fields for each expert and designs finer-grained sharing patterns among tasks through a set of coding variables that selectively choose which cells to route for a given task. But the learned routing parameters of these methods are static for all the samples, which can hardly warrant optimal performance. Finally, the methods of dynamic gating learn optimized weights for each task based on the input sample to effectively combine the outputs of lower-level networks and achieve success in industrial applications. The MMoE [12] model adapts the Mixture-of-Experts (MoE)[9] structure to multi-task learning by sharing the expert sub-networks across all tasks, while also maintaining separate gating network optimized for each task. And Zhao et al. [26] extend the MMoE model [12] and apply it to learn multiple ranking objectives in Youtube video recommender systems. Moreover, PLE [17] achieves superior performance for news recommendation by assigning both shared expert sub-networks among tasks and dedicated expert sub-networks for each task. hard parameter sharing hard parameter sharing soft parameter sharing soft parameter sharing customized routing dynamic gating dynamic gating AITM [24] is the most similar method, which also augments the architecture of task towers. Nevertheless, as a concrete implementation, it is not validated to enhance the performance of various multi-task models. Moreover, it can only work for the tasks with sequential dependence relations. Admittedly, the aforementioned methods achieve impressive performance. However, as the task towers exert a direct effect on the prediction results, the standalone task towers tend not to be the most effective design for promoting positive knowledge transfer by exploiting the task relationships. First, for each task, the information selected by the relevant tasks is extremely valuable. Accordingly, we introduce the shared component of Cross Task Feature Mining(CTFM), which utilizes delicate attention mechanisms to extract relevant information from other tasks at the entry layer of the task tower. With the common attention mechanisms, the explicit task-specific information distilled by lower-level networks are mingled together and one task’s loss will undesirably affect the inputs of other task towers, which is the task awareness missing problem and can hinder the learning of lower-level networks. In contrast to the usual attention mechanisms, our design can ensure appropriate information separation. We argue that reserving explicit task-specific knowledge has a positive effect on performance, which is validated in the experimental section. Second, because the tasks for recommender systems are related, the training labels of multiple tasks should obey a joint distribution. The prediction results for these tasks should not densely fall into the low-density areas. Therefore, a dedicated network named Global Knowledge Distillation(GKD) is introduced for each task to distill valuable global knowledge from the results of the upper layer task towers. For each task, the distilled global information helps to ensure consistent predictions with other tasks. We summarize our main contributions below. • We propose the framework of D eep M utual L earning across task towers( DML ), which is compatible with various backbone multi-task models D M L DML • The proposed novel sharing structure helps to enhance effective knowledge transfer across different tasks. • We conduct offline experiments and online A/B testing to evaluate and understand the effectiveness of our method. 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

Stories, News & Mentions about LEAR

Lear’s Stories on HackerNoon

Lear's News & Mentions