In a past life, I used to do a lot of software engineering. I have listed a few open-source projects I started (apart from research codebases). I have presented a few talks at software conferences which are also listed here.
selected software
torch-diffsim
Parallelizable, differentiable, deformable simulation entirely in PyTorch
Simplicits Nerfstudio
Qualitative, speed improvements and a more complete implementation for Simplicits which allows simulating a mesh, Gaussian Splat, or a NeRF.
MIRNet-TFJS
TensorFlow JS models for MIRNet for low-light💡 image enhancement that can run entirely on your browser.
(GitHub Trending)
Fast-Transformer
An optimized implementation of Additive Attention.
(GitHub Trending)
3D Transforms
A library to easily work with 3D data and make 3D transformations.
Gradient-Centralization
Instantly improve your training performance by implementing Gradient Centralization in optimizers.
(GitHub Trending)
Perceiver
An optimized implementation of Perceiver.
(GitHub Trending)
Greenathon
Originally a hackathon submission, shows how to train models specifically for deploying them to run entirely on browsers.
(GitHub Trending)
ISAB
A framework to use Permutation-Invariant Neural Networks.
ML With Android 11
Popular samples for optimized inference for machine learning models on Android using TensorFlow Lite using capabilities introduced in Android 11.
(GitHub Trending)
Face-Recognition Flutter
Popular samples for optimized inference for machine learning models on Android using Flutter and Firebase ML Kit.
TF Watcher
A tool to monitor your ML jobs remotely.
(GitHub Trending)
Nystromformer
An optimized implementation of using Nyström Method for approximation self-attention.
Transformer in Transformer
An optimized implementation of performing attention inside local patches for image classification.
Conformer
One of the first implementations of the popular Conformer.
GLOM
One of the first implementations of the popular Hinton's GLOM with optimization to make it runnable.
GLU
Gated Linear Units and many of their variants.
Compositional Attention
An optimized implementation of Compositional Attention and their variants around Disentangling Search and Retrieval.
Invariant Point Attention
Invariant Point Attention from AlphaFold 2 for all problems.
Wasm FAAS
A proof of concept to run Machine Learning models as serverless functions with Wasm.
conference talks
