Mitigating Gradient Bias in Multi-objective Learning: A Provably Convergent Approach
Mitigating Gradient Bias in Multi-objective Learning: A Provably Convergent Approach
复制标题
DOI：
--
复制
发表时间：
2023
期刊：
影响因子：
--
通讯作者：
H. Fernando;Han Shen;Miao Liu;Subhajit Chaudhury;K. Murugesan;Tianyi Chen
中科院分区：
文献类型：
--
作者：
H. Fernando;Han Shen;Miao Liu;Subhajit Chaudhury;K. Murugesan;Tianyi Chen
具有多个目标的机器学习问题要么出现在具有多个标准的学习中，其中学习必须在公平、安全和准确性等多个性能指标之间进行权衡；要么出现在多任务学习中，其中多个任务被联合优化，其中多个任务之间存在归纳偏差。这些多目标学习问题通常由多目标优化框架来解决。然而，现有的随机多目标梯度方法及其最近的变种(如MGDA、PCGrad、CAGrad等)。所有这些都采用了有偏差的梯度方向，这导致了经验绩效的下降。为此，我们提出了一种求解多目标优化问题的随机多目标梯度校正方法。我们的方法的独特之处在于，即使在非凸的情况下，它也可以在不增加批处理大小的情况下保证收敛。在监督学习和强化学习上的仿真实验证明了该方法相对于最新方法的有效性。
Machine learning problems with multiple objectives appear either i) in learning with multiple criteria where learning has to make a trade-off between multiple performance metrics such as fairness, safety and accuracy; or, ii) in multi-task learning where multiple tasks are optimized jointly, sharing inductive bias among them. These multiple-objective learning problems are often tackled by the multi-objective optimization framework. However, existing stochastic multi-objective gradient methods and their recent variants (e.g., MGDA, PCGrad, CAGrad, etc.) all adopt a biased gradient direction, which leads to degraded empirical performance. To this end, we develop a stochastic multi-objective gradient correction (MoCo) method for multi-objective optimization. The unique feature of our method is that it can guarantee convergence without increasing the batch size even in the nonconvex setting. Simulations on supervised and reinforcement learning demonstrate the effectiveness of our method relative to state-of-the-art methods.