Machine Learning Resources
A curated guide to understanding models, training, and inference from first principles.
Where to Start
Three paths depending on how you learn. Pick one and follow it before branching out — depth beats breadth early on.
If you learn by doing
1
fast.ai Practical Deep Learning — train real models on day one, fill in theory as you go. Then add Kaggle competitions for practice datasets.
2
TensorFlow Playground — adjust networks in real time to build intuition about depth, width, and activation functions.
3
Andrej Karpathy: Zero to Hero — build a neural network from scratch in Python. Bridges doing and understanding.
If you prefer structured theory first
1
3Blue1Brown Neural Networks — watch all four videos. Best visual intuition for how networks and backprop actually work.
2
Andrew Ng's ML Specialization — systematic coverage of supervised learning, cost functions, regularization, and neural networks.
3
Deep Learning book (Goodfellow) — read alongside the course for rigorous foundations. Free online.
If you want applied engineering fast
1
Hands-On ML with Scikit-Learn & TensorFlow — covers the full pipeline from data prep to deployment with working code.
2
Hugging Face courses — transformers, diffusion models, NLP — the models you'll actually deploy. Interactive notebooks throughout.
3
Kaggle competitions — solve real problems with messy data, read winning solutions. The fastest way to calibrate your skills.
Courses & Videos
Four beautifully animated videos covering how neural networks work, what gradient descent is actually doing, and how backpropagation flows. Essential before anything else — 40 minutes total.
Eight-part video series building everything from scratch: micrograd (autograd engine), makemore (bigram model), all the way to GPT-2. If you only pick one hands-on series, pick this one.
Andrew Ng — Machine Learning Specialization Free to audit
Three-course sequence covering supervised learning, neural networks, decision trees, and unsupervised learning. The gold-standard introduction — methodical, clear, and well-paced.
Top-down approach: you fine-tune a state-of-the-art image classifier in lesson one, then work backward to understand why it works. Unusually effective for people who learn by building.
Full graduate-level course: linear models, SVMs, learning theory, probabilistic models, and RL. More rigorous than Ng's Coursera series — best for building theoretical depth.
Compact introduction with interactive exercises. Good for getting oriented quickly before committing to a full course. Covers linear regression through neural nets in a few hours.
Courses on NLP, transformers, diffusion models, and RL — directly tied to the tools most practitioners use today. Hands-on notebooks throughout, practical and up to date.
Books
Free online
The authoritative technical reference. Covers linear algebra, probability, optimization, and then the full stack of deep learning methods. Dense but definitive — best used alongside a course rather than read start to finish.
Modern two-volume treatment of ML from a probabilistic perspective, with accompanying JAX/PyTorch code notebooks. More current than Bishop's PRML and available free online.
Interactive textbook with runnable code in PyTorch, JAX, and TensorFlow. Covers CNNs, RNNs, attention, transformers, and more — each concept paired immediately with implementation.
Widely recommended (paid)
Hands-On Machine Learning with Scikit-Learn, Keras & TensorFlow — Aurélien Géron
The best practical guide: covers the full pipeline from data cleaning and feature engineering through training neural networks and deploying to production. Code-heavy, example-driven.
Pattern Recognition and Machine Learning — Christopher Bishop
Bayesian foundations of ML. Slower going than Géron but invaluable for understanding why models are designed the way they are. Pairs well with Murphy's probabilistic books above.
Interactive Tools & Playgrounds
Learn by experimenting. These tools let you manipulate networks and training parameters directly and see the effects in real time.
TensorFlow Playground Browser
Visualize a neural network training in real time. Adjust depth, width, learning rate, and activation functions and watch the decision boundary form. Hands-down the fastest way to build intuition.
Teachable Machine — Google Browser
Train an image, audio, or pose classifier in minutes using your webcam or uploaded files — no code required. Export as TensorFlow.js for deployment. Good for demystifying the training loop.
Kaggle Learn Free
Micro-courses (intro ML, intermediate ML, feature engineering, NLP) each ending with a hands-on competition. Fastest way to practice on real-world messy datasets.
Google Colab Free GPU
Free cloud Jupyter notebooks with GPU access. The standard environment for running ML experiments without local setup. Most tutorials and courses post their code as Colab notebooks.
Hugging Face Spaces Browser
Thousands of live model demos: image generation, text classification, translation, speech recognition — run state-of-the-art models directly in your browser to understand what they can and can't do.
Papers & Key Reading
The landmark papers, with accessible guides to each. Reading the original papers builds intuition that courses alone don't give you.
The foundation of GPT, BERT, and every large language model in use today. Read the paper alongside Jay Alammar's illustrated guide which walks through every component visually.
Introduces word embeddings — the idea that you can represent meaning as vectors in high-dimensional space. Foundational for understanding how all modern NLP models represent language.
How can we understand what a neural network has learned? This interactive article visualizes what individual neurons respond to inside image classifiers. Read anything on distill.pub.
The clearest explanation of recurrent networks and LSTMs ever written. Even if RNNs are largely superseded by transformers, this post builds intuition about sequence modeling and gates that carries forward.
A celebrated blog post demonstrating what character-level RNNs can generate — from Shakespeare to Linux source code. Motivating and approachable even as a first read.
Communities & Blogs
A journal dedicated to clear, visual explanations of ML research. Every article is interactive. The closest thing to what a good ML textbook should be — unfortunately on hiatus but the archive is invaluable.
Christopher Olah's essays on neural network interpretability, attention, and visualization. Sets the standard for how to explain ML concepts. Read the LSTM post and the visual introduction to attention.
Deep, well-researched survey posts on diffusion models, policy gradients, attention mechanisms, and more. Weng is an OpenAI researcher — these posts are how ML practitioners get up to speed on new areas.
Competitions, public notebooks, and a community of practitioners. The winning solutions for past competitions — particularly the "writeups" — are some of the best applied ML reading available.
Every paper linked to its implementation. State-of-the-art benchmarks across tasks. Useful for understanding what the current frontier looks like and finding code to run new architectures.