Projects

rlkit – RL framework with algorithms implemented in PyTorch: Imagined Goals, TD models, Hindsigh Expirience Replay, Double DQN et al.

Optuna – Bayesian hyperparameter optimisation framework with pruning and parallelisation. Features an imperative and modular define-by-run style API.

SleepWalk – R package for interactive embeddings exploration (e.g., t-SNE, UMAP). Shows color coded "real" distances of all the cells to the one under your cursor.

Nevergrad – Python toolbox for performing gradient-free optimization from Facebook.

Flair – a very simple framework for state-of-the-art NLP by Zalando Tech. Features multilingual NER, PoS, BERT & ELMo embeddings et al. Builds on PyTorch.

code2vec – a neural network for learning distributed representations of code. POPL 2019. Demo.

Research Papers and Books

"Variational Bayesian Monte Carlo" – novel Bayesian inference framework that combines variational inference with active-sampling Bayesian quadrature for models with expensive black-box likelihood. NIPS 2018. Source.

"An Introduction to Deep Reinforcement Learning" – introduction to deep RL models, algorithms and techniques. Particular focus is on generalization and how deep RL can be used for practical applications.[140pp]

"Elimination of All Bad Local Minima in Deep Learning" – proves, without any strong assumption, that adding one neuron per output unit can eliminate all suboptimal local minima for multi-class classification/regression with an arbitrary loss function.

"A Geometric Theory of Higher-Order Automatic Differentiation" – differential geometric treatment of AD with practical high-performance implementations. [55pp].

"Mathematics and Computation" – book draft by Avi Wigderson, Princeton Univ. Press. Introduction to computational complexity theory, its connections and interactions with math & its central role social sciences and technology.

"Visualizing the Loss Landscape of Neural Nets" – explores how network architecture affects the loss landscape and its effect of generalization. NIPS 2018. Source.

"Deep Reinforcement Learning" – the overview draws a big picture, filled with details. Discusses 6 core elements, 6 important mechanisms, and 12 applications, focusing on contemporary work, and in historical contexts. [150pp].

"Automl: Methods, Systems, Challenges" – book draft by Prof. Frank Hutter.

Posts, Articles, Tutorials

Key Papers in Deep RL by OpenAI – list of papers in deep RL that should provide a useful starting point for someone looking to do research in the field.

NLP Overview – an up-to-date learning resource that integrates important information related to NLP research, such as: SoTA, emerging concepts, applications, benchmark, datasets, code etc.

A Full Hardware Guide to Deep Learning. CPU: 1-2 cores per GPU > 2GHz. PCIe: lanes do not matter. RAM: clock rate do not matter. PSU: add up watts of GPUs + CPU. Then multiple by 110% et al.

Open Courses, Textbooks, Lectures and Cheatsheets – curated resources for learning statistics and machine learning.

A Concise Handbook of TensorFlow – online book for those who already knows ML/DL theories and want to focus on learning TensorFlow itself.

Scipy Lecture Notes – one document to learn numerics, science, and data with Python. Quick introduction to central tools and techniques with increasing level of expertise, from beginner to expert.

Video Lectures and Talks

The International Conference on Probabilistic Programming. Talks from the PROBPROG 2018 Conference, held at the MIT Media Lab in Cambridge.

NeurIPS 2018. Online Videos.

Deep Learning and Reinforcement Learning Summer School, Toronto 2018. Hosted by CIFAR and the Vector Institute.

Conference on Learning Theory – COLT 2018 Online Videos.