From Gradient Volume to Shapley Fairness: Towards Fair Multi-Task Learning

Multi-task learning often suffers from gradient conflicts, leading to task-level unfair optimization and degraded overall performance. To address this, we present SVFair, a Shapley value-based framework for fair gradient aggregation that explicitly targets task-level fairness under such conflicts. Unlike heuristic scalarization or pairwise conflict penalties, SVFair combines a geometric view of gradient interaction with a cooperative-game view of fair contribution. We propose two scalable geometric conflict metrics: VolDet, a gram determinant volume metric, and VolDetPro, its sign-aware extension distinguishing antagonistic gradients. By integrating these metrics into Shapley value computation, SVFair quantifies each task's deviation from the overall gradient and rebalances updates toward fairness. In parallel, our Shapley value computation admits controllable complexity. Extensive experiments show that SVFair achieves state-of-the-art results across diverse supervised and reinforcement learning benchmarks, and further improves existing methods when integrated as a fairness-enhancing module.