Blending Optimal Control and Biologically Plausible Learning for Noise-Robust Physical Neural Networks
arXiv:2502.19053 · doi:10.1103/PhysRevLett.134.017301
Abstract
The rapidly increasing computational demands for artificial intelligence (AI) have spurred the exploration of computing principles beyond conventional digital computers. Physical neural networks (PNNs) offer efficient neuromorphic information processing by harnessing the innate computational power of physical processes; however, training their weight parameters is computationally expensive. We propose a training approach for substantially reducing this training cost. Our training approach merges an optimal control method for continuous-time dynamical systems with a biologically plausible training method--direct feedback alignment. In addition to the reduction of training time, this approach achieves robust processing even under measurement errors and noise without requiring detailed system information. The effectiveness was numerically and experimentally verified in an optoelectronic delay system. Our approach significantly extends the range of physical systems practically usable as PNNs.
28 pages, 10 figures
References in corpus (13)
- Deep Learning with Coherent Nanophotonic Circuits
- Machine learning and the physical sciences
- Recent Advances in Physical Reservoir Computing: A Review
- Parallel convolution processing using an integrated photonic tensor core
- Photonics for artificial intelligence and neuromorphic computing
- Deep physical neural networks enabled by a backpropagation algorithm for arbitrary physical systems
- Brain-inspired computing: We need a master plan
- Wave Physics as an Analog Recurrent Neural Network
- Delocalized Photonic Deep Learning on the Internet's Edge
- Complex Dynamics and Synchronization of Delayed-Feedback Nonlinear Oscillators
- Demonstration of Decentralized, Physics-Driven Learning
- Spin waves propagating through a stripe magnetic domain structure and their applications to reservoir computing
- Physical deep learning based on optimal control of dynamical systems