Sam

hamiltonian-nn. Code for our paper "Hamiltonian Neural Networks"

scribe. Realistic Handwriting with Tensorflow

mnist1d. A 1D analogue of the MNIST dataset for measuring spatial biases and answering Science of Deep Learning questions.

baby-a3c. A high-performance Atari A3C agent in 180 lines of PyTorch

crypto-rnn. Learning the Enigma with Recurrent Neural Networks

visualize_atari. Code for our paper "Visualizing and Understanding Atari Agents" (https://goo.gl/AMAoSc)

pythonic_ocr. A convolutional neural network implemented in pure numpy.

excitationbp. Visualizing how deep networks make decisions

cursivetransformer. Training a transformer to generate cursive handwriting

psi0nn. A neural network quantum ground state solver

dnc. Differentiable Neural Computer in TensorFlow

optimize_wing. We simulate a wind tunnel, place a rectangular occlusion in it, and then use gradient descent to turn the occlusion into a wing.

ncf. Nature's Cost Function (NCF). Finding paths of least action with gradient descent.

studying_growth. Studying Cell Growth with Neural Cellular Automata

structural_optimization. Coding structural optimization, from scratch, in 200 lines of Python

greydanus.github.io. My academic blog

stereograms. Code for playing with random dot stereograms.

mr_london. A LSTM recurrent neural network implemented in pure numpy

mnist-gan. Generative Adversarial Networks for the MNIST dataset

piecewise_node. Temporal abstraction for autoregressive sampling

rlzoo. A central location for my reinforcement learning experiments

subspace-nn. Optimizing neural networks in subspaces

flowchart. A Python script that converts boolean expressions to flowcharts or directed acyclic graphs.

fractal_tree. A numerical model of fractal dynamics

np_nets. Neural network experiments written purely in numpy

compton. Exploring the quantum nature of light with compton scattering

regularization. I use a one-layer neural network trained on the MNIST dataset to give an intuition for how common regularization techniques affect learning.

deep_thesaurus. Use a pretrained NLP model to rank thesaurus suggestions

baselines. Simple MNIST baselines for 1) numpy backprop 2) dense nns 3) cnns 3) seq2seq

karpathy.github.io. my blog

dlfun. Forays into the world of deep learning using TensorFlow

evostrats. A minimal evolution strategies benchmark

lagrangian_nns. Fork of codebase from Miles Cranmer's GitHub

billiards. A simple RL environment for studying planning.

dissipative_hnns. Fork from Sosanya's GitHub of our code for "Dissipative HNNs"