Multiplexed gradient descent: Fast online training of modern datasets on hardware neural networks without backpropagation

Abstract

We present multiplexed gradient descent (MGD), a gradient descent framework designed to easily train analog or digital neural networks in hardware. MGD utilizes zero-order optimization techniques for online training of hardware neural networks. We demonstrate its ability to train neural networks on modern machine learning datasets, including CIFAR-10 and Fashion-MNIST, and compare its performance to backpropagation. Assuming realistic timescales and hardware parameters, our results indicate that these optimization techniques can train a network on emerging hardware platforms orders of magnitude faster than the wall-clock time of training via backpropagation on a standard GPU, even in the presence of imperfect weight updates or device-to-device variations in the hardware. We additionally describe how it can be applied to existing hardware as part of chip-in-the-loop training, or integrated directly at the hardware level. Crucially, the MGD framework is highly flexible, and its gradient descent process can be optimized to compensate for specific hardware limitations such as slow parameter-update speeds or limited input bandwidth.

References in corpus (14)

• Deep Learning with Coherent Nanophotonic Circuits
• Neuromorphic Silicon Photonic Networks
• Deep physical neural networks enabled by a backpropagation algorithm for arbitrary physical systems
• Large-scale neuromorphic optoelectronic computing with a reconfigurable diffractive processing unit
• Brain-inspired computing: We need a master plan
• A one femtojoule athermal silicon modulator
• Eligibility Traces and Plasticity on Behavioral Time Scales: Experimental Support of neoHebbian Three-Factor Learning Rules
• High-speed programmable photonic circuits in a cryogenically compatible, visible-NIR 200 mm CMOS architecture
• Design of optical neural networks with component imprecisions
• Hardware error correction for programmable photonics
• Gradient learning in spiking neural networks by dynamic perturbation of conductances
• Gradients without Backpropagation
• Short-term plasticity as cause-effect hypothesis testing in distal reward learning
• Scaling Forward Gradient With Local Losses

Cited by in corpus (8)

• Single chip photonic deep neural network with accelerated training
• Training Convolutional Neural Networks with the Forward-Forward algorithm
• Model-free Optical Processors using In Situ Reinforcement Learning with Proximal Policy Optimization
• A spiking photonic neural network of 40.000 neurons, trained with rank-order coding for leveraging sparsity
• Self-training superconducting neuromorphic circuits using reinforcement learning rules
• Scaling of hardware-compatible perturbative training algorithms
• Online training and pruning of multi-wavelength photonic neural networks
• Wave Computing based on Dynamical Networks: Applications in Optimization Problems