Fantastic Multi-Task Gradient Updates and How to Find Them In a Cone
This addresses the challenge of gradient conflicts in multi-task learning for AI practitioners, representing an incremental improvement over existing dynamic gradient methods.
The paper tackles the problem of balancing competing objectives in multi-task learning due to conflicting gradients by proposing ConicGrad, a method that dynamically regulates gradient update directions within a cone, achieving state-of-the-art performance on standard benchmarks.
Balancing competing objectives remains a fundamental challenge in multi-task learning (MTL), primarily due to conflicting gradients across individual tasks. A common solution relies on computing a dynamic gradient update vector that balances competing tasks as optimization progresses. Building on this idea, we propose ConicGrad, a principled, scalable, and robust MTL approach formulated as a constrained optimization problem. Our method introduces an angular constraint to dynamically regulate gradient update directions, confining them within a cone centered on the reference gradient of the overall objective. By balancing task-specific gradients without over-constraining their direction or magnitude, ConicGrad effectively resolves inter-task gradient conflicts. Moreover, our framework ensures computational efficiency and scalability to high-dimensional parameter spaces. We conduct extensive experiments on standard supervised learning and reinforcement learning MTL benchmarks, and demonstrate that ConicGrad achieves state-of-the-art performance across diverse tasks.