arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2205.03076cs.LGmath.OC
收藏

超越反向传播:通过隐式微分与均衡传播的双层优化

Beyond backpropagation: bilevel optimization through implicit differentiation and equilibrium propagation

ETH Zurich(苏黎世联邦理工学院);University of Zurich(苏黎世大学)

Nicolas Zucchet, João Sacramento

AI总结:

本文综述了基于梯度的双层优化方法,重点介绍隐式微分与均衡传播两类方法的数学基础、梯度估计算法及其竞争优势。

摘要:

本文回顾了用于求解双层优化问题的基于梯度的方法。双层优化是一种通用框架,用于学习那些通过其最小化的量而被隐式定义的系统。这种刻画方式可应用于神经网络、优化器、算法求解器乃至物理系统,并且与对此类系统的显式定义相比,能够提供更大的建模灵活性。本文聚焦于求解此类问题的基于梯度的方法。我们将这些方法区分为两类:一类植根于隐式微分,另一类利用均衡传播定理。我们介绍了这些方法背后的数学基础,详细阐述了梯度估计算法,并比较了不同方法的竞争优势。

英文摘要:

This paper reviews gradient-based techniques to solve bilevel optimization problems. Bilevel optimization is a general way to frame the learning of systems that are implicitly defined through a quantity that they minimize. This characterization can be applied to neural networks, optimizers, algorithmic solvers and even physical systems, and allows for greater modeling flexibility compared to an explicit definition of such systems. Here we focus on gradient-based approaches that solve such problems. We distinguish them in two categories: those rooted in implicit differentiation, and those that leverage the equilibrium propagation theorem. We present the mathematical foundations that are behind such methods, introduce the gradient-estimation algorithms in detail and compare the competitive advantages of the different approaches.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑