A Unified Framework for Gradient Aggregation in Multi-Objective Optimization

In three linesUnified framework for gradient aggregation in multi-objective optimization. Authors establish convergence rates to Pareto stationarity via sufficient alignment condition, showing non-conflicting directions within gradient convex hull ensure convergence. Introduces capped MGDA from CVaR formulation, validated on synthetic and practical benchmarks.

Summary generated by Claude — human-verified