Machine Learning

An Introduction to Variational Autoencoders

The canonical tutorial on Variational Autoencoders by Kingma and Welling — the original VAE inventors. Covers the ELBO, reparameterization trick, and extensions to deeper generative models. Essential background for anyone working with latent variable models or modern diffusion/flow models.

Zvec: in-process vector database from Alibaba

Zvec is Alibaba's in-process vector database built on Proxima, their production search engine. No server to run — embed it directly in your app for billion-vector similarity search in milliseconds.

Neuronpedia Gemma Scope: interactive mechanistic interpretability for Gemma 2

Neuronpedia's Gemma Scope microscope lets you scan Gemma 2's internal features using Sparse Autoencoders — activating features, steering behavior, browsing SAE decompositions. Interactive mechanistic interpretability for a production model.

Linear Representations and Superposition in LLMs

A clear explainer on how LLMs encode thousands of features in relatively small embedding spaces via superposition — the Johnson-Lindenstrauss lemma applied to neural representations. Good primer on the theoretical foundations behind SAE-based interpretability.

Isomorphic Labs Drug Design Engine

Isomorphic Labs' Drug Design Engine (IsoDDE) goes beyond AlphaFold 3 to predict binding affinity and identify cryptic drug pockets from amino acid sequences alone. The step from structure prediction to actual drug optimization.

Understanding Neural Networks, Visually

An interactive web visualization that walks through how neural networks work — weights, activations, backpropagation — through direct manipulation. Good at building intuition rather than just showing diagrams.

Learning World Model Learning: inverse dynamics

Inverse dynamics learning as part of a 'learning world model learning from scratch' series — teaches robots to infer what actions produced observed state transitions. A ground-up educational approach to the robotics data problem.

How Attention Got So Efficient: GQA, MLA, DSA explained

A YouTube explainer on how attention mechanisms evolved from full multi-head attention to the efficient variants powering modern LLMs — covering GQA (Grouped Query Attention), MLA (Multi-head Latent Attention from DeepSeek), and DSA (Dynamic Sparse Attention). Useful for understanding why modern inference is faster and cheaper than it was two years ago.

Deep-ML: LeetCode-Style Machine Learning Practice

Deep-ML is a free, open-source LeetCode-style platform for machine learning — coding challenges across ML fundamentals, deep neural networks, computer vision, and NLP, with in-browser Python execution and immediate test feedback.

High-Dimensional Data Analysis: Wright & Ma Textbook

John Wright and Yi Ma's graduate textbook on high-dimensional data analysis — sparse models, low-rank models, and deep networks unified under a single mathematical framework. Adopted at Berkeley, Columbia, UIUC, Tsinghua. Published Cambridge University Press 2022.

How to Read Deep Learning Papers as a Software Engineer

A YouTube guide for software engineers on reading deep learning papers — framing the paper-reading skill as an engineering competency rather than a research one, with practical strategies for extracting implementation-relevant understanding from dense notation.

Why Deep Learning Works Unreasonably Well

Part 3 of a 'How Models Learn' series on why deep learning works unreasonably well — addressing the apparent paradox that overparameterized models generalize when classical statistics says they shouldn't. Covers implicit regularization, loss landscape geometry, and the lottery ticket hypothesis.

GPU Calculator for LLM Inference

An interactive calculator for estimating GPU requirements for LLM inference — input model size, quantization, batch size, and context length to get memory and throughput estimates. Practical first tool when planning self-hosted or cloud inference deployments.

A Theory of Usable Information Under Computational Constraints

The breakthrough behind modern AI image generators: diffusion models (3Blue1Brown)

3Blue1Brown's visual explanation of diffusion models — part 1 covering the core breakthrough behind modern AI image generators. Classic 3B1B treatment: intuition-first, mathematically grounded, visually clear.

McFly: neural network-powered shell history search

McFly replaces Ctrl+R shell history search with a small neural network that ranks suggestions by context — recent directory, exit codes, and command frequency all influence ranking. Works with bash, zsh, fish.

Kiln Fine-Tuning LLM Models Guide

Kiln's practical fine-tuning guide walks through generating synthetic training data (the 'Ladder' strategy: 10 manual → 30 LLM → 1000 synthetic) and running training jobs on OpenAI, Fireworks, and Unsloth. Emphasizes data quality over cost during generation.

MLGarden: Visual Neural Network Editor

MLGarden is an interactive visual editor for building and training neural networks without linear algebra — you construct computation graphs visually, watch backpropagation happen, and train on classification tasks. Built in C++ for performance, runs in the browser for accessibility.

Flash Attention: Derived and Coded from First Principles with Triton

A from-scratch implementation of Flash Attention using Triton (Python GPU programming), deriving the algorithm from first principles before coding it. Valuable for anyone wanting to understand how Flash Attention achieves its memory efficiency — learning by building rather than just using the library.

fast.ai

fast.ai is Jeremy Howard and Rachel Thomas's free deep learning course and library, famous for teaching neural networks top-down — use them effectively first, understand the math later. Widely credited with democratizing deep learning education.

Deep Learning Systems (CMU 10-414/714)

10-414/714: Deep Learning Systems at CMU — a publicly available course on building the components of a deep learning framework from scratch, including automatic differentiation, optimization, and hardware acceleration. One of the best resources for understanding how frameworks like PyTorch actually work.

Nightshade: Protecting Copyright Through Adversarial Poisoning

Nightshade is a tool that lets artists poison AI training data by adding imperceptible perturbations to images — images look normal to humans but cause AI models trained on them to produce corrupted outputs. An offensive countermeasure for artists against unauthorized scraping.

Inspectus: LLM Analytics and Visualization

Inspectus is labml.ai's tool for visualizing LLM internals — attention maps, token distributions, and model analytics rendered as interactive visualizations in Jupyter notebooks. Makes transformer attention patterns inspectable without custom code.

Fine-tuning language models: a practical overview

A clear introductory overview of LLM fine-tuning from the GenAI Guidebook — covers why and when to fine-tune, the mechanics of weight updates, and the major techniques including LoRA and RLHF.

Neural Networks from Scratch in Python

Neural Networks from Scratch (NNFS) is a book by Harrison Kinsley and Daniel Kukiela that builds neural networks in pure Python with no frameworks — the go-to resource for understanding what's actually happening inside backpropagation and gradient descent.

Heimdall: Local ML App for macOS

Heimdall is a macOS app that brings machine learning tools to a local native interface — positioning itself as an accessible way to run ML models without needing a Python environment or cloud API. Part of the 2023 wave of ML-powered local productivity tools.

Heinsen Routing: Vector Routing for Sequences and Capsule Networks

Heinsen routing is a novel algorithm for routing vectors in sequences and capsule networks — an alternative to Hinton's dynamic routing that generalizes across domains. A niche but foundational contribution to the capsule network research line.

From Deep to Long Learning?

Stanford Hazy Research argues the next frontier is moving from deep networks to networks that can process very long sequences — motivating state space models like Mamba as a shift away from transformer attention's O(n²) complexity. A prescient 2023 post about where sequence modeling was headed.

Google Research Deep Learning Tuning Playbook

Google Research's Deep Learning Tuning Playbook — a systematic guide to maximizing model performance through hyperparameter optimization. Written by Braxton Osting and team, it covers the science and art of tuning learning rates, batch sizes, regularization, and the full training configuration.

Replicate Cog: Containers for Machine Learning

Cog is Replicate's open-source tool for packaging machine learning models into standardized Docker containers with automatic HTTP APIs. It's the packaging layer that powers Replicate's model hosting platform — define inputs/outputs in Python, get a container that serves the model.

NLP Demystified

NLP Demystified is a free video course covering NLP fundamentals from text preprocessing through transformers and modern language models — aimed at practitioners who want a solid conceptual foundation rather than just API usage. One of the cleaner free NLP curricula available.

Byte Size ArXiv

Byte Size ArXiv summarizes cutting-edge ML research papers in accessible, digestible form. A useful filter layer over the torrent of ArXiv submissions for people who want to track research without reading full papers.

Aporia — ML Observability

Aporia is a cloud-native ML observability platform for monitoring deployed models — drift detection, performance monitoring, and explainability tooling. Addresses the gap between training-time metrics and what models actually do in production.

Q2 2022 AI/ML Industry Report (Gradient Flow Preview)

Gradient Flow's Q2 2022 preview report surveying the AI/ML industry landscape — adoption patterns, infrastructure tooling, and where enterprise ML investment was flowing mid-2022. A snapshot of the field right before the generative AI wave broke, useful as a baseline for how quickly the priorities shifted.

Synced — AI Research and Industry News

Synced is an AI and tech news publication covering ML research, industry developments, and AI applications — originally focused on the Chinese AI ecosystem but expanded globally. One of the more substantive English-language sources tracking AI research news.

Monolith: Real Time Recommendation System With Collisionless Embedding Table

Monolith is ByteDance's production recommendation system purpose-built for online training with dynamic sparse features, featuring a collisionless embedding table with expirable embeddings and frequency filtering. It demonstrates that general-purpose ML frameworks like TensorFlow are ill-suited for industrial recommendation with non-stationary data distributions.

ROOTS Search Tool — BigScience

A Hugging Face Space for searching ROOTS — the massive multilingual dataset used to train BLOOM, the BigScience open LLM. Lets researchers trace which training documents a model might have learned from.

Bleeding Edge — AI News Feed

Bleeding Edge is a curated feed of noteworthy AI developments — research papers, product launches, and industry news filtered for signal over noise. A useful daily tracker for keeping up with the pace of AI progress.

AI Winter Is Well On Its Way

Filip Piekniewski's 2018 blog post predicting an AI winter due to deep learning's fundamental limitations — a contrarian view written before the GPT era proved the scaling hypothesis. Useful as a document of expert skepticism that turned out to be largely wrong.

Vectors Are Over? Hashes as the Future of AI Search

Algolia's provocative post arguing that hash-based retrieval outperforms vector search for many real-world search use cases — a counterargument to the vector database hype of 2022. Grounds the comparison in production search quality metrics.

The Illustrated Stable Diffusion

Jay Alammar's visual explainer of how Stable Diffusion works under the hood — covering latent diffusion, the CLIP text encoder, and the U-Net denoiser. Alammar's illustrated series is one of the best entry points for building intuition about complex ML architectures.

MIT 6.S898: Deep Learning (Fall 2022)

MIT's 6.S898 Deep Learning course taught by Phillip Isola, covering the full modern deep learning stack from fundamentals through generative models and transformers. One of the cleaner academic deep learning curricula, with public materials.

Language Models as Models of the Visual World

Research showing language models can use linear projections of image representations as soft prompts for vision-language tasks — without tuning the LM or image encoder. An early signal for the efficiency of frozen model feature reuse in multimodal architectures.

How Diffusion Models Work: The Math from Scratch

AI Summer's mathematical walkthrough of how diffusion models work from scratch — covering the forward noising process, reverse denoising, DDPM training objective, and score matching. The most math-forward accessible introduction to the field.

SE-ML: Software Engineering Practices for Machine Learning

SE-ML is a catalog of software engineering practices adapted for machine learning systems — covering data management, model training, deployment, and monitoring. A structured reference for treating ML development with the same rigor as traditional software engineering.

Didact AI: The Anatomy of an ML-Powered Stock Picking Engine

A technical teardown of Didact AI's ML-powered stock picking engine — covering feature engineering, model architecture, training pipeline, and how uncertainty quantification informs position sizing. Rare public documentation of a production ML trading system.

MLOps: Machine Learning Operations

A comprehensive guide to MLOps — the practices, tools, and culture for deploying and maintaining machine learning models in production. Covers the full lifecycle from experiment tracking through model serving, monitoring, and retraining pipelines.

The Vector Database Index

Gradient Flow's landscape map of vector databases — published in 2022 when the category was forming, covering Pinecone, Weaviate, Qdrant, Chroma, Milvus, and others. A useful historical snapshot of the vector DB market at the moment it became strategically important.

ML YouTube Courses

A curated catalog of machine learning courses available on YouTube, maintained by DAIR.AI. Covers ML fundamentals, deep learning, NLP, computer vision, and specialized topics — a single index for free university-grade ML education.

Banana: Serverless GPU Inference Hosting

Banana is a serverless GPU inference hosting platform for machine learning models — deploy your model in three steps, pay only for inference time. Targets the gap between spinning up a GPU VM yourself and the complexity of cloud ML serving infrastructure.

Attention? Attention! (Lilian Weng)

Lilian Weng's canonical blog post on attention mechanisms — written in 2018, it covers sequence-to-sequence attention, self-attention, and multi-head attention. Still the clearest single-page reference for understanding how attention works before reading the Transformer paper.

Memory-Based Model Editing at Scale

ICML 2022 paper on memory-based model editing at scale — a method for locally updating a neural network's factual knowledge without full retraining. Uses a cache of explicit edits that override the base model's predictions for targeted inputs.

Editable Neural Networks

ICLR paper proposing 'editable neural networks' — a method for making targeted modifications to a model's behavior on specific inputs without disrupting performance elsewhere. An early formalization of the model editing problem.

Alpa: Automated Distributed Training for Large Models

Alpa is a system for automatically parallelizing large neural network training across distributed hardware — finding optimal parallelism strategies without manual configuration. From a Berkeley/CMU research collaboration, it targets the challenge of scaling models beyond single-GPU memory.

How AI Transformers Mimic Parts of the Brain (Quanta Magazine)

Quanta Magazine's September 2022 piece on emerging research showing that transformer attention patterns converge with how the brain processes language and vision — a surprising empirical finding with implications for both AI interpretability and computational neuroscience.

AI Content Generation, Part 1: Machine Learning Basics (Jon Stokes)

Jon Stokes' accessible introduction to machine learning as the foundation of AI content generation — Part 1 of a series aimed at readers with no ML background. Stokes covers the statistical learning framing without requiring math, making it one of the better on-ramps for non-technical audiences.

Made With ML: MLOps Curriculum

Made With ML is a free, project-based curriculum for learning ML engineering and MLOps — covering not just model training but the full production pipeline from data to deployment. One of the most practical and comprehensive self-study resources for applied ML.

Attention and Augmented Recurrent Neural Networks (Distill)

Distill's 2016 interactive visual explainer covering four augmented RNN variants — attention, memory networks, neural Turing machines, and neural programmers. A foundational reference for understanding how external memory and attention were grafted onto recurrent networks before the transformer era.

A Short Chronology of Deep Learning for Tabular Data

Sebastian Raschka's chronological survey of deep learning approaches for tabular data — the domain where gradient boosted trees still dominate. A clear-eyed accounting of why deep learning hasn't won on structured data despite winning everywhere else.

"Hey Machine, What's My Invoice Total?" (Impira Blog)

Impira's explainer on document understanding ML — how to extract structured data like invoice totals from unstructured documents using layout-aware models. Covers why document AI is harder than it looks and how layout-aware transformers like LayoutLM changed the problem.

The Cost of Training NLP Models: A Concise Overview

AI21 Labs' 2020 overview of the financial and compute cost of training large NLP models, tracing the exponential growth in training expenditure from BERT through GPT-3. An early quantitative lens on the economics of scale in language model development.

A Library for Representing Python Programs as Graphs for Machine Learning

Google Research library paper introducing python_graphs — an open-source tool for constructing graph representations of Python programs via static analysis, producing control flow graphs, data flow graphs, and program dependence graphs. Standardizes the infrastructure for ML-on-code research so researchers don't each rebuild the graph extraction layer.

Artificial Images: Derrick Schultz's Generative AI Practice

Artificial Images is Derrick Schultz's practice combining generative AI art with ML education — workshops on diffusion models, ComfyUI, and computational filmmaking for artists and designers. A sustained bridge between the ML research community and working artists.

Explaining Transformer Model Predictions

A practical comparison of SHAP, Transformers Interpret, and Ferret for explaining Hugging Face transformer predictions. Key takeaway: different methods give different results for the same prediction — all require careful interpretation.

Getting into Machine Learning in 2022 (HN Discussion)

Hacker News discussion on the best paths into ML/DL in 2022 — debating resources (fast.ai, Andrew Ng, Bishop), career tracks (researcher vs. engineer vs. data scientist), and prerequisites. A snapshot of community wisdom on the ML learning path before ChatGPT shifted the landscape.

Stochastic Processes and Simulations: An ML Perspective

Stochastic Processes and Simulations: A Machine Learning Perspective by Vincent Granville — covers point processes, Poisson-binomial processes, GPU clustering, and extreme value theory with Python code. Bridges probability theory and ML's simulation-heavy approach to uncertainty.

KDD 2022 Keynote

Keynote slides from the ACM SIGKDD 2022 conference on knowledge discovery and data mining. Content covers advances in large-scale ML, graph learning, or responsible AI — exact topic depends on which keynote this slide deck is from.

Awesome Colab Notebooks

Curated collection of Google Colab notebooks for ML experiments — covering Stable Diffusion, GANs, NLP models, and more. A snapshot of what was freely runnable in browser-based GPU compute in 2022, before Hugging Face Spaces simplified deployment further.

Machine Learning — Tom Mitchell (1997)

Tom Mitchell's 1997 Machine Learning textbook from CMU — the foundational academic reference that defined the field's core concepts before deep learning. Still assigned in graduate courses for its clear treatment of decision trees, Bayesian methods, and the canonical formal definition of machine learning.

Abacus.AI

Abacus.AI is an enterprise ML platform providing AI capabilities to companies without large AI teams — real-time feature stores, automated training, and deployment. Bookmarked via the jobs page, suggesting career interest in AI platform infrastructure work.

Sematic: Python-First ML Pipeline Orchestration

Sematic is an open-source ML pipeline orchestration framework using Python decorators — no YAML, seamless local-to-cloud execution, built-in artifact tracking dashboard. Competes with MLflow and Kubeflow but stays in pure Python.

Piranha: A GPU Platform for Secure Computation

Piranha (USENIX Security 2022) is a GPU platform for secure multi-party computation that exploits GPU parallelism to accelerate MPC protocols by 10-15× over CPU implementations. It shows that GPU hardware can close the practical performance gap for privacy-preserving machine learning at scale.

Weights & Biases — Fully Connected ML Blog

The Weights & Biases Fully Connected blog covers ML research, tutorials, and MLOps practices. Saved as a resource hub — W&B's writing quality is high, and their reports feature interactive experiment tracking visualizations unavailable elsewhere.

Ivy: The Unified Machine Learning Framework

Ivy is a unified machine learning framework that transpiles code between NumPy, PyTorch, TensorFlow, and JAX — write once, run anywhere across ML backends. An ambitious attempt to solve the framework fragmentation problem that plagues ML research and production.

Distill: Clear Explanations of Machine Learning

Distill.pub is a peer-reviewed ML journal focused on unusually clear explanations of machine learning concepts, using interactive visualizations alongside rigorous technical content. Went on hiatus in 2021 after 5 years, but the archive remains the gold standard for interactive ML pedagogy.

CatalyzeX: AI/ML Papers with Code Everywhere

CatalyzeX is a Chrome extension that adds 'Find Code' buttons directly to ML papers on arXiv, Google Scholar, Semantic Scholar, and NeurIPS — linking you from a paper to its GitHub implementation without leaving the page. Essential for practitioners who read papers and want to run the code.

Random Forest — MLU-Explain

MLU-Explain's visual, interactive introduction to the Random Forest algorithm — animated trees, bootstrap sampling, and feature importance built into the browser. Part of Amazon's ML University series of interactive learning tools.

AI Infrastructure Ecosystem 2022

A mid-2022 industry report mapping the AI/ML infrastructure ecosystem across three build strategies (build-your-own, end-to-end platforms, best-of-breed composable stacks) and the full toolchain from orchestration through model serving. A useful time-capsule of the MLOps landscape just before the LLM era reshaped it.

Scoring Aave Accounts for Creditworthiness

Cred Protocol's paper (arXiv:2207.07008) proposing a credit scoring system for Aave v2 accounts using a tree-based classifier trained to predict 'position delinquency' — whether a borrow position's health factor will drop below 1 within 90 days. It's a direct attempt to bring FICO-style credit infrastructure to DeFi, enabling undercollateralized lending.

Colossal-AI: Distributed Deep Learning System

Colossal-AI is an open-source distributed deep learning framework that makes training very large models more accessible — cutting GPU memory requirements by up to 10x versus standard PyTorch. One of several systems research projects responding to the GPU memory wall problem in 2022.

Linear Algebra — Math for Machine Learning

A YouTube course on linear algebra specifically framed for machine learning — covering vectors, matrices, eigenvalues, and the operations that underlie neural networks. Useful complement to theoretical ML reading when the math intuition is missing.

Jay Alammar — Visualizing Machine Learning

Jay Alammar's blog is the go-to resource for visually understanding modern ML architectures — transformers, BERT, GPT, and more through hand-crafted diagrams. His illustrated explainers have become canonical references for practitioners who want intuition before equations.

Colah's Blog — Chris Olah on Neural Networks

Chris Olah's personal blog is foundational reading for anyone trying to understand deep learning from first principles — his LSTM explainer, neural network topology post, and attention posts have shaped how a generation of researchers think about these models. Olah went on to co-found Anthropic's interpretability team, and the blog reflects his interest in mechanistic understanding.

Top arXiv Machine Learning Papers in 2021

A community-ranked list of the top ML papers on arXiv in 2021 according to metacurate.io. A useful snapshot of what the ML research community considered most impactful at a pivotal year — just before diffusion models and ChatGPT reshaped the field.

Text Embeddings Visually Explained

Cohere's visual primer on text embeddings explains how words and sentences become vectors in high-dimensional space, and what operations on those vectors mean semantically. A good conceptual foundation before diving into practical embedding-based applications like semantic search or classification.

Exploring Production-Ready Recommender Systems with NVIDIA Merlin

NVIDIA Merlin is a framework for building GPU-accelerated production recommender systems — covering feature engineering (NVTabular), training (HugeCTR, Merlin Models), and serving (Triton). This post explores the end-to-end pipeline for large-scale recommendation.

ART: Questions Are All You Need to Train a Dense Passage Retriever

ART (Autoencoding-based Retriever Training) trains dense retrievers without labeled question-passage pairs — only questions and an unpaired document collection. This removes the main bottleneck for deploying dense retrieval in new domains where annotation is expensive.

A Survey on Graph Representation Learning Methods

Khoshraftar and An (York University) provide a comprehensive survey of graph representation learning covering node embeddings, GNNs, and knowledge graph methods, with attention to both spectral and spatial approaches. It's the right starting point for anyone orienting to the GRL landscape — methodical coverage from DeepWalk through GAT.

Near-Linear Scaling of Gigantic Model Training on AWS

Amazon Science's post on achieving near-linear scaling when training gigantic language models across thousands of GPUs on AWS infrastructure. Covers parallelism strategies (tensor, pipeline, data) that make training 100B+ parameter models economically feasible.

Pen & Paper Exercises in Machine Learning

Michael Gutmann's (Edinburgh) collection of pen-and-paper exercises covering the mathematical foundations of machine learning — linear algebra, optimization, graphical models, density estimation, and classification. Designed to build mathematical fluency that coding-first courses skip, using derivation rather than implementation as the primary learning mode.

Privacy-Preserving Machine Learning with Fully Homomorphic Encryption for Deep Neural Networks

This paper demonstrates running deep neural network inference entirely on encrypted data using fully homomorphic encryption, so the server never sees plaintext inputs or outputs. It makes encrypted ML inference practical by combining FHE with approximation-friendly neural network architectures.

How fast can we perform a forward pass?

A deep technical investigation into the theoretical and practical speed limits of a single transformer forward pass, accounting for compute, memory bandwidth, and hardware constraints. The answer matters for inference costs, latency SLAs, and understanding where efficiency gains are actually possible.

Proof of Useful Work: Blockchain Web Services

A 2022 technical whitepaper proposing Proof of Useful Work (PoUW) as an alternative to Bitcoin's proof-of-work, directing miners' computation toward paying clients' ML training jobs and scientific simulations instead of purposeless hash puzzles. A direct response to the energy waste criticism of PoW mining.

Gradient Explanations for HuggingFace BERT Classification

A tutorial by Victor Dibia on generating gradient-based explanations for HuggingFace BERT text classification models in TensorFlow 2.0 — visualizing which tokens most influenced the model's prediction. Explainability for transformer classifiers was a practical gap in 2022 since attention maps alone are insufficient.

Slither-simil: ML-Assisted Smart Contract Audits

Trail of Bits introduces Slither-simil, a tool that uses ML embeddings to find smart contracts similar to a known-vulnerable one — dramatically accelerating audits by surfacing candidate contracts before manual review. An early example of applying ML to the smart contract security problem.

MLU-Explain: Visual ML Education

MLU-Explain is Amazon's collection of interactive visual explainers for core machine learning concepts — decision trees, random forests, bias-variance tradeoff, cross-validation, and more. Built as interactive articles in the style of Distill.pub, targeting practitioners who want intuition over math.

Modeling Vocabulary for Big Code Machine Learning

An empirical study of vocabulary modeling decisions for machine learning systems on source code, evaluated across 14,436 projects. It matters because the choices made when tokenizing and preprocessing code vocabularies have an outsized impact on neural language model accuracy, yet were poorly documented before this work.

BERTopic: The Future of Topic Modeling

A Pinecone explainer on BERTopic — a topic modeling library that uses transformer embeddings instead of bag-of-words statistics, producing semantically coherent topics. BERTopic largely made LDA obsolete for practitioners who have access to modern embeddings.

Quant-MELO-Portfolio: Bayesian Portfolio Optimization

Quant-MELO-Portfolio is a Python project applying Bayesian architecture to stock portfolio optimization — finding optimal weights via Global Minimum Variance and Tangency portfolios. A concrete implementation of mean-variance optimization with Bayesian uncertainty quantification.

OPT Training Chronicles

Meta's OPT-175B training logbook — a day-by-day record of hardware failures, instabilities, loss spikes, and interventions during a ~2 month distributed training run. Rare public documentation of what large language model training actually looks like at scale.

Proof of Learning (PoLe): Blockchain Consensus via Neural Network Training

Proof of Learning (PoLe) repurposes the computation wasted on Proof-of-Work consensus by directing it toward training neural networks, making blockchain nodes useful ML workers. The cheating-prevention mechanism (Secure Mapping Layer) is the interesting technical contribution — a linear layer baked into the network architecture that makes it costly to fake training progress.

Self-Organising Textures

Distill's 2021 interactive article on Neural Cellular Automata for texture synthesis — a small local rule, learned via gradient descent, produces complex global textures through iteration. One of the clearest demos that local rules can encode rich global structure.

Differentiable Self-Organizing Systems Workshop

Workshop site for the Differentiable Self-Organizing Systems research thread — covering gradient-based approaches to systems that grow, repair, and self-organize. The academic home of the Neural Cellular Automata research program from Mordvintsev et al.

All Roads Lead to Rome: The ML Job Market in 2022

Eric Jang's 2022 essay on the ML job market — titled 'All Roads Lead to Rome,' arguing that different entry points (research labs, industry, startups) all converge on the same destination if you're technically strong. Candid advice from a DeepMind robotics researcher.

MLOps Toys: Curated List of Machine Learning Projects

MLOps Toys is a curated directory of ML projects and tools — organized by category (data, model training, serving, monitoring, etc.). A snapshot of the MLOps ecosystem circa 2022, useful as a reference map of what tools existed before the category consolidated.

cleanlab 2.0: Automatically Find Errors in ML Datasets

cleanlab 2.0 is an open-source Python framework for automatically finding and fixing errors in ML datasets — mislabeled examples, out-of-distribution samples, near-duplicates. Built on the 'confident learning' statistical framework for label noise estimation.

ML Paper Volume: Moore's Law for arXiv

A tweet noting that ML paper volume on arXiv is growing exponentially — keeping current requires reading 100+ papers per day. A Moore's Law analogy applied to research output rather than transistor density, capturing the 2022 ML research explosion.

CrypTen: Secure Multi-Party Computation Meets Machine Learning

CrypTen is a PyTorch-compatible framework from Facebook AI Research that wraps secure multi-party computation protocols behind a familiar tensor API, making private inference and training accessible to ML practitioners without cryptography expertise. The bet is that adoption bottlenecks for privacy-preserving ML are mostly about developer experience, not theoretical limits.

DISCO: Decentralized Privacy-Preserving ML Training

DISCO (from EPFL) is a decentralized, privacy-preserving ML training framework using peer-to-peer networking — no central server, nodes train locally and share model updates directly. Extends federated learning to remove the trusted aggregator requirement.

An Open Source AutoML Benchmark

This paper presents an open-source, extensible benchmark for comparing AutoML systems across 39 classification datasets, finding that no single system consistently dominates a tuned random forest baseline. It establishes best practices for fair AutoML evaluation and provides a living framework that accepts community contributions.

Transformer Inference Arithmetic

Carol Chen (kipply) works through the arithmetic of transformer inference — compute vs. memory bandwidth, KV cache sizing, and how batch size shapes throughput/latency tradeoffs. Essential reference for anyone reasoning about LLM serving costs.

Ranking YC Companies with a Neural Net

Eric Jang trains a neural net to rank Y Combinator companies by prestige/success using only their names and descriptions — an experiment in whether language models encode meaningful startup quality signals. The results are surprisingly coherent.

Graph-Powered Machine Learning

Alessandro Negro's Manning book covers the intersection of graph theory and machine learning, from knowledge graphs and GNNs to fraud detection and recommendations using Neo4j. It's a practical end-to-end treatment that bridges graph databases and ML for practitioners.

Deep Learning with PyTorch

Manning's 2020 practical guide to deep learning with PyTorch by Stevens, Antiga, and Viehmann, covering tensors through CNNs, RNNs, generative models, and production deployment. The go-to book for practitioners who want to understand PyTorch from first principles rather than copy-paste patterns.

Machine Learning in Finance: From Theory to Practice

Springer 2020 textbook by Dixon, Halperin, and Bilokon bridging ML theory and quantitative finance practice — covering supervised learning, NLP for financial texts, RL for trading, and deep learning for derivatives pricing. The most rigorous academic treatment of ML applied to finance.

Transformers for Software Engineers

Nelson Elhage's explainer on transformers pitched at software engineers — treats the architecture as a data structure rather than mysterious ML magic. Grounding for anyone who codes but hasn't internalized what attention actually does.

Haystack: Question Answering at Scale

Haystack by deepset is an open-source NLP framework for building question-answering and search systems at scale — Retriever-Reader pipelines over large document corpora. An early entrant in what became the RAG ecosystem.

Uberduck: Voice Cloning and Text-to-Speech

Uberduck is a text-to-speech and voice cloning platform that gained early viral traction through celebrity voice imitations and rap generation. Early consumer-facing example of generative audio before ElevenLabs dominated the space.

Ntropy: Transaction Enrichment API

Ntropy is a transaction enrichment API that cleans, labels, and categorizes raw banking transaction data using ML — turning the noisy strings in bank feeds into structured, queryable merchant and category data.

An Intuitive Guide to Linear Algebra

BetterExplained's intuitive guide to linear algebra — builds geometric intuition for vectors, matrices, and transformations rather than drilling algebraic procedures. The right starting point before the formal treatment.

All That Glitters Is Not Gold: Comparing Backtest and Out-of-Sample Performance on a Large Cohort of Trading Algorithms

GPT-J: Self-Hosted Open-Source LLM

A guide to running GPT-J, EleutherAI's open-source GPT-3 analog, in Docker — a 2022 snapshot of what self-hosting a large language model looked like before the Llama era democratized this further. Early example of the open-weight model movement.

Word2Vec Explained

A clear walkthrough of Word2Vec's intuition and mechanics — skip-gram vs. CBOW, negative sampling, and why the king-queen analogy works. Good primer before reading papers on later embedding methods.

Data50: The World's Top Data Startups

Andreessen Horowitz's 2022 list of the 50 most important data startups — a snapshot of the data infrastructure landscape at peak cloud/ML investment. Useful map of which companies a16z thought were winning in data tooling, observability, and analytics.

Metarank: ML-Powered Ranking Engine

Metarank is an open-source ML-powered ranking engine — takes user feedback signals (clicks, purchases, bookmarks) and trains a Learn-to-Rank model to personalize product listings and search results. Low-code alternative to building a custom LTR pipeline.

Introduction to Machine Learning Interviews Book

Chip Huyen's free ML interviews book — covers both the process of landing ML roles and the technical depth (math, coding, ML concepts) that hiring loops test. Dual-purpose: career strategy and technical review.

R2D3: Visual Introduction to Machine Learning

R2D3 is a visual introduction to machine learning and statistics — scrollytelling essays that use animated D3.js visualizations to explain concepts like decision trees, bias-variance tradeoff, and model comparison. One of the best examples of the genre.

Will Transformers Take Over Artificial Intelligence?

Quanta Magazine's 2022 look at whether transformer architectures will dominate all of AI — following their success in NLP and early incursion into image classification. A useful time-capsule of the moment when the transformer paradigm started feeling inevitable.

The Bitter Lesson

Rich Sutton's 2019 essay arguing that the dominant lesson from 70 years of AI research is that general methods leveraging computation always win over human-engineered knowledge — a humbling argument against clever domain-specific tricks. One of the most cited and debated essays in ML.

Anatomy of an AI System

Kate Crawford and Vladan Joler's 2018 visual essay dissecting an Amazon Echo as a case study in the full global supply chain, labor, and infrastructure behind AI systems — materials, data labor, algorithmic labor, user labor. One of the most cited critical AI artifacts.

MLOps Maturity Models: Google and Microsoft Frameworks

ZenML's overview of MLOps maturity models from Google and Microsoft — frameworks for thinking about how ML organizations can systematically improve how they develop and deploy models. Useful if you're trying to level up a team's ML practices from ad-hoc to automated.

MLOps Notebooks Syllabus

A Jupyter notebook curriculum for MLOps — covering model deployment, monitoring, CI/CD for ML, feature stores, and data versioning. Practical operational coverage for the gap between training a model and running it reliably in production.

Swarm Training

Shawn Presser's Swarm Training explores distributed ML training across many commodity machines with low-bandwidth interconnects — democratizing large model training beyond clusters with expensive NVLink. Part of the broader open-source effort to train large models outside of big lab infrastructure.

Probabilistic Machine Learning (Kevin Murphy)

Kevin Murphy's Probabilistic Machine Learning book series — a comprehensive treatment of ML through a probabilistic/Bayesian lens, freely available online. The 2012 original (MLPP) and its 2022 follow-ups are standard graduate-level references.

Neural Instrument Cloning from Very Few Samples

Research on cloning musical instrument sounds using neural audio synthesis from very few samples — few-shot timbre transfer. Relevant to AI music generation tools and the question of how much training data audio models need.

GAN Lab: Interactive GANs in the Browser

GAN Lab is an interactive browser-based tool for playing with Generative Adversarial Networks — visualizing the generator and discriminator training dynamics in real time. One of the best tools for building intuition about how GANs work and fail.

Dive into Deep Learning Compiler

A free textbook companion to Apache TVM covering deep learning compiler design — how neural networks get transformed into optimized code for CPUs, GPUs, and specialized accelerators. Essential for anyone working on ML infrastructure or hardware-software co-design.

Block-NeRF: City-Scale Neural Neighborhoods

Block-NeRF from Waymo and UC Berkeley extends Neural Radiance Fields to city-scale scenes by dividing them into individually trained blocks that stitch together. A step toward photorealistic neural reconstruction of entire neighborhoods from street-level imagery.

Document Processing in Multiple Languages

Nanonets' overview of multilingual document processing — extracting structured data from invoices, forms, and documents in non-English languages. Covers the ML challenges of multilingual OCR and key-value extraction at scale.

Task-Specific Knowledge Distillation for BERT

A tutorial on task-specific knowledge distillation for BERT using Hugging Face Transformers and Amazon SageMaker — compressing a 109M-parameter BERT-base into a 4M-parameter student with 90%+ performance retention. Demonstrates that you don't need a massive model in production if you can distill task knowledge from one.

Interpretable Machine Learning

Christoph Molnar's free online book covering the theory and practice of interpretable machine learning — from inherently interpretable models (decision trees, linear regression) to post-hoc methods (SHAP, LIME, counterfactuals). The standard reference for understanding and explaining ML model behavior.

ML and NLP Research Highlights of 2021

Sebastian Ruder's annual ML/NLP research highlights for 2021 — covering foundation models, the prompting revolution, AlphaFold 2, diffusion models, and the growing focus on efficiency and responsible AI. The most useful single-document summary of where the field moved that year.

LambdaMART In Depth

An in-depth technical explainer on LambdaMART — the gradient boosted tree algorithm for learning-to-rank that underlies most production search and recommendation systems. Explains lambda values, pairwise swapping, and DCG optimization in a way that builds genuine intuition.

Interviews.AI — Data Science Interview Preparation

A GitHub-hosted book of data science and ML interview preparation material aimed at quantitative candidates facing competitive ML engineer and data scientist interviews. Covers statistics, probability, ML theory, and coding challenges.

Real-Time Machine Learning: Challenges and Solutions

Chip Huyen's definitive survey of real-time ML challenges — the engineering gap between batch ML pipelines and systems that must make predictions on live data with low latency. Covers online prediction, continual learning, and the infrastructure needed to bridge training and serving.

Lessons on ML Platforms from Netflix, DoorDash, Spotify, and More

Ernest Chan's Towards Data Science survey of ML platform design at Netflix, DoorDash, Spotify, and other tech companies — extracting common patterns and lessons from how production ML infrastructure evolved at scale. Practical systems thinking for ML platform builders.

Hidden Technical Debt in Machine Learning Systems

Sculley, Holt, Golovin et al. at Google extend the software engineering concept of technical debt to ML systems, identifying ML-specific forms that accumulate invisibly at the system level: boundary erosion, entanglement, hidden feedback loops, undeclared consumers, and data dependencies. The canonical paper explaining why ML systems are uniquely expensive to maintain.

Feature Stores — A Hierarchy of Needs

Eugene Yan's survey of feature stores organized as a hierarchy of needs — from basic feature reuse to online serving to real-time streaming. Practical taxonomy for understanding when you need a feature store and what tier of sophistication your use case requires.

How Sber Built ruDALL-E — Interview with Sergei Markov

Serokell's interview with Sergei Markov of SberDevices about building ruDALL-E — a 12B parameter Russian-language text-to-image model. Covers the engineering and research challenges of training massive multimodal models, plus the open-source culture argument in ML.

TheSequence 2022 ML Reading List

TheSequence's 2022 ML reading list — curated books and papers for going deep on machine learning and AI. A practitioner-oriented guide to the foundational and cutting-edge literature in the field heading into 2022.

ML Zoomcamp — Free Cohort Machine Learning Course

Alexey Grigorev's ML Zoomcamp — a free cohort-based machine learning course covering regression, classification, deployment, and MLOps fundamentals. A comprehensive practical curriculum from the author of Machine Learning Bookcamp.

Stanford CS 329S — Machine Learning Systems Design

Stanford CS 329S Machine Learning Systems Design — Chip Huyen's course on building production ML systems. Covers the full lifecycle from problem framing through data, training, deployment, and monitoring with real-world case studies.

Operationalizing Machine Learning: Forrester Research Report

Forrester Research report on operationalizing machine learning in enterprise settings, covering the organizational and technical requirements for moving models from development into production reliably. Useful snapshot of where enterprise ML practice stood in late 2021 and the key gaps between experimentation and production.

Faiss: The Missing Manual

Pinecone's comprehensive tutorial on FAISS (Facebook AI Similarity Search) — the foundational open-source library for approximate nearest neighbor search over dense vectors. Essential reading before choosing or building any vector search system.

Machine Learning Engineering for the Real World

A practical guide to ML engineering as a discipline — applying software engineering processes (agile, simplicity, iterative development) to ML projects from scoping through production. Makes the case that ML projects fail not from algorithmic complexity but from lack of engineering discipline around planning, experimentation, and deployment.

Models and Algorithms for Unlabelled Data

Vaibhav Verdhan's Manning book on unsupervised learning algorithms, covering clustering, dimensionality reduction, and anomaly detection with Python implementations on real-world datasets. Practical in orientation — stronger on applied case studies than mathematical rigor.

applied-ml: Papers and Tech Blogs on ML in Production

Eugene Yan's curated list of papers and engineering blog posts from companies sharing real-world ML in production — classification, recommendation, search, NLP, and more. One of the most useful ML reference repositories because it focuses on what actually shipped, not just what was published.

Collaborative Filtering Doesn't Work for Us

Chatroulette's engineering post explaining why traditional collaborative filtering didn't work for their video chat matching problem — no persistent user history, no item catalog, and the need for real-time matching under hard constraints. A useful case study on where standard recommender patterns break down.

Sentence Transformers: Pretrained Models

The Sentence-BERT pretrained models documentation — a reference for choosing the right sentence embedding model for semantic similarity, semantic search, or paraphrase detection tasks. The go-to resource when you need to pick a model before training your own.

Semantic Textual Similarity — Papers with Code Benchmark

Papers with Code's benchmark page for Semantic Textual Similarity — a task measuring how similar two sentences are in meaning, scored against human judgments. The canonical reference for comparing embedding models on this fundamental NLP task.

An Introduction to Statistical Learning (2nd Edition)

The second edition of James, Witten, Hastie, and Tibshirani's canonical intro-level statistical learning textbook, updated in 2021 to include deep learning, survival analysis, and multiple testing. It sits between undergraduate statistics and the more demanding Elements of Statistical Learning — the best entry point for practitioners who want rigorous but accessible ML foundations.

jupyter-text2code: English to Python in Jupyter

jupyter-text2code is a proof-of-concept Jupyter extension that converts English queries into Python code — an early 2021 demonstration of natural language to code generation in notebooks, before Copilot and ChatGPT made this mainstream.

Inferrd: Managed GPU Model Inference

Inferrd was a managed ML model serving platform with GPU support — deploy models via API without managing infrastructure. A 2021-era MLOps tool in the model deployment category that eventually got acquired.

Ray: A Distributed Framework for Emerging AI Applications

Micah Lerner's paper summary of Ray — UC Berkeley's distributed computing framework for AI workloads. Ray unified task-parallel and actor-based distributed computing in a Python-native API, becoming the foundation for libraries like RLlib, Tune, and later Ray Serve.

How UMAP Works

The official UMAP documentation explaining the mathematical intuition behind the algorithm — Riemannian geometry and fuzzy simplicial sets as the conceptual foundation. More rigorous than most t-SNE/UMAP comparisons and explains why UMAP preserves global structure better.

Alien Dreams: CLIP-Guided Image Generation

UC Berkeley ML blog's 2021 post on using CLIP for guided image generation — an early exploration of the CLIP+VQGAN/diffusion pipeline that preceded Stable Diffusion. Historically significant as a snapshot of generative AI before it became mainstream.

Made With ML

Made With ML is a free, project-oriented ML curriculum covering the full stack from data to deployment — written by Goku Mohandas. One of the most practical ML learning resources available, because it treats production concerns as first-class alongside model building.

Berkeley Document Summarizer

The Berkeley Document Summarizer is a learning-based extractive summarization system that uses syntactic compression and coreference constraints. Academic research code from Greg Durrett, representing the pre-neural era of NLP summarization work.

CS 7646: Machine Learning for Trading

CS 7646 is Georgia Tech's OMSCS course applying machine learning to financial trading — covering technical indicators, reinforcement learning for portfolio management, and backtesting. One of the few accredited graduate courses that bridges quant finance and ML.

Scikit-Learn Cheat Sheet (2021)

A cheat sheet for scikit-learn's main API patterns — estimator interface, preprocessing, model selection, and pipelines. Useful for quickly recalling the consistent fit/predict/transform pattern across all sklearn objects.

Your AI Skills Are Worth Less Than You Think

Ryszard Szopa's contrarian take on the AI skills premium — arguing that ML expertise is overvalued relative to domain knowledge and problem formulation skills, and that most ML work is commodity engineering that will commoditize further. A useful corrective to 2021 ML hype.

Deep Daze: Text to Image with CLIP and Siren

Deep Daze is Phil Wang's early text-to-image tool combining OpenAI's CLIP with Siren (implicit neural representations) — one of the first accessible open-source implementations of text-guided image generation, predating DALL-E and Stable Diffusion by over a year.

Many Models Workflows in Python

Alex Hayes's port of the R 'many models' workflow pattern to Python — fitting many models across groups using tidy data conventions. Bridges the gap between R's purrr/broom/tidymodels idioms and Python's pandas/scikit-learn ecosystem.

GPT-Neo: Open Source GPT-3 Scale Language Models

GPT-Neo is EleutherAI's open-source implementation of GPT-style language models at GPT-3 scale — the first serious open attempt to replicate GPT-3's capabilities before open-weight models became common. Historically significant as the origin of the open LLM movement.

Self-Supervised Learning: The Dark Matter of Intelligence

Yann LeCun and Ishan Misra's Facebook AI blog post arguing that self-supervised learning — learning from unlabeled data — is the key to human-level AI, analogous to the dark matter that makes up most of the universe's mass. Published ahead of a wave of self-supervised breakthroughs.

Shapash: Making Machine Learning Models Transparent

Shapash is MAIF's Python library for making ML models interpretable to non-technical stakeholders — wrapping SHAP and LIME with better visualizations and business-friendly explanations. Targets the gap between data scientists and decision-makers.

Powering Search and Recommendations at DoorDash

DoorDash's engineering blog post on their search and recommendation systems — covering how they rank restaurants and dishes, handle cold-start problems, and personalize results. A practical look at production recommendation systems at a major food delivery platform.

FACET: Human-Explainable AI

FACET is BCG Gamma's Python library for human-explainable AI — extending SHAP with interaction effects and redundancy-aware feature importance, plus simulation tools for model-based what-if analysis. More sophisticated than vanilla SHAP for understanding feature relationships.

SHAP: SHapley Additive exPlanations

SHAP (SHapley Additive exPlanations) is the standard Python library for explaining individual predictions from any ML model using game-theoretic Shapley values. It works across tree models, deep neural networks, and linear models, and produces both local and global interpretability.

LabML Neural Networks: Annotated Implementations

LabML Neural Networks is a collection of PyTorch implementations of ML papers with line-by-line annotations — making research papers readable by walking through the actual code. Covers transformers, diffusion models, GANs, and RL algorithms side-by-side with the paper math.

LazyPredict: Fit All scikit-learn Models in One Line

LazyPredict fits and evaluates all scikit-learn classifiers or regressors on a dataset with a single call, returning a sorted comparison table. A fast baseline scanner for figuring out which model family is worth investing in before tuning.

CS231n: Convolutional Neural Networks for Visual Recognition

Stanford CS231n: Convolutional Neural Networks for Visual Recognition — Andrej Karpathy's course that became the de facto entry point into deep learning for computer vision. The lecture notes remain among the best written explanations of CNNs, backprop, and training practice.

The Illustrated Transformer

Jay Alammar's illustrated walkthrough of the Transformer architecture — the most widely cited visual explainer of attention, encoder-decoder structure, and multi-head attention. A must-read before diving into any BERT, GPT, or T5 paper.

Sparse Matrices in SciPy

A visual explainer of sparse matrix formats in SciPy (COO, CSR, CSC, LIL, DOK) with animated illustrations showing how data is stored. Essential reading before working with high-dimensional feature matrices in ML or graph algorithms.

Probabilistic Machine Learning: An Introduction (Murphy)

Kevin Murphy's Probabilistic Machine Learning: An Introduction is the modern update to his 2012 ML textbook — free online, covering everything from linear models through deep learning with a probabilistic framing. The definitive graduate-level ML reference for 2022 onward.

10 Categories of Deep Recommendation Systems

James Le's survey of 10 categories of deep learning-based recommendation systems — from MLP and autoencoder approaches through attention-based and graph neural network methods. A useful taxonomy for understanding how the field moved beyond matrix factorization.

Crossminds: ML Research Video Platform

Crossminds was a platform for discovering and watching technical ML research videos — conference talks, paper presentations, and lectures organized by topic and institution. It aimed to be YouTube but with research-quality signal-to-noise filtering.

Machine Learning Mastery

Machine Learning Mastery is Jason Brownlee's prolific tutorial blog — hundreds of practical, code-first ML tutorials covering scikit-learn, Keras, time series, NLP, and more. Known for quantity and accessibility rather than depth, but a go-to reference for how-to implementations.

mlart.co — Machine Learning Art Gallery

mlart.co is a curated gallery of machine learning art projects — generative works created using neural networks, GANs, style transfer, and other ML techniques. A showcase of creative applications of AI at the intersection of art and research.

Emerging Architectures for Modern Data Infrastructure

Andreessen Horowitz maps the emerging stack for modern data infrastructure — separating analytics, ML, and operational workloads into distinct architectural patterns. A useful taxonomy of the data tooling landscape circa 2020.

Gradient Boosted Decision Trees

An illustrated explainer of gradient boosted decision trees — how sequential weak learners correct prior errors by fitting residuals, and how this connects to gradient descent. Covers regression, binary classification, and multi-class variants.

Rules of Machine Learning: Best Practices for ML Engineering

Martin Zinkevich's 43-rule guide from Google on practical ML engineering, organized around the principle that most gains come from good features and solid infrastructure rather than clever algorithms. A pragmatic counterweight to academic ML papers — the kind of advice that separates production systems from demos.

ELI5 — sklearn Explainability Module

ELI5's sklearn module provides model explanation tools for scikit-learn estimators — feature importance, prediction decomposition, and permutation-based importance across linear models, tree ensembles, and SVMs. The explainability companion for sklearn workflows.

sktime — Unified Machine Learning with Time Series

sktime is a Python library providing a unified scikit-learn-compatible interface for time series machine learning — forecasting, classification, regression, clustering, and anomaly detection. Solves the ecosystem fragmentation problem for temporal data.

Multi-Armed Bandits and the Stitch Fix Experimentation Platform

Stitch Fix's blog on multi-armed bandits as an alternative to A/B testing — Thompson Sampling routes traffic toward better-performing arms dynamically, reducing wasted exposure. Strong motivation for when bandits beat traditional experimentation.

Data Science Interview Questions and Answers

A community-maintained GitHub repo of data science and ML interview questions and answers — covering statistics, machine learning theory, algorithms, and coding. A useful study guide and signal for what interviewers actually test.

Unpopular Opinion — Data Scientists Should Be More End-to-End

Eugene Yan argues that data scientists deliver more value when they own the full problem lifecycle — from identifying the problem through production deployment. Fewer handoffs, better context, faster iteration, and stronger ownership.

ML in Production — Best Practices for Real-World ML Systems

ML in Production is a blog and newsletter focused on building and operating real-world ML systems — covering experimentation programs, deployment, monitoring, and the organizational practices that make ML succeed in production environments.

Explicit Recommender System — Matrix Factorization in PyTorch

A tutorial implementing explicit recommender systems via matrix factorization in PyTorch — using Embedding layers for user and item factors, trained with alternating gradient descent. Concrete implementation of collaborative filtering fundamentals.

Machine Learning from Scratch

Machine Learning from Scratch is a free online book deriving seven core ML algorithms from first principles — linear regression, logistic regression, naive Bayes, decision trees, ensembles, and neural networks. Mathematically rigorous, aimed at practitioners who want to understand how algorithms work mechanistically.

Effective Testing for Machine Learning Systems

Jeremy Jordan's framework for testing machine learning systems — pre-train checks, invariance tests, directional expectation tests, and minimum functionality tests. Organizes tests around model 'skills' rather than code structure.

TikTok and the Sorting Hat

Eugene Wei's landmark essay arguing TikTok's algorithm is the product — not the content or creators — and that it functions as a 'sorting hat' that matches users to their taste tribes faster than any previous social network. Required reading for anyone thinking about content recommendation systems.

Data Science Meets DevOps: MLOps with Jupyter, Git, and Kubernetes

Kubeflow's blog post on MLOps practices combining Jupyter notebooks, Git, and Kubernetes — the infrastructure side of making ML reproducible, collaborative, and deployable. Establishes the canonical MLOps stack for Kubernetes-based ML teams.

Mathematics for Machine Learning: Multivariate Calculus

Imperial College London's Mathematics for Machine Learning Multivariate Calculus course on YouTube — free coverage of partial derivatives, gradients, the Jacobian, Hessian, and optimization techniques needed to understand how machine learning models are trained.

DeepDream: How Alexander Mordvintsev Excavated the Computer's Hidden Layers

The story behind Google's DeepDream — how researcher Alexander Mordvintsev discovered that running gradient ascent on a convolutional network's hidden layers produces psychedelic imagery that reveals what features the network learned. A landmark moment in neural network interpretability.

150+ Best Machine Learning, NLP, and Python Tutorials

Robbie Allen's curated list of 150+ machine learning, NLP, and Python tutorials, organized by topic and difficulty. A high-signal link aggregation for 2020-era ML self-study, with commentary on what each resource covers.

Top Machine Learning Books Recommended by Experts (2020)

MentorCruise's expert-recommended machine learning book list for 2020 — recommendations from actual ML practitioners and researchers, covering foundational math, statistical learning, deep learning, and applied ML.

Speech and Language Processing (Jurafsky & Martin)

The free online draft of Jurafsky and Martin's 'Speech and Language Processing' — the canonical NLP textbook, covering everything from regex to transformers, now in its 3rd edition. Freely available online in draft form and unusually well-written for a technical textbook.

Machine Learning Roadmap

Daniel Bourke's visual roadmap connecting the core concepts of machine learning, what to learn first, and what tools to use — built as a community-oriented guide for self-learners entering the field in 2020.

The Overfitted Brain: Dreams Evolved to Assist Generalization

A 2020 paper proposing that dreams evolved as a biological regularization mechanism — the brain 'trains' on noisy, hallucinated data during sleep to prevent overfitting to waking experience. A striking bridge between ML theory and sleep neuroscience.

Using Machine Learning to Rank Search Results (Part 2)

A practical tutorial on applying machine learning to re-rank search results — part 2 of a series covering feature engineering, training data construction, and evaluation for LTR in production search. Hands-on complement to the more theoretical LTR literature.

Learning to Rank with Vespa

Thiago Martins' tutorial on learning-to-rank with Vespa for text search — covers how Vespa's native ML integration makes LTR simpler than Elasticsearch/Solr plugins, with models evaluated inside the search engine. Vespa's approach to embedding LTR natively.

Learning to Rank for Personalised Search (Yandex Kaggle Competition)

Yanir Seroussi's Kaggle competition post-mortem on Yandex Search Personalisation — applying learning-to-rank techniques to personalized search with behavioral signals. A practical case study of LTR on real search logs.

Learning to Rank 101: Linear Models

OpenSource Connections' foundational explainer on linear models for learning-to-rank — the first step before gradient boosted trees. Covers feature engineering and the intuition for why linear LTR models are both a useful starting point and a useful baseline.

Is Your Infrastructure Ready for Learning to Rank?

OpenSource Connections' guide to the infrastructure required before you can deploy learning-to-rank — covering the judgment lists, feature logging pipelines, training data management, and model versioning that most LTR tutorials skip. The operational reality of LTR in production.

Intuitive Explanation of Learning to Rank: RankNet, LambdaRank, and LambdaMART

Nikhil Dandekar's intuitive explainer on the evolution from pointwise to pairwise to listwise learning-to-rank, covering RankNet, LambdaRank, and LambdaMART. One of the best conceptual introductions to how modern search ranking algorithms actually work.

Introduction to Personalized Search

Recombee's introduction to personalized search — connecting the recommendation systems world to search, covering how behavioral signals (clicks, purchases) can be used to personalize result ranking per user. Bridges the gap between generic LTR and user-specific personalization.

Best Data Science Books According to the Experts

Built In's curated list of best data science books by category, with expert commentary on what each covers and who it's for. Spans introductory statistics through advanced ML, with a practical rather than academic bias.

Applied ML: Papers and Blogs on ML in Production (2020)

Eugene Yan's curated GitHub list of papers and blog posts on ML in production — original 2020 bookmark of this now-landmark repository. Covers recommendation, search, NLP, data quality, feature engineering, and more from companies that actually shipped these systems.

Why GPT-3 Matters

Leo Gao's early analysis of why GPT-3 is qualitatively different from prior language models — written in May 2020 when GPT-3 was first announced. One of the clearer technical takes on what the scaling breakthrough meant, from someone who later worked on it.

Reflecting on a Year of Making Machine Learning Actually Useful

Shreya Shankar's honest reflection on a year trying to make machine learning actually useful in industry — covering the gap between academic ML and production, the underappreciated role of data work, and why most ML projects fail before the model stage. One of the most cited personal essays in the MLOps space.

Building a Recommender System Using Embeddings

Drop Engineering's walkthrough of building a brand recommender using learned embeddings — training entity embeddings from user-brand interaction data to capture brand similarity in continuous vector space. A practical case study in embedding-based recommendations.

PRML: Pattern Recognition and Machine Learning Algorithms in Python

Python implementations of algorithms from Bishop's 'Pattern Recognition and Machine Learning' — the canonical probabilistic ML textbook. Bridges the gap between the math in the book and working code.

160 Data Science Interview Questions

Alexey Grigorev's compilation of 160 data science interview questions across statistics, machine learning, SQL, and programming — a broad coverage reference for data science interview preparation, organized by topic.

Your Ultimate Learning Path to Become a Data Scientist in 2020

Analytics Vidhya's structured learning path to become a data scientist in 2020 — a step-by-step curriculum covering statistics, Python, ML algorithms, and tools, with specific resource recommendations at each stage.

Superpixel Segmentation — IVRL

The IVRL lab at EPFL's research page on superpixel segmentation — home of the SLIC algorithm, which became the dominant superpixel method due to its speed and perceptual uniformity. Superpixels are a fundamental preprocessing step in classical computer vision pipelines.

Practical Deep Learning for Coders — fast.ai

fast.ai's Practical Deep Learning for Coders — Jeremy Howard and Rachel Thomas's free course that inverted the standard pedagogy: start with working image classifiers, then learn the theory underneath. Democratized deep learning at a moment when most education assumed a PhD on-ramp.

Over 150 of the Best Machine Learning, NLP, and Python Tutorials

A 2017 curated list of 150+ ML, NLP, and Python tutorials organized by topic — a snapshot of the best practitioner learning resources from the pre-transformer era. Useful as a historical reference for what the field considered canonical learning material at the time.

Neuro-Fuzzy

Neuro-fuzzy systems combine neural network learning with fuzzy logic's ability to handle imprecise, rule-based reasoning — a hybrid AI approach popular in consumer electronics and control systems during the 1990s-2000s. Famous for appearing on washing machine panels before deep learning made these labels obsolete.

Speeding up scikit-learn Workflow Using a High-Performance Go Proxy

A 2016 blog post on using HoverPy — a Go-based HTTP recording/replay proxy — to speed up scikit-learn workflows that make external HTTP calls. An early example of applying API mocking techniques to ML pipelines for faster, deterministic testing.

How to Use t-SNE Effectively

Martin Wattenberg, Fernanda Viégas, and Ian Johnson's interactive Distill article on how t-SNE maps actually work — and how to avoid misreading them. Essential reading before trusting any t-SNE plot.

LexVec

LexVec is a Go implementation of a word embedding model that factorizes the PPMI matrix with position-dependent weighting, outperforming Word2Vec on several NLP benchmarks at the time. An interesting mid-2010s data point in the race to improve on Word2Vec before attention-based models took over.

Gradient Boosting Explained

Alex Rogozhnikov's interactive 3D visualization of gradient boosting — shows how decision boundaries evolve as the ensemble builds up trees. One of the cleaner intuition-builders for gradient boosting before XGBoost dominance made it feel like a black box.

Approaching (Almost) Any Machine Learning Problem

Abhishek Thakur's systematic framework for tackling any supervised ML problem — from data cleaning and feature engineering through model selection and stacking. One of the most-shared practical ML workflow guides from the Kaggle blog era.

Automatic Sample Layout (VAE)

Kyle McDonald's EYEO 2016 demonstration of automatic audio sample organization using a Variational Autoencoder — the VAE learns a latent representation of sounds and arranges them in 2D space so similar samples cluster together. An early application of generative models to creative tools.

Fizz Buzz in TensorFlow

Joel Grus deliberately solves FizzBuzz using TensorFlow — a satirical demonstration of using a neural network where simple conditional logic would do. A sharp critique of ML overcomplexity disguised as a tutorial.

The ROC Curve Explained

A visual explanation of the Receiver Operating Characteristic (ROC) curve and AUC for binary classifier evaluation. One of the clearest introductions to the concept for practitioners learning to assess model performance beyond accuracy.

The Three Cultures of Machine Learning

Jason Eisner's JHU tutorial describing three distinct cultures within machine learning — the statistical/probabilistic, the algorithmic/computational, and the geometric/optimization traditions. Useful framing for understanding why ML researchers sometimes talk past each other.

Machine Learning's First Cheating Scandal

A post-mortem on a Kaggle-era incident where participants gamed the public leaderboard through repeated test-set probing — effectively overfitting to held-out data via the submission API. Raised serious questions about how ML benchmarks and competitions should be designed.

Where Are the Opportunities for Machine Learning Startups?

A 2015 VC perspective on where machine learning startups have genuine opportunities — identifying verticals where ML adds defensible value vs. where it's a feature, not a company. Notable for emphasizing ML education as an overlooked category.

Overfitting, Regularization, and Hyperparameters

DS Walter's practitioner explainer on overfitting, regularization techniques, and hyperparameter tuning — covering L1/L2 penalties, dropout, and cross-validation. A clear introduction to the bias-variance tradeoff for working data scientists.

Dive into Machine Learning

A curated GitHub guide for learning machine learning hands-on with Jupyter notebooks and scikit-learn — one of the most-starred beginner ML resources of 2015. Its strength is pairing interactive notebooks with curated external readings rather than building yet another tutorial from scratch.

That Time When "Recurrent Neural Network" Was Replaced with "Dog"

The joke where replacing 'recurrent neural network' with 'dog' in ML papers produces sentences that remain plausible — a 2015 meme that captured real frustration with AI hype language. Still a useful test for vague ML claims.

What is TF-IDF? The 10 Minute Guide

A 10-minute introduction to TF-IDF — the classic term-weighting scheme that balances how often a word appears in a document against how rare it is across the corpus. Still one of the most useful baselines in text analysis despite being decades old.

benchm-ml: ML Algorithm Benchmark Comparison

A systematic benchmark of machine learning algorithms across platforms and implementations — comparing gradient boosting, random forests, neural networks, and others on speed and accuracy. One of the best empirical references for choosing between ML tools in 2015, when the xgboost vs sklearn debate was live.

A Visual Introduction to Machine Learning

A scrollytelling visual explanation of decision trees by Stephanie Yee and Tony Chu — widely considered one of the best examples of data visualization as pedagogy. It makes the algorithm's splitting logic tangible without any math.

10 Types of Regressions. Which One to Use?

A reference guide to 10 regression types and when to use each — covering linear, logistic, ridge, lasso, polynomial, and more. The kind of practical decision map that was useful before sklearn docs became as thorough as they are now.

Implicit Feedback and Collaborative Filtering

A technical post on implicit feedback in collaborative filtering — when you have views, clicks, and dwell time instead of explicit ratings. Covers the ALS (alternating least squares) approach and linear algebra tricks for making matrix factorization tractable at scale.

The Perilous World of ML: Pipeline Jungles and Hidden Feedback Loops

John Foreman on the hidden technical debt in ML systems — pipeline jungles and feedback loops that make production ML fragile in ways that pure model metrics never reveal. Anticipates the 'Hidden Technical Debt in Machine Learning Systems' Google paper by months.

The Geometry of Classifiers

Nina Zumel's geometric treatment of machine learning classifiers — how decision boundaries, margins, and probability regions look in feature space. Builds visual intuition for why different classifier families make the tradeoffs they do.

Interpreting Random Forests

A deep dive into techniques for interpreting random forest models — going beyond accuracy to understand what the model has learned. Particularly useful because random forests were the dominant ensemble method before gradient boosting took over, and interpretability was the main complaint against them.

The Current State of Machine Intelligence (2014)

Shivon Zilis's 2014 landscape of machine intelligence companies — an early attempt to map the ML startup ecosystem before deep learning had fully taken over. A historical snapshot of what the AI industry looked like before the transformer era.

Introduction to Principal Component Analysis (PCA)

Laura Diane Hamilton's accessible introduction to Principal Component Analysis — covering the geometric and algebraic intuition without requiring a linear algebra background. One of the cleaner beginner explanations of a technique that's notoriously hard to explain well.

Calibrating Classifier Probabilities

Daniel Nee's post on calibrating classifier output probabilities — the underappreciated problem that a model's predicted probability of 0.8 doesn't always mean there's an 80% chance of the positive class. Essential reading before using model outputs for decision-making.

Kernel PCA

Sebastian Raschka's tutorial on Kernel PCA — extending standard PCA to capture non-linear structure using the kernel trick with RBF kernels. Includes Python implementation, making it one of the clearest practical explanations of the technique available in 2014.

Y Combinator 2014 Data Science Startups

MLWave's survey of Y Combinator's 2014 data science and machine learning startups — a snapshot of where the industry was investing in applied ML before the current deep learning era. Shows which problem domains were being commercialized in the pre-GPT wave.

New to Machine Learning? Avoid These Three Mistakes

A Medium post on the three common mistakes beginners make in machine learning — likely covering data leakage, premature model selection, and ignoring baselines. The kind of counterintuitive advice that every bootcamp graduate needs before their first real project.

Bayesian Machine Learning on Apache Spark

Cloudera's engineering blog post on implementing Bayesian machine learning on Apache Spark — combining probabilistic inference with distributed computation. A technically ambitious combination that was ahead of most production ML stacks in 2014.

Scikit-learn Pipelines and FeatureUnions

Zac Stewart's deep dive into composing scikit-learn Pipelines and FeatureUnions — showing how to chain preprocessing steps, branch feature transformations, and combine them back together while preventing data leakage. The definitive 2014 guide to production-ready sklearn code.

Security Data Science Papers

Covert.io's curated list of academic papers applying data science and machine learning to security problems — covering network intrusion detection, malware classification, anomaly detection, and more. A reference for practitioners working at the intersection of ML and cybersecurity.

Pattern Classification Data Glossary

Sebastian Raschka's data science and machine learning glossary from his pattern_classification GitHub repo — a community reference that grew into one of the most-linked ML terminology resources on GitHub. Useful for grounding conversations in shared vocabulary.

10 Tips for Better Deep Learning Models

Laura Diane Hamilton's ten practical tips for improving deep learning model performance, covering data preparation, architecture choices, regularization, and training tricks. A snapshot of practitioner wisdom circa 2014, before the era of giant pretrained models made many of these tradeoffs less urgent.

A Tour of Machine Learning Algorithms

Jason Brownlee's taxonomy of machine learning algorithms organized by learning style and similarity — a map of the algorithm space useful for orienting newcomers. The categorization gives a mental model for when to reach for which algorithm family.

Data in Practice

Data in Practice is a tutorial blog by Daniel Baumgartel covering coding, algorithms, data science, machine learning, and distributed computing — practical implementations with working code. A representative example of the practitioner-written technical blogs that shaped the 2014 data science self-education ecosystem.

An Introduction to Gradient Descent and Linear Regression

A clear walkthrough of gradient descent applied to linear regression, with code — a good foundational tutorial connecting the mathematical update rule to a concrete implementation. One of the most-linked introductions to the algorithm for newcomers to ML.

Must Read Before Attending Any Data Science Interview

Data Science Central's pre-interview reading list covering statistics, machine learning, coding, and business acumen questions asked in 2014 data science interviews. A historical snapshot of what the field considered core practitioner knowledge.

Zipfian Academy: Week 3 — Or: "Bridging the Gap"

Week 3 of Zipfian Academy: 'Bridging the Gap' — the week where statistics met machine learning, building the conceptual bridge from probability distributions to predictive models. The moment in the curriculum where the pieces start connecting.

Google Turns to Machine Learning to Build a Better Data Centre

A 2014 report on Google applying machine learning to optimize data center cooling — one of the first public disclosures that Google was using neural networks to automate infrastructure decisions. DeepMind later published the full methodology in 2016.

20 Short Tutorials All Data Scientists Should Read and Practice

Data Science Central's curated list of 20 short tutorials spanning statistics, machine learning, SQL, R, and Python — designed as quick-read practice exercises. A community-sourced map of what foundational skills were considered essential for data scientists in 2014.

Bayes Impact Initiative

Bayes Impact is a nonprofit that applies data science and machine learning to social problems — healthcare, criminal justice, public policy. Founded in 2014, it embodies the 'data for good' ethos that was prominent in the San Francisco tech community at the time.

How Airbnb Used Conditional Probability Models to Improve Search

How Airbnb used conditional probability models and data visualization to improve search ranking — an early public example of a major consumer company describing ML-driven search personalization. Significant for its transparency about the modeling approach.

Converting Categorical Data into Numbers with Pandas and Scikit-Learn

FastML tutorial on converting categorical variables to numeric form using Pandas and scikit-learn's LabelEncoder and OneHotEncoder. A foundational data preprocessing step that trips up many beginners.

Programmatically Understanding the Expectation Maximization Algorithm

Nipun Batra's programmatic walkthrough of the Expectation Maximization algorithm — showing the E and M steps in code to build intuition for how EM converges. Makes the algorithm's alternating optimization structure tangible.

Outlier Detection via Markov Chain Monte Carlo

Bugra Akyildiz's walkthrough of outlier detection using Markov Chain Monte Carlo via PyMC — fitting a Bayesian mixture model to separate inliers from outliers using posterior inference. A more principled alternative to distance-based outlier methods.

How to Get Started with Machine Learning in Python

Prismatic story aggregating a 'How to Get Started with Machine Learning in Python' tutorial — a 2014 entry point to scikit-learn, NumPy, and Pandas for ML practitioners. Reflects the era's onboarding gap before dedicated ML courses existed.

ConvNetJS Demo: Classify Toy 2D Data

Andrej Karpathy's ConvNetJS demo for classifying 2D toy datasets — a real-time browser visualization of a neural network training on user-drawn boundaries. One of the first compelling interactive deep learning visualizations.

Machine Learning for Growth — Zipfian Academy Workshop

Zipfian Academy's Machine Learning for Growth workshop — a short-format course on applying ML to product growth problems like churn prediction, A/B testing, and cohort analysis. Reflects the 2014 convergence of data science with growth hacking.

My Solution for the Galaxy Zoo Challenge — Sander Dieleman

Sander Dieleman's winning solution for the Galaxy Zoo Kaggle challenge using convolutional neural networks to classify galaxy morphology from images. A landmark result showing CNNs achieving human-level performance on a citizen science dataset.

The Kernel Trick

Eric Kim's explanation of the kernel trick in support vector machines — how kernels enable SVMs to classify non-linearly separable data by implicitly mapping it to a higher-dimensional space. One of the cleaner intuitive explanations of a mathematically dense concept.

Scikit-Learn: Model Validation and Testing (PyCon 2013 Notebook)

Jake VanderPlas's PyCon 2013 notebook on model validation and testing in scikit-learn — covers train/test splits, cross-validation, and model selection in executable notebook form. A practical tutorial that shaped how Python practitioners learned to evaluate models.

Some Useful Machine Learning Libraries

A 2013-era survey of machine learning libraries across Python, R, Java, and C++ — a snapshot of the fragmented ML tooling landscape before scikit-learn and deep learning frameworks consolidated the field.

Not Just for Search: Using ElasticSearch with Machine Learning Algorithms

An early (2013) case for using Elasticsearch beyond full-text search — specifically as a substrate for machine learning applications like nearest-neighbor lookup and feature indexing. Pre-dates the vector search era but anticipates the same pattern.

Statistical Data Mining Tutorials — AutonLab (CMU)

Carnegie Mellon's AutonLab tutorial series on statistical data mining — covering Bayesian networks, clustering, density estimation, dimensionality reduction, and anomaly detection. A rigorous academic reference that was freely available online before MOOC culture normalized that.

How To Choose The Right Test Options When Evaluating Machine Learning Algorithms

Jason Brownlee's guide to choosing between hold-out validation, k-fold cross-validation, and bootstrap estimation when evaluating ML algorithms. Covers when each approach is appropriate given dataset size and computational budget.

Cloudera Rebuilding Machine Learning for Hadoop with Oryx

GigaOm's coverage of Cloudera launching Oryx — an open-source ML-on-Hadoop framework using the Lambda Architecture for batch retraining plus real-time serving. An early attempt to make production machine learning first-class on Hadoop.

Design and Run Your First Experiment in Weka

Jason Brownlee's Machine Learning Mastery guide to designing and running experiments in Weka — the GUI-based ML tool from Waikato. Shows how to set up proper comparative experiments with statistical testing, not just running one algorithm.

NIPS 2012 Paper #1338

A NIPS 2012 (NeurIPS 2012) conference paper — saved in the same timeframe as deep learning and distributed neural network research. Likely related to neural architecture or distributed/large-scale training given the surrounding bookmarks.

Why You Should Be Spot-Checking Algorithms on Your Machine Learning Problems

Jason Brownlee's Machine Learning Mastery argument for spot-checking — trying 10-15 algorithms quickly on a new dataset before committing to tuning any single one. Saves time by letting the data tell you which algorithm families are worth investing in.

LinkedIn Acquires Bright: Big Data Job Matching

LinkedIn's ~$120M acquisition of Bright, a job-matching startup that used machine learning to score candidate-job fit from resume and job description text. A signal of how ML-powered matching was becoming central to professional network platforms.

5 Mistakes Programmers Make When Starting in Machine Learning

Jason Brownlee's list of five mistakes programmers make when transitioning into machine learning — over-focus on theory, skipping problem definition, ignoring data quality, neglecting model evaluation, and treating ML as a programming challenge. The practitioner's onramp.

Optimism in the Face of Uncertainty: the UCB1 Algorithm

Jeremy Kun's accessible treatment of the UCB1 algorithm — the principle of 'optimism in the face of uncertainty' formalized as a bandit algorithm with proven regret bounds. Shows why adding a confidence bonus to estimated rewards elegantly solves the exploration-exploitation tradeoff.

Multi-Armed Bandits

Cameron Davidson-Pilon's blog post on multi-armed bandits from a Bayesian perspective — draws on the same probabilistic programming intuition as his 'Bayesian Methods for Hackers' book. Frames bandits as the natural application of iterative belief updating.

Multi-Armed Bandit Experiments

Analytics blog post on multi-armed bandit experiments as a replacement for static A/B testing in website optimization — covers epsilon-greedy, UCB, and Thompson Sampling with practical framing for product teams.

A Book About Bandit Algorithms

John Myles White's free book on bandit algorithms for website optimization — covers epsilon-greedy, softmax, UCB, and Thompson Sampling with practical web application examples. An accessible bridge from theory to product experimentation.

Understanding Multi-Armed Bandit Algorithms

DataBozo's conceptual explanation of multi-armed bandit algorithms — epsilon-greedy, UCB, and Thompson Sampling compared from first principles. Aimed at practitioners who want to understand the tradeoffs before implementing.

Visualizing K-Means Clustering

Naftali Harris's interactive visualization of k-means clustering — place points on a canvas and watch the algorithm converge step by step. Exposes why initialization matters and where k-means fails.

Immunity Project: Using ML to Combat HIV

Immunity Project used machine learning on HIV genomics data to design a vaccine targeting the virus's conserved regions — accepted into Y Combinator as only the second nonprofit in its history. An early example of ML applied to high-stakes biology.

Troll Detection with Scikit-Learn

Impermium's Kaggle blog post on building a troll detection classifier with scikit-learn — text features, gradient boosting, and the practical challenges of training on imbalanced toxic comment data.

k-Nearest Neighbor Classification for Harrisburg Crimes

k-Nearest Neighbor classification applied to predicting crime type in Harrisburg, PA from location and time features. A typical early applied ML blog post that makes the algorithm concrete with a public dataset.

Computer Science: The Learning Machines

Nature News's January 2014 feature on machine learning and deep neural networks going mainstream — the moment the field started reaching a broader scientific audience. Documents the Hinton-LeCun-Bengio wave before it was inevitable.

Reproducible Machine Learning Results By Default

Jason Brownlee's Machine Learning Mastery post on setting random seeds and other practices to make ML experiments reproducible by default. A basic hygiene checklist that surprisingly many practitioners skip.

Vowpal Wabbit Tutorial for the Uninitiated

Zinkov's beginner tutorial for Vowpal Wabbit — the fast online learning library from Microsoft Research. Shows the input format, basic invocation, and when VW's out-of-core learning beats batch methods.

An Easy Way to Bridge Between Python and Vowpal Wabbit

Steve's Machine Learning Blog post on using Python to feed data to Vowpal Wabbit via subprocess — a simple bridge for calling VW from Python workflows without a native binding.

The Remarkable k-means++

Larry Wasserman's Normal Deviate blog post on k-means++ — the 2007 initialization trick from Arthur and Vassilvitskii that gives k-means an O(log k) approximation guarantee and better convergence in practice.

Regularizing Neural Networks with Dropout and DropConnect

FastML's comparison of Dropout and DropConnect — two techniques for regularizing neural networks by randomly zeroing activations or weights during training. Clarifies that DropConnect's CIFAR-10 SOTA came from model ensembling, not the technique itself.

A Not-So-Basic Neural Network in Python

Daniel Rodriguez's practical tutorial on implementing a non-trivial neural network in Python from scratch — going beyond the toy perceptron examples to show backpropagation and training on real data. A mid-2013 hands-on coding reference.

Teaching a Computer to Read: NLP Hacking in Python

Scripted blog's introduction to NLP in Python — using NLTK for tokenization, part-of-speech tagging, named entity recognition, and sentiment analysis. A practical hands-on introduction to text processing written for a content-tech company's engineering blog.

Self-Study Guide to Machine Learning

Machine Learning Mastery's self-study guide to machine learning — Jason Brownlee's staged curriculum for practitioners who want to learn ML without a formal background. Prescriptive and pragmatic: pick an algorithm, implement it, run it on data, iterate.

Metacademy — Bayesian Machine Learning Roadmap

Roger Grosse's curated Metacademy roadmap for learning Bayesian machine learning from scratch — a sequenced path through Bayesian inference, probabilistic graphical models, approximate inference, and nonparametric methods. The systematic route through a complex prerequisite landscape.

Metacademy

Metacademy is a 'web of knowledge' for machine learning — a dependency graph of concepts where each node links to learning resources and its prerequisites. The idea: show exactly what you need to know before you can understand any given topic.

Bayesian Machine Learning via Category Theory

Culbertson and Sturtz's 2013 paper applying category theory to Bayesian machine learning — using the Kleisli category of the Giry monad to formalize supervised learning, stochastic processes as priors, and the Kalman filter. Heavy theory, but part of the broader push to give probabilistic ML rigorous foundations.

BayesDB

BayesDB from MIT CSAIL's probabilistic computing group — a database system that lets you query statistical relationships using a SQL-like language (BQL) without specifying a model. Automatically infers the right probabilistic model from data.

Advantages of Different Classification Algorithms

Quora thread on the tradeoffs between classification algorithms — naive Bayes, SVM, decision trees, logistic regression, k-NN, and neural networks. A practical reference for choosing the right algorithm given your data characteristics and constraints.

Deep Learning 101

Introductory survey of deep learning for non-specialists — covers hierarchical representation learning, RBMs, autoencoders, and the four core obstacles. Written to help readers mentally filter hype from substance in 2013.

Statistical Models Can Predict Kickstarter's Success Within 4 Hours

Ars Technica coverage of research showing statistical models can predict Kickstarter campaign success within the first four hours of launch. Early-momentum signals turn out to be highly predictive — social proof compounds quickly.

Why Machine Learning Fails

Louis Dorard's analysis of why machine learning projects fail in practice — usually not because the algorithms are wrong but because the problem setup, data quality, or evaluation approach is broken. The engineering side of ML is where most projects die.

Machine Learning Tutorial: The Naive Bayes Text Classifier

DatumBox's tutorial on the Naive Bayes text classifier — explaining the math, the conditional independence assumption, and how to implement it from scratch. A standard first classifier for anyone learning NLP and ML simultaneously.

sklearn-pandas: Bridge Between pandas and scikit-learn

sklearn-pandas is a library bridging pandas DataFrames and scikit-learn's pipeline API — enabling column-level transformations with named features rather than anonymous numpy arrays. Fills a friction point that frustrated every data scientist using both libraries together.

Python Extensions to Do Machine Learning

A roundup of Python extensions and libraries for machine learning circa 2013 — the ecosystem before it had fully consolidated around scikit-learn, NumPy, pandas, and matplotlib as the canonical stack.

IBM Wants to Use Big Data to Predict Heart Disease Long Before It Strikes

VentureBeat on IBM's initiative to use big data and predictive models to forecast heart disease years before symptoms appear. A 2013 case study in applying ML to healthcare at scale — ambitious but illustrative of where the field thought it was going.

Gaining Access to the Best Machine-Learning Methods

Strata/O'Reilly piece on how access to the best machine learning methods was democratizing — cloud computing, open-source libraries (scikit-learn, Spark MLlib), and MOOCs were putting frontier ML within reach of non-specialists.

Teaching a Computer to Read: NLP Hacking in Python

Scripted's NLP hacking tutorial in Python — covering tokenization, part-of-speech tagging, named entity recognition, and text classification with NLTK and scikit-learn. An applied introduction to NLP for data scientists.

Random Forests Algorithm

An introduction to the Random Forests algorithm — explaining how ensembling many decorrelated decision trees reduces variance and produces a robust classifier. A standard explainer from the Data Science Central era.

Why Recommendation Engines Are About to Get Much Better

Coverage of advances in recommendation engine technology in 2013 — driven by larger datasets, better collaborative filtering, and contextual signals. The moment when personalization was transitioning from a luxury to an expectation.

Machine Learning Skills for Jobs (2013)

A Quora thread on what skills machine learning jobs required in 2013 — the answer set reflects the early data science job market before the role fragmented into ML engineer, data scientist, and AI researcher specializations. A snapshot of what practitioners thought mattered at the time.

Designing Machine Learning Frameworks

Mikio Braun's reflection on the design principles behind ML frameworks — the tensions between flexibility and usability, scikit-learn's consistency, and what makes a good ML library API. Written by the author of jBLAS.

Six Steps in Data Science

A 2013 blog post laying out six practical steps in a data science workflow — from problem framing through data collection, exploration, modeling, evaluation, and deployment. A snapshot of how practitioners were thinking about the discipline before MLOps and production ML tooling matured.

Deep Support Vector Machines

A video lecture on Deep Support Vector Machines from ROKS 2013 — hybrid architectures combining deep feature learning with SVM classification. A snapshot of the moment researchers explored whether SVMs and deep learning could coexist before end-to-end networks won out.

Jeremy Howard on the Big Data Obsession

Jeremy Howard's Quora answer on why the 'big data' obsession was somewhat misplaced — arguing that algorithms and predictive modeling matter more than raw data volume, and that the real value was in applying machine learning, not just collecting more data. A contrarian view from someone who knew ML deeply before the hype peaked.

Variable Importance in Neural Networks

R-bloggers post on measuring variable importance in neural networks — techniques like the Garson algorithm and Olden's method for attributing prediction contributions to input features. An early attempt at neural network interpretability before SHAP, LIME, and modern explainability tools existed.

Stanford CS229: Machine Learning Course Materials

Stanford CS229 Machine Learning course materials page — Andrew Ng's graduate ML course whose lecture notes became canonical self-study references. The notes on supervised learning, SVMs, neural networks, and probabilistic models were some of the clearest technical writing in the field.

Target Practice: The Power of Predictive Analytics

Forbes coverage of Target's predictive analytics program — the famous pregnancy prediction story that showed retail chains could infer major life events from purchase patterns. A case study in both the power and the social friction of behavioral prediction at scale.

Introduction to One-Class Support Vector Machines

A practical introduction to one-class SVMs — the variant of Support Vector Machines used for anomaly detection and novelty detection when you only have examples of normal behavior. Useful when collecting labeled anomaly examples is impractical or impossible.

The State of Recommender Technology (2013)

A 2013 survey of recommender system technology covering collaborative filtering, content-based approaches, and the state of the field before deep learning took over. Published by Data Community DC alongside coverage of CoBrain, a startup working on recommendation infrastructure.

Mining of Massive Datasets (Stanford)

The Stanford textbook by Rajaraman and Ullman on algorithms for mining massive datasets — locality-sensitive hashing, PageRank, collaborative filtering, stream algorithms, and more. Freely available online and a standard reference for large-scale data algorithms.

Machine Learning Cheat Sheet

Emanuel Ferm's machine learning cheat sheet — a compact reference covering the main supervised and unsupervised learning algorithms with notes on when to apply each. A quick-reference for practitioners who know the algorithms but want a memory aid for their properties.

Where Can I Find Large Datasets Open to the Public?

A 2013 Quora thread aggregating large public datasets (≥1 GB) for machine learning and data science research. A community-curated snapshot of the open data landscape before Kaggle, HuggingFace Datasets, and government open data portals became the primary discovery mechanisms.

Some Datasets Available on the Web

Data Wrangling Blog's curated list of publicly available datasets for machine learning and data analysis. An early community resource for finding training data before Kaggle and HuggingFace centralized dataset discovery.

A Thousand-Foot View of Machine Learning

A high-level orientation to machine learning from 2009 — the major paradigms (supervised, unsupervised, reinforcement), the core families of algorithms, and when to apply each. A useful framing piece for someone entering the field.

MIT Can Predict How Many Retweets You'll Get

Wired's coverage of MIT research predicting retweet counts from tweet content and network features. An early demonstration that social network propagation could be modeled predictively, with implications for understanding how information spreads.

First Targeted Ads, Now Data Scientists Think They Can Change the World

GigaOm piece on a wave of data scientists pivoting from ad targeting to social good applications — healthcare, education, poverty prediction. A 2013 snapshot of the idealism that accompanied the data science boom and the question of whether these techniques could address harder problems.

A Very Short History of Data Science

Gil Press's Forbes timeline tracing data science from John Tukey's 1962 'The Future of Data Analysis' through the 2000s emergence of the term. A useful historical grounding for understanding why 'data science' coalesced as a separate discipline rather than just being statistics.

How to Get Hilary Mason to Build Your Recommender for Free

Mortar Data's post on building a free recommender system using Hilary Mason's approach — a practical guide to collaborative filtering on Hadoop using Mahout. A snapshot of the state of accessible recommendation infrastructure in 2013.

Career Advice: How Do I Become a Data Scientist?

Quora's canonical 2013 answer on becoming a data scientist — one of the most-read career guides in the early data science field. A snapshot of what skills and background paths were considered credible entry points before formal data science degrees existed.

Machine Learning and Link Spam: My Brush With Insanity

A practitioner's account of applying machine learning to link spam detection at SEOMoz — the messy reality of building a classifier with noisy labels, adversarial inputs, and shifting distributions. An honest account of ML in production before MLOps was a field.

I'm a Bandit — Sébastien Bubeck's Optimization Blog

I'm a Bandit is Sébastien Bubeck's research blog on optimization and online learning — covering convex optimization, bandit algorithms, and randomized methods at a level between textbook and paper. One of the best technical blogs in theoretical ML/optimization.

Don Turnbull's Data Science Book Recommendations

DataScholars post on Don Turnbull's data science book recommendations — a 2013 curated reading list for practitioners entering the field. A period artifact showing what an experienced data scientist thought was essential reading at the height of the big data hype cycle.

Sarkar — Graph Processing Paper (CMU AutonLab)

Prithwish Sarkar's paper from CMU's AutonLab on large-scale graph analysis — bookmarked alongside Apache Giraph as a reference for distributed graph processing. CMU's AutonLab works on scalable machine learning for graph-structured data.

Locality-Sensitive Hashing

Locality-sensitive hashing (LSH) is a family of algorithms for approximate nearest-neighbor search — hashing high-dimensional vectors so that similar items hash to the same bucket with high probability. The practical solution to similarity search at scale when exact methods are too slow.

Big Data and the Topologist

Low Dimensional Topology blog post on what topologists bring to big data — specifically topological data analysis (TDA) and the Mapper algorithm for finding shape in high-dimensional datasets. An unusual perspective on why geometric intuition matters for data analysis.

Introduction to Machine Learning — Alex Smola, CMU 2013

Alex Smola's Introduction to Machine Learning course at CMU (10-701, 2013) — a graduate-level survey of ML theory and methods from one of the field's top researchers. Course materials publicly available, covering optimization, probabilistic models, and learning theory.

A Road to Smarter Payments with Machine Learning and Predictive Analytics

Payments Views on applying machine learning to payment processing — primarily fraud detection, risk scoring, and authorization optimization. An early 2013 take on what would become a core fintech ML use case.

Natural Language Processing for the Working Programmer

Natural Language Processing for the Working Programmer — a free online book teaching NLP concepts using Haskell. Unusual choice of language for NLP education; valuable for functional programming practitioners interested in text processing.

Data-Visualization Firm's New Software Autonomously Finds Abstract Connections

Wired's profile of Ayasdi and its Iris software — a topological data analysis platform that finds abstract structure in high-dimensional data without the analyst specifying what to look for. Raises the question of whether a machine can find connections a human wouldn't have thought to look for.

Ayasdi: Automatic Insight Discovery

Ayasdi's homepage — the Stanford TDA spinout offering automatic pattern discovery in high-dimensional data. Saved alongside the Wired article about their Iris platform.

UC Berkeley Course: Analyzing Big Data with Twitter

UC Berkeley's I290 course on analyzing big data with Twitter — lecture videos posted publicly covering Hadoop, NLP, streaming analytics, and social network analysis using Twitter's data firehose. One of the first openly-published university courses on the emerging data science field.

Social Machine Learning — Cambridge Computer Lab Slides

Slides from a Cambridge Computer Lab course on social machine learning — applying ML to social network data, link prediction, community detection, and behavior modeling. A 2010-era academic reference captured at the boundary between network science and machine learning.

Hidden Markov Models on Hadoop — Isabel Drost

Isabel Drost's slides on Hidden Markov Models and Hadoop — covering how HMMs can be implemented and trained at scale using MapReduce. A 2012-era reference for scaling sequence models before deep learning displaced them.

Recommender Systems

Wikipedia's overview of recommender systems — collaborative filtering, content-based filtering, and hybrid approaches. Saved in 2012 when Netflix Prize aftermath and the rise of personalization made recommendation algorithms a hot research area.

Kittydar: JavaScript Cat Face Detection

Kittydar is a JavaScript library for cat face detection in images — a real implementation of neural-network-based object detection in the browser, released in 2012 when running ML in JavaScript was novel and this kind of project showed what was becoming possible.

High Dimensional Undirected Graphical Models

Larry Wasserman's 'Normal Deviate' blog post on high-dimensional undirected graphical models — a rigorous but accessible treatment of the statistical theory for learning sparse graphs from data when the number of variables exceeds the sample size. Core to modern genomics and causal inference.

Machine Learning: Genetic Algorithms in JavaScript

Burak Kanber's accessible tutorial implementing genetic algorithms in JavaScript from scratch — part of a machine learning series aimed at web developers. One of the early 'ML for JS devs' resources that predated the current wave of browser-based ML by several years.

So You Call Yourself a Data Scientist?

VentureBeat's 2012 examination of what 'data scientist' actually means — a job title proliferating faster than the field had consensus on its definition. Published at the height of the 'sexiest job of the 21st century' hype cycle.

An Algorithm to Pick Startup Winners

MIT Technology Review's 2012 piece on early attempts to predict startup success algorithmically — using team composition, funding history, and network signals. The prediction accuracy was modest; the methodological interest was in what features correlated with outcomes.

Great Machine Learning Products

O'Reilly Radar's 2012 analysis of what distinguishes great machine learning products from mediocre ones — written during the pre-deep-learning ML era when practitioners were figuring out how to ship ML-powered features users would actually trust and use.