Some resources for ML research

Personal and biased selection of ML resources.

Disclaimer: I'm a noivce in ML research, and I read only a few of the list.

Table of Contents

• Beginner's Guide
• Machine Learning
• Deep Learning
• Reinforcement Learning
• Graphical Model
• Optimization
• Learning Theory
• Statistics
• Topics in Machine Learning
• Math Backgrounds
• Blogs

Beginner's Guide

Must Read

• Machine Learning: A Probabilistic Perspective (Murphy)
• Deep Learning (Goodfellow et al.)
• Reinforcement Learning: An Introduction (Sutton & Barto)

Recommended

• Convex Optimization (Boyd & Vandenberghe)
• Graphical Models, Exponential Families, and Variational Inference (Wainwright & Jordan)
• Learning from Data (Abu-Mostafa) -> for whom interested in learning theory

Machine Learning

There are many ML books, but most of them are encyclopedic.
I recommend to take a course using Murphy or Bishop book.

Textbook

• Machine Learning: A Probabilistic Perspective (Murphy) ✨
• Pattern Recognition and Machine Learning (Bishop) ✨
• The Elements of Statistical Learning (Hastie et al.)
• Pattern Classification (Duda et al.)
• Bayesian Reasoning and Machine Learning (Barber)

Lecture

• Stanford CS229 Machine Learning ✨
• CMU 10701 Mahine Learning
• Oxford Machine Learning

Deep Learning

Goodfellow et al. is the new classic.
For vision and NLP, Stanford lectures would be helpful.

Textbook

• Deep Learning (Goodfellow et al.) ✨

Lecture (Practice)

• Stanford CS231n Convolutional Neural Networks for Visual Recognition ✨
• Stanfrod CS224d Deep Learning for Natural Language Processing
• Oxford Deep Natural Language Processing

Lecture (Theory)

• Stanford Stat385 Theories of Deep Learning

Tutorial

• ICML 2017 Sequence-To-Sequence Modeling with Neural Networks

Reinforcement Learning

For classic (non-deep) RL, Sutton & Barto is the classic.
For deep RL, lectures from Berkeley/CMU looks good.

Textbook

• Reinforcement Learning: An Introduction (Sutton & Barto) ✨

Lecture

• UCL Reinforcement Learning ✨
• Berkeley CS294 Deep Reinforcement Leanring ✨
• CMU 10703 Deep Reinforcement Learing and Control

Tutorial

• ICML 2017 Deep Reinforcement Learning, Decision Making, and Control

Graphical Model

Koller & Friedman is comprehensive, but too encyclopedic.
I recommend to take an introductory course using Koller & Friedman book.

Wainwright & Jordan only focuses on variational inference,
but it gives really good intuition for probabilistic models.

Textbook

• Probabilistic Graphical Models: Principles and Techniques (Koller & Friedman)
• Graphical Models, Exponential Families, and Variational Inference (Wainwright & Jordan) ✨

Lecture

• Stanford CS228 Probabilistic Graphical Models
• CMU 10708 Probabilistic Graphical Models ✨

Optimization

Boyd & Vandenberghe is the classic, but I think it's too boring.
Reading chapter 2-5 would be enough.

Bertsekas more concentrates on convex analysis.
Nocedal & Wright more concentrates on optimization.

Textbook

• Convex Optimization (Boyd & Vandenberghe) ✨
• Convex Optimization Theory (Bertsekas)
• Numerical Optimization (Nocedal & Wright)

Lecture

• Stanford EE364a Convex Optimization I ✨
• Stanford EE364b Convex Optimization II
• MIT 6.253 Convex Analysis and Optimization

Tutorial

• ICML 2017 Recent Advances in Stochastic Convex and Non-Convex Optimization

Learning Theory

In my understanding, there are two major topics in learning theory:

• Learning Theory: VC-dimension, PAC-learning
• Online Learning: regret bound, multi-armed bandit

For learning theory, Kearns & Vazirani is the classic; but it's too old-fashined.
Abu-Mostafa is a good introductory book, and I think it's enough for most people.

For online learning, Cesa-Bianchi & Lugosi is the classic.
For multi-armed bandit, Bubeck & Cesa-Bianchi provides a good survey.

Textbook (Learning Theory)

• Learning from Data (Abu-Mostafa) ✨
• Foundations of Machine Learning (Mohri et al.)
• An Introduction to Computational Learning Theory (Kearns & Vazirani)

Textbook (Online Learning)

• Prediction, Learning, and Games (Cesa-Bianchi & Lugosi)
• Regret Analysis of Stochastic and Nonstochastic Multi-armed Bandit Problems (Bubeck & Cesa-Bianchi)

Lecture

• Caltech Learning from Data ✨
• CMU 15859 Machine Learning Theory
• Berkeley CS281b/Stat241b Statistical Learning Theory
• MIT 9.520 Statistical Learning Theory and Applications

Tutorial

• ICML 2017 Real World Interactive Learning

Statistics

Statistics is a broad area; hence, I listed only a few of them.
For advanced topics, lectures from Berkeley/Stanford/CMU/MIT looks really cool.

Textbook (Statistical Inference)

• All of Statistics (Wasserman)
• Computer Age Statistical Inference (Efron & Hastie) ✨
• Time Series Analysis and Its Applications: With R Examples (Shumway & Stoffer)

Textbook (Nonparametrics)

• All of Nonparametric Statistics (Wasserman)
• Introduction to Nonparametric Estimation (Tsybakov)
• Gaussian Process and Machine Learning (Rasmussen & Williams) ✨
• Bayesian Nonparametrics (Ghosh & Ramamoorthi) ✨

Textbook (Advanced Topics)

• High-Dimensional Statistics: A Non-Asymptotic Viewpoint (Wainwright) ✨
• Statistics for High-Dimensional Data (Bühlmann & van de Geer)
• Asymptotic Statistics (van der Vaart)
• Empirical Processes in M-Estimation (van der Vaart)

Lecture

• Berkeley Stat210a Theoretical Statistics I
• Berkeley Stat210b Theoretical Statistics II
• Stanford Stat300a Theory of Statistics
• Stanford CS369m Algorithms for Massive Data Set Analysis
• CMU 36755 Advanced Statistical Theory I
• MIT 18.S997 High-Dimensional Statistics

Topics in Machine Learning

Miscellaneous topics related to machine learning.
There are much more subfields, but I'll not list them all.

Information Theory

• Elements of Information Theory (Cover & Thomas)
• Information Theory, Inference, and Learning Algorithms (MacKay)

Network Science

• Networks, Crowds, and Markets (Easley & Kleinberg)
• Social and Economic Networks (Jackson)

Markov Chain

• Markov Chains (Norris)
• Markov Chains and Mixing Times (Levin et al.)

Game Theory

• Algorithmic Game Theory (Nisan et al.)
• Multiagent Systems (Shoham & Leyton-Brown)

Combinatorics

• The Probabilistic Method (Alon & Spencer)
• A First Course in Combinatorial Optimization (Lee)

Algorithm

• Introduction to Algorithms (Cormen et al.)
• Randomized Algorithms (Motwani & Raghavan)
• Approximation Algorithms (Vazirani)

Geometric View

• Topological Data Analysis (Wasserman)
• Methods of Information Geometry (Amari & Nagaoka)
• Algebraic Geometry and Statistical Learning Theory (Watanabe)

Some Lectures

• MIT 18.409 Algorithmic Aspects of Machine Learning
• MIT 18.409 An Algorithmist's Toolkit

Tutorial

• ICML 2017 Interpretable Machine Learning
• ICML 2017 Robustness Meets Algorithms (and Vice-Versa)

Math Backgrounds

I selected essential topics for machine learning.
Personally, I think more analysis / matrix / geometry never hurts.

Probability

• Probability: Theory and Examples (Durrett)
• Theoretical Statistics (Keener)
• Stochastic Processes (Bass)
• Probability and Statistics Cookbook (Vallentin)

Linear Algebra

• Linear Algebra (Hoffman & Kunze)
• Matrix Analysis (Horn & Johnson)
• Matrix Computations (Golub & Van Loan)
• The Matrix Cookbook (Petersen & Pedersen)

Large Deviations

• Concentration Inequalities and Martingale Inequalities (Chung & Lu)
• An Introduction to Matrix Concentration Inequalities (Tropp)

Blogs

TBD
Acknowledge to https://stats385.github.io/blogs.

Research Group

• DeepMind Blog
• OpenAI Blog
• FAIR Blog
• BAIR Blog
• Distill
• Cambridge MLG

Persoanal Blog

• Andrej Karpathy
• Colah
• inFERENCe