CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning. This changelog does not include internal changes that do not affect the user.

[Unreleased]

Added

• Added usage example showing how to combine TorchJD with automatic mixed precision (AMP).

Changed

• Refactored the underlying optimization problem that UPGrad and DualProj have to solve to project onto the dual cone. This may minimally affect the output of these aggregators.
• Refactored internal verifications in the autojac engine so that they do not run at runtime anymore. This should minimally improve the performance and reduce the memory usage of backward and mtl_backward.

Fixed

• Fixed the behavior of backward and mtl_backward when some tensors are repeated (i.e. when they appear several times in a list of tensors provided as argument). Instead of raising an exception in these cases, we are now aligned with the behavior of torch.autograd.backward. Repeated tensors that we differentiate lead to repeated rows in the Jacobian, prior to aggregation, and repeated tensors with respect to which we differentiate count only once.
• Removed arbitrary exception handling in IMTLG and AlignedMTL when the computation fails. In practice, this fix should only affect some matrices with extremely large values, which should not usually happen.

[0.5.0] - 2025-02-01

Added

• Added new aggregator ConFIG from ConFIG: Towards Conflict-free Training of Physics Informed Neural Networks.

[0.4.2] - 2025-01-30

Added

• Added Python 3.13 classifier in pyproject.toml (we now also run tests on Python 3.13 in the CI).

[0.4.1] - 2025-01-02

Fixed

• Fixed a bug introduced in v0.4.0 that could cause backward and mtl_backward to fail with some tensor shapes.

[0.4.0] - 2025-01-02 [YANKED]

Changed

• Changed how the Jacobians are computed when calling backward or mtl_backward with parallel_chunk_size=1 to not rely on torch.autograd.vmap in this case. Whenever vmap does not support something (compiled functions, RNN on cuda, etc.), users should now be able to avoid using vmap by calling backward or mtl_backward with parallel_chunk_size=1.

• Changed the effect of the parameter retain_graph of backward and mtl_backward. When set to False, it now frees the graph only after all gradients have been computed. In most cases, users should now leave the default value retain_graph=False, no matter what the value of parallel_chunk_size is. This will reduce the memory overhead.

Added

• RNN training usage example in the documentation.

[0.3.1] - 2024-12-21

Changed

• Improved the performance of the graph traversal function called by backward and mtl_backward to find the tensors with respect to which differentiation should be done. It now visits every node at most once.

[0.3.0] - 2024-12-10

Added

• Added a default value to the inputs parameter of backward. If not provided, the inputs will default to all leaf tensors that were used to compute the tensors parameter. This is in line with the behavior of torch.autograd.backward.
• Added a default value to the shared_params and to the tasks_params arguments of mtl_backward. If not provided, the shared_params will default to all leaf tensors that were used to compute the features, and the tasks_params will default to all leaf tensors that were used to compute each of the losses, excluding those used to compute the features.
• Note in the documentation about the incompatibility of backward and mtl_backward with tensors that retain grad.

Changed

• BREAKING: Changed the name of the parameter A to aggregator in backward and mtl_backward.
• BREAKING: Changed the order of the parameters of backward and mtl_backward to make it possible to have a default value for inputs and for shared_params and tasks_params, respectively. Usages of backward and mtl_backward that rely on the order between arguments must be updated.
• Switched to the PEP 735 dependency groups format in pyproject.toml (from a [tool.pdm.dev-dependencies] to a [dependency-groups] section). This should only affect development dependencies.

Fixed

• BREAKING: Added a check in mtl_backward to ensure that tasks_params and shared_params have no overlap. Previously, the behavior in this scenario was quite arbitrary.

[0.2.2] - 2024-11-11

Added

• PyTorch Lightning integration example.
• Explanation about Jacobian descent in the README.

Fixed

• Made the dependency on ecos explicit in pyproject.toml (before cvxpy 1.16.0, it was installed automatically when installing cvxpy).

[0.2.1] - 2024-09-17

Changed

• Removed upper cap on numpy version in the dependencies. This makes torchjd compatible with the most recent numpy versions too.

Fixed

• Prevented IMTLG from dividing by zero during its weight rescaling step. If the input matrix consists only of zeros, it will now return a vector of zeros instead of a vector of nan.

[0.2.0] - 2024-09-05

Added

• autojac package containing the backward pass functions and their dependencies.
• mtl_backward function to make a backward pass for multi-task learning.
• Multi-task learning example.

Changed

• BREAKING: Moved the backward module to the autojac package. Some imports may have to be adapted.
• Improved documentation of backward.

Fixed

• Fixed wrong tensor device with IMTLG in some rare cases.
• BREAKING: Removed the possibility of populating the .grad field of a tensor that does not expect it when calling backward. If an input t provided to backward does not satisfy t.requires_grad and (t.is_leaf or t.retains_grad), an error is now raised.
• BREAKING: When using backward, aggregations are now accumulated into the .grad fields of the inputs rather than replacing those fields if they already existed. This is in line with the behavior of torch.autograd.backward.

[0.1.0] - 2024-06-22

Added

• Basic project structure.
• aggregation package:
  • Aggregator base class to aggregate Jacobian matrices.
  • AlignedMTL from Independent Component Alignment for Multi-Task Learning.
  • CAGrad from Conflict-Averse Gradient Descent for Multi-task Learning.
  • Constant to aggregate with constant weights.
  • DualProj adapted from Gradient Episodic Memory for Continual Learning.
  • GradDrop from Just Pick a Sign: Optimizing Deep Multitask Models with Gradient Sign Dropout.
  • IMTLG from Towards Impartial Multi-task Learning.
  • Krum from Machine Learning with Adversaries: Byzantine Tolerant Gradient Descent.
  • Mean to average the rows of the matrix.
  • MGDA from Multiple-gradient descent algorithm (MGDA) for multiobjective optimization.
  • NashMTL from Multi-Task Learning as a Bargaining Game.
  • PCGrad from Gradient Surgery for Multi-Task Learning.
  • Random from Reasonable Effectiveness of Random Weighting: A Litmus Test for Multi-Task Learning.
  • Sum to sum the rows of the matrix.
  • TrimmedMean from Byzantine-Robust Distributed Learning: Towards Optimal Statistical Rates.
  • UPGrad from Jacobian Descent for Multi-Objective Optimization.
• backward function to perform a step of Jacobian descent.
• Documentation of the public API and of some usage examples.
• Tests:
  • Unit tests.
  • Documentation tests.
  • Plotting utilities to verify qualitatively that aggregators work as expected.