Library for Jacobian descent with PyTorch. It enables the optimization of neural networks with multiple losses (e.g. multi-task learning).

Repository for hosting the built documentation of torchjd

This repo is used to compare different torchjd functions (e.g. aggregators), potentially from different commits of torchjd, on a variety of topics (computation time, precision, etc.)

Optimization of simple functions with JD and plot of the trajectories.