game-theoretic multi-task loss balancing

transform

List<LossTensor> -> ScaledLossTensor

Balance conflicting gradients and magnitudes from multiple task losses dynamically using Nash bargaining or Pareto optimization.

Problem it solves

Multi-task models often suffer from one task's loss dominating or degrading the performance of other tasks during training.

Consumes

List<LossTensor>

Emits

ScaledLossTensor

Distilled from 1 source

The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.