heatdrop.ai

DIPY has spent a decade turning raw diffusion MRI data into maps of neural pathways that researchers actually argue about.

An educational reinforcement learning project where an agent learns to buy and sell a single stock through rewards and penalties, not explicit instructions.

A 2016 tutorial repo that still runs on TensorFlow 2.x, comparing how feedforward, convolutional, and recurrent networks handle city noise.

Matminer collects scattered materials-science datasets and featurizers into one library so researchers can stop writing the same data-prep scripts.

A research prototype that squeezes road segmentation, car detection, and street classification into a single real-time encoder-decoder model.

Phillip is a Super Smash Bros. Melee AI trained with deep reinforcement learning to brawl in Dolphin emulator—though its creator has since moved on to imitation learning.

A readable reimplementation of guided policy search for robotic control, aimed at researchers who want to understand the machinery rather than just run it.

Irwin is the anti-cheat engine that learns how cheaters play by grinding through games with Stockfish and TensorFlow.

A 2017 experiment that treats price charts as images and asks AlexNet to guess which way the candlestick winds.

An older TensorLayer reference implementation for BRATS brain tumor segmentation, with some honest notes about its own rough edges.

A 2016 Tel Aviv University project that swaps Tao Feng's RBM for a TensorFlow CNN and scrapes 30-second previews to classify ten music genres.

A 2016-era TensorFlow project that treats OHLCV price data as a 1D "image" for CNN pattern recognition.

A faithful TensorFlow reimplementation of a 2015 paper that uses recurrent neural networks to predict multiple medical conditions from messy, irregular vital-sign sequences.

A teaching-oriented AI bot that abstracts away the differences between Brood War and SC2 so you can focus on strategy, not API archaeology.

A PyTorch A3C implementation that keeps per-agent networks on GPU but shoves updates to a CPU-shared model via Hogwild, trading elegance for raw speed.

A PyTorch DOOM bot that won the ViZDoom AI Competition by learning to frag with deep reinforcement learning.

A 2017 RL framework that treats multi-hop KG reasoning as pathfinding with embedding-based states and a reward function that cares about accuracy, diversity, and efficiency—not just getting there.

ETH Zurich master's thesis project that separates moving objects from static background in real-time 3D reconstruction.

A hands-on Python comparison of Q-learning and Sarsa for grid-world path planning, complete with commented code and experimental charts.

An old-school Keras notebook that detects credit card fraud by training autoencoders only on normal transactions, then punishing anything that looks weird.