Deep physical neural networks enabled by a backpropagation algorithm for arbitrary physical systems

Abstract

Deep neural networks have become a pervasive tool in science and engineering. However, modern deep neural networks' growing energy requirements now increasingly limit their scaling and broader use. We propose a radical alternative for implementing deep neural network models: Physical Neural Networks. We introduce a hybrid physical-digital algorithm called Physics-Aware Training to efficiently train sequences of controllable physical systems to act as deep neural networks. This method automatically trains the functionality of any sequence of real physical systems, directly, using backpropagation, the same technique used for modern deep neural networks. To illustrate their generality, we demonstrate physical neural networks with three diverse physical systems-optical, mechanical, and electrical. Physical neural networks may facilitate unconventional machine learning hardware that is orders of magnitude faster and more energy efficient than conventional electronic processors.

References in corpus (8)

• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Training and Operation of an Integrated Neuromorphic Network Based on Metal-Oxide Memristors
• Survey of Machine Learning Accelerators
• A flexible and fast PyTorch toolkit for simulating training and inference on analog crossbar arrays
• Carbon Emissions and Large Neural Network Training
• Supervised learning in a mechanical system
• Building high accuracy emulators for scientific simulations with deep neural architecture search
• Physical deep learning based on optimal control of dynamical systems

Cited by in corpus (78)

• The physics of optical computing
• Experimentally realized in situ backpropagation for deep learning in nanophotonic neural networks
• Single chip photonic deep neural network with accelerated training
• Neural Network Potentials for Chemistry: Concepts, Applications and Prospects
• Polarization Multiplexed Diffractive Computing: All-Optical Implementation of a Group of Linear Transformations Through a Polarization-Encoded Diffractive Network
• Learning without neurons in physical systems
• Nonlinear optical encoding enabled by recurrent linear scattering
• Massively Parallel Universal Linear Transformations using a Wavelength-Multiplexed Diffractive Optical Network
• Demonstration of Decentralized, Physics-Driven Learning
• Pattern recognition in the nucleation kinetics of non-equilibrium self-assembly
• Roadmap for Unconventional Computing with Nanotechnology
• Optical Generative Models
• Deep Photonic Reservoir Computer Based on Frequency Multiplexing with Fully Analog Connection Between Layers
• Soft Metamaterials: Adaptation and Intelligence
• Machine Learning Without a Processor: Emergent Learning in a Nonlinear Electronic Metamaterial
• Neuromorphic Overparameterisation and Few-Shot Learning in Multilayer Physical Neural Networks
• Measuring, processing, and generating partially coherent light with self-configuring optics
• Residual-based Attention Physics-informed Neural Networks for Spatio-Temporal Ageing Assessment of Transformers Operated in Renewable Power Plants
• Heavy tails and pruning in programmable photonic circuits
• Overcoming the Coherence Time Barrier in Quantum Machine Learning on Temporal Data
• Self-learning Machines based on Hamiltonian Echo Backpropagation
• Training neural networks with end-to-end optical backpropagation
• Quantum-limited stochastic optical neural networks operating at a few quanta per activation
• Intelligent Soft Matter: Towards Embodied Intelligence
• Ultrafast single-channel machine vision based on neuro-inspired photonic computing
• Tackling Sampling Noise in Physical Systems for Machine Learning Applications: Fundamental Limits and Eigentasks
• Mesoscopic ultrafast nonlinear optics -- The emergence of multimode quantum non-Gaussian physics
• Learning to self-fold at a bifurcation
• Referenceless characterisation of complex media using physics-informed neural networks
• Predicting nonlinear optical scattering with physics-driven neural networks
• Programmable metasurfaces for future photonic artificial intelligence
• Mechanical intelligence via fully reconfigurable elastic neuromorphic metasurfaces
• Training Coupled Phase Oscillators as a Neuromorphic Platform using Equilibrium Propagation
• Deep Photonic Reservoir Computer for Speech Recognition
• Asymmetrical estimator for training encapsulated deep photonic neural networks
• Realistic quantum photonic neural networks
• Noise mitigation strategies in physical feedforward neural networks
• Principles and Metrics of Extreme Learning Machines Using a Highly Nonlinear Fiber
• Multiplexed gradient descent: Fast online training of modern datasets on hardware neural networks without backpropagation
• Bias-Scalable Near-Memory CMOS Analog Processor for Machine Learning
• Hardware-Efficient Photonic Tensor Core: Accelerating Deep Neural Networks with Structured Compression
• Intelligent mechanical metamaterials towards learning static and dynamic behaviors
• XLuminA: An Auto-differentiating Discovery Framework for Super-Resolution Microscopy
• High-performance real-world optical computing trained by in situ gradient-based model-free optimization
• Qudit Machine Learning
• Predicting mode-locked fiber laser output using feed-forward neural network
• High-precision programming of large-scale ring resonator circuits with minimal pre-calibration
• Physical Neural Networks with Self-Learning Capabilities
• Optofluidic memory and self-induced nonlinear optical phase change for reservoir computing in silicon photonics
• Pushing the Boundary of Quantum Advantage in Hard Combinatorial Optimization with Probabilistic Computers
• Optical neuromorphic computing via temporal up-sampling and trainable encoding on a telecom device platform
• High-speed coherent photonic random-access memory in long-lasting sound waves
• A Reconfigurable Linear RF Analog Processor for Realizing Microwave Artificial Neural Network
• Arbitrary control over multimode wave propagation for machine learning
• Physical Computing for Materials Acceleration Platforms
• Blending Optimal Control and Biologically Plausible Learning for Noise-Robust Physical Neural Networks
• From masses and radii of neutron stars to EOS of nuclear matter through neural network
• Model-free Optical Processors using In Situ Reinforcement Learning with Proximal Policy Optimization
• Mechanical in-sensor computing: a programmable meta-sensor for structural damage classification without external electronic power
• Differentiable master equation solver for quantum device characterisation
• Self-training superconducting neuromorphic circuits using reinforcement learning rules
• The Physics of Learning
• Deep learning methods for Hamiltonian parameter estimation and magnetic domain image generation in twisted van der Waals magnets
• A universal approximation theorem for nonlinear resistive networks
• Scaling of hardware-compatible perturbative training algorithms
• Internal noise in hardware deep and recurrent neural networks helps with learning
• Dense Associative Memory in a Nonlinear Optical Hopfield Neural Network
• Quantum-secure multiparty deep learning
• Disentangling Aleatoric and Epistemic Uncertainty in Physics-Informed Neural Networks. Application to Insulation Material Degradation Prognostics
• A self-learning magnetic Hopfield neural network with intrinsic gradient descent adaption
• Fisher information flow in artificial neural networks
• Metrics for spin-based computing
• Talking with a ghost: semi-virtual coupled levitated oscillators
• Harnessing intuitive local evolution rules for physical learning
• Irregular Metamaterial Networks
• Polaritonic Machine Learning for Graph-based Data Analysis
• Streamlined optical training of large-scale modern deep learning architectures with direct feedback alignment
• A ghost mechanism: An analytical model of abrupt learning in recurrent networks