Bandits for Recommender Systems: A Practical Guide
A field-oriented look at how multi-armed bandits balance exploration and exploitation in real-world recommender systems.
AI & ML
Frontier models, research results, agents, and developer tooling.
Python's Hidden Patterns: Lessons from Popular Libraries
A new essay highlights unconventional Python techniques used inside widely adopted libraries, offering practical insights for developers.
Defining a Data Team's Vision: Advice for New Leads
Eugene Yan's mailbag post offers practical guidance for new data team leads on setting a vision and roadmap.
Starting Data Science Projects: Lessons from Experience
Eugene Yan shares practical, hard-won advice on kicking off data science initiatives effectively.
Bootstrapping Labels From Scratch: Weak, Semi, Active, and Human Review
How to build training data from unlabeled examples using semi-supervised learning, active learning, weak supervision, and human review.
Autoencoders vs. Diffusers: Key Differences Explained
A concise look at how variational and denoising autoencoders compare with diffusion models.
Applying Eugene Yan's Writing Advice to AI/ML Communication
A writer's online journey offers transferable lessons for AI/ML practitioners who need to share their work clearly.
How Data Scientists Can Influence Without Formal Authority
Eugene Yan offers practical tactics for data scientists to shape decisions by showing data, asking questions, and building trust.
Eugene Yan: Why an Amazon Scientist Writes for Readers
An Amazon applied scientist explains the value of translating complex machine learning work into accessible writing.
Real-Time RecSys: From Industry Architecture to MVP
A practitioner's overview of why recommender systems need real-time updates, how industrial designs handle the problem, and what an MVP should look like.
Reinforcement Learning Tackles Long-Term Rewards in Recommendations and Search
A technical overview by Eugene Yan shows why RL matters for recommender systems and search when effects are long term and catalogs are fluid.
Start With Heuristics, Not Machine Learning
Eugene Yan argues that teams should first build simple rule-based baselines before turning to machine learning.
Eugene Yan’s 2021 Review: Goals Met and ApplyingML.com Built
A machine learning practitioner’s annual review reports progress on most goals and the launch of a new resource site.
Bootstrapping Relevance Labels for Semantic Search Evaluation
Eugene Yan tackles the chicken-and-egg problem of measuring search recall when no ground-truth relevance labels exist.
A Framework for Designing Search and Recommendation Systems
A concise mental model separates offline and online environments from retrieval and ranking when building discovery systems.
Text-to-Image Explained: Diffusion, Conditioning, and Guidance
A practical primer on how modern text-to-image models work, from diffusion basics to conditioning and guidance.
SyGra Framework Aims to Simplify Data Building for Language Models
ServiceNow AI has introduced SyGra, a one-stop framework for generating and managing training data for both large and small language models.
Hugging Face's Gaia2 and ARE Open New Paths for Agent Research
The release pairs a new benchmark with a research environment to let the community study AI agents more openly.