Gradient Surgery for Multi-Task Learning

TLDR

Deep learning has achieved impressive results in many domains, yet data efficiency remains a major challenge, and while multi‑task learning promises shared structure to improve efficiency, its optimization difficulties are not fully understood. This study identifies three conditions in the multi‑task optimization landscape that cause detrimental gradient interference and proposes a general method to avoid such interference. The method performs gradient surgery by projecting each task’s gradient onto the normal plane of any conflicting task’s gradient. Across challenging multi‑task supervised and reinforcement learning tasks, the approach yields substantial efficiency and performance gains, is model‑agnostic, and can be combined with existing multi‑task architectures.

Abstract

While deep learning and deep reinforcement learning (RL) systems have demonstrated impressive results in domains such as image classification, game playing, and robotic control, data efficiency remains a major challenge. Multi-task learning has emerged as a promising approach for sharing structure across multiple tasks to enable more efficient learning. However, the multi-task setting presents a number of optimization challenges, making it difficult to realize large efficiency gains compared to learning tasks independently. The reasons why multi-task learning is so challenging compared to single-task learning are not fully understood. In this work, we identify a set of three conditions of the multi-task optimization landscape that cause detrimental gradient interference, and develop a simple yet general approach for avoiding such interference between task gradients. We propose a form of gradient surgery that projects a task's gradient onto the normal plane of the gradient of any other task that has a conflicting gradient. On a series of challenging multi-task supervised and multi-task RL problems, this approach leads to substantial gains in efficiency and performance. Further, it is model-agnostic and can be combined with previously-proposed multi-task architectures for enhanced performance.