Physical deep learning with biologically inspired training method: gradient-free approach for physical hardware

Nakajima, M.; Inoue, Katsuma; Tanaka, Kenji; Kuniyoshi, Yasuo; Hashimoto, Toshikazu; Nakajima, Kohei

doi:10.1038/s41467-022-35216-2

Cited by 43 publications

(32 citation statements)

References 67 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, the training algorithm called direct feedback alignment was introduced in Ref. [300], see Figure 13. This is a universal method based on random projection with alternative nonlinear activation and requires no information about the nature of the physical system.…”

Section: Alternative Learning Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Analog Photonics Computing for Information Processing, Inference, and Optimization

Nikita

Berloff

2023

Adv Quantum Tech

View full text Add to dashboard Cite

This review presents an overview of the current state‐of‐the‐art in photonics computing, which leverages photons, photons coupled with matter, and optics‐related technologies for effective and efficient computational purposes. It covers the history and development of photonics computing and modern analogue computing platforms and architectures, focusing on optimization tasks and neural network implementations. The authors examine special‐purpose optimizers, mathematical descriptions of photonics optimizers, and their various interconnections. Disparate applications are discussed, including direct encoding, logistics, finance, phase retrieval, machine learning, neural networks, probabilistic graphical models, and image processing, among many others. The main directions of technological advancement and associated challenges in photonics computing are explored, along with an assessment of its efficiency. Finally, the paper discusses prospects and the field of optical quantum computing, providing insights into the potential applications of this technology.

show abstract

Section: Alternative Learning Methodsmentioning

confidence: 99%

“…Augmented DFA enables parallel, scalable, and physically accelerable training of deep physical networks based on random projection with alternative nonlinearity g(a). Reproduced with permission [300]. Copyright 2022, Springer Nature.…”

mentioning

confidence: 99%

Analog Photonics Computing for Information Processing, Inference, and Optimization

Nikita

Berloff

2023

Adv Quantum Tech

View full text Add to dashboard Cite

show abstract

“…The second category, deep physical neural networks (PNNs), focuses on training the hardware’s physical transformations directly to perform the desired computations. PNNs hold the promise of more scalable, energy-efficient, and faster neural network hardware by exploiting physical transformations and eliminating the conventional software-hardware separation ( 22 , 23 ).…”

mentioning

confidence: 99%

Backpropagation-free training of deep physical neural networks

Momeni,

Rahmani,

Malléjac

et al. 2023

Science

View full text Add to dashboard Cite

Recent successes in deep learning for vision and natural language processing are attributed to larger models but come with energy consumption and scalability issues. Current training of digital deep-learning models primarily relies on backpropagation that is unsuitable for physical implementation. In this work, we propose a simple deep neural network architecture augmented by a physical local learning (PhyLL) algorithm, which enables supervised and unsupervised training of deep physical neural networks without detailed knowledge of the nonlinear physical layer’s properties. We trained diverse wave-based physical neural networks in vowel and image classification experiments, showcasing the universality of our approach. Our method shows advantages over other hardware-aware training schemes by improving training speed, enhancing robustness, and reducing power consumption by eliminating the need for system modeling and thus decreasing digital computation.

show abstract

“…It has been reported that BPTT can be approximated so that RNNs can be trained in a biologically plausible manner [62,64]. Nakajima et al extended these methods and successfully trained multilayer physical RC [65]. By applying those methods to SM-RC, it is expected that SM-RC can be trained in analog and physical systems.…”

mentioning

confidence: 99%

Learning Reservoir Dynamics with Temporal Self-Modulation

Sakemi¹,

Yoshimura²,

Matsuki³

et al. 2023

Preprint

View full text Add to dashboard Cite

Reservoir computing (RC) can efficiently process time-series data by transferring the input signal to randomly connected recurrent neural networks (RNNs), which are referred to as a reservoir. The highdimensional representation of time-series data in the reservoir significantly simplifies subsequent learning tasks. Although this simple architecture allows fast learning and facile physical implementation, the learning performance is inferior to that of other state-of-the-art RNN models. In this paper, to improve the learning ability of RC, we propose self-modulated RC (SM-RC), which extends RC by adding a selfmodulation mechanism. The self-modulation mechanism is realized with two gating variables: an input gate and a reservoir gate. The input gate modulates the input signal, and the reservoir gate modulates the dynamical properties of the reservoir. We demonstrated that SM-RC can perform attention tasks where input information is retained or discarded depending on the input signal. We also found that a chaotic state emerged as a result of learning in SM-RC. This indicates that self-modulation mechanisms provide RC with qualitatively different information-processing capabilities. Furthermore, SM-RC outperformed RC in NARMA and Lorentz model tasks. In particular, SM-RC achieved a higher prediction accuracy than RC with a reservoir 10 times larger in the Lorentz model tasks. Because the SM-RC architecture only requires two additional gates, it is physically implementable as RC, providing a new direction for realizing edge AI.

show abstract

Physical deep learning with biologically inspired training method: gradient-free approach for physical hardware

Cited by 43 publications

References 67 publications

Analog Photonics Computing for Information Processing, Inference, and Optimization

Analog Photonics Computing for Information Processing, Inference, and Optimization

Backpropagation-free training of deep physical neural networks

Learning Reservoir Dynamics with Temporal Self-Modulation

Contact Info

Product

Resources

About