v0.13.0
🥝 MoDo, STCH, GeometricMean
This release introduces:
- MoDo, a weighting method from Three-Way Trade-Off in Multi-Objective Learning: Optimization, Generalization and Conflict-Avoidance. Thanks a lot to @KhusPatel4450 for the contribution!
- STCH, a scalarizer from Smooth Tchebycheff Scalarization for Multi-Objective Optimization. Thanks a lot to @ppraneth for the contribution!
- GeometricMean, a scalarizer from MultiNet++: Multi-Stream Feature Aggregation and Geometric Loss Strategy for Multi-Task Learning. Thanks a lot to @ppraneth for the contribution!
Changelog
Added
- Added
STCHfrom Smooth Tchebycheff Scalarization for Multi-Objective
Optimization, aScalarizerthat combines the input
tensor of values into a smooth approximation of their (weighted, shifted) maximum. - Added
MoDoWeightingfrom Three-Way Trade-Off in Multi-Objective Learning: Optimization, Generalization and Conflict-Avoidance (JMLR 2024). It is a statefulWeightingthat maintains task weights across calls via a simplex-projected gradient step on a cross-batch matrixG = J_1 @ J_2.T, computed from two independent mini-batches usingautojac.jac. - Added
GeometricMean(also known as GLS) studied in MultiNet++: Multi-Stream Feature
Aggregation and Geometric Loss Strategy for Multi-Task
Learning,
aScalarizerthat returns the geometric mean of the input tensor of values.