arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.09569cs.LGphysics.app-ph
收藏

利用局部物理信息瓶颈训练深度物理神经网络

Training deep physical neural networks with local physical information bottleneck

Department of Precision Instrument, Tsinghua University(精密仪器系,清华大学);Laboratoire Kastler Brossel, École Normale Supérieure - Paris Sciences et Lettres (PSL) Research University, Sorbonne Université, Centre National de la Recherche Scientifique (CNRS), UMR 8552, Collège de France(Kastler Brossel实验室,巴黎科学与文学研究大学(PSL)研究大学,索邦大学,法国国家科学研究中心(CNRS),UMR 8552,法国学院);School of Integrated Circuits, Beijing Advanced Innovation Center for Integrated Circuits, BNRist, Tsinghua University(集成电路学院,北京集成电路先进创新中心,BNRist,清华大学);Department of Electrical and Electronic Engineering, The University of Hong Kong(电子与电气工程系,香港大学);State Key Laboratory of Precision Space-time Information Sensing Technology, Department of Precision Instrument, Tsinghua University(精密时空信息感知技术国家重点实验室,精密仪器系,清华大学);Key Laboratory of Photonic Control Technology, Ministry of Education, Tsinghua University(光电控制技术重点实验室,教育部,清华大学);Department of Electrical Engineering, City University of Hong Kong(电气工程系,城市大学)

Hao Wang, Ziao Wang, Xiangpeng Liang, Han Zhao, Jianqi Hu, Junjie Jiang, Xing Fu, Jianshi Tang, Huaqiang Wu, Sylvain Gigan, Qiang Liu

AI总结:

本文提出物理信息瓶颈框架,通过整合信息论与局部学习,实现深度物理神经网络在任意物理动态下的高效训练,适用于多种物理基质。

摘要:

深度学习已彻底改变了现代社会,但面临日益严峻的能量和延迟限制。深度物理神经网络(PNNs)是相互连接的计算系统,直接利用类比动力学实现高效、超快的AI执行。然而,要实现这一潜力,需要通用的训练方法,以适应物理的复杂性。本文提出了物理信息瓶颈(PIB),一种通用且高效的框架,将信息论与局部学习相结合,使深度PNNs能够在任意物理动态下学习。通过将矩阵基于的信息瓶颈分配给每个单元,我们展示了在电子忆阻芯片和光计算平台上,监督、无监督和强化学习的实现。PIB还能够适应严重的硬件故障,并通过地理分布的资源实现并行训练。通过绕过辅助数字模型和对比测量,PIB将PNN训练重新表述为内在、可扩展的信息论过程,与多种物理基质兼容。

英文摘要:

Deep learning has revolutionized modern society but faces growing energy and latency constraints. Deep physical neural networks (PNNs) are interconnected computing systems that directly exploit analog dynamics for energy-efficient, ultrafast AI execution. Realizing this potential, however, requires universal training methods tailored to physical intricacies. Here, we present the Physical Information Bottleneck (PIB), a general and efficient framework that integrates information theory and local learning, enabling deep PNNs to learn under arbitrary physical dynamics. By allocating matrix-based information bottlenecks to each unit, we demonstrate supervised, unsupervised, and reinforcement learning across electronic memristive chips and optical computing platforms. PIB also adapts to severe hardware faults and allows for parallel training via geographically distributed resources. Bypassing auxiliary digital models and contrastive measurements, PIB recasts PNN training as an intrinsic, scalable information-theoretic process compatible with diverse physical substrates.

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。

↑