 Название: Photonic Reservoir Computing: Optical Recurrent Neural Networks
Автор: Daniel Brunner, Miguel C. Soriano, Guy Van der Sande
Издательство: De Gruyter
Год: 2019
Страниц: 373
Язык: английский
Формат: pdf (true), epub
Размер: 10.1 MB, 28.7 MB
Photonics has long been considered an attractive substrate for next generation implementations of machine-learning concepts. Reservoir Computing tremendously facilitated the realization of recurrent neural networks in analogue hardware. This concept exploits the properties of complex nonlinear dynamical systems, giving rise to photonic reservoirs implemented by semiconductor lasers, telecommunication modulators and integrated photonic chips.
- Ultra-high speed photonic reservoirs.
- Fully implemented analogue photonic neural networks.
- Theoretical framework of computing with such devices.
This book is devoted to a comprehensive compilation of the first hardware platforms employed for photonic reservoir computing. Reservoir computing is a machine learning technique that is susceptible by design to be implemented in hardware. We here show how it has been successfully ported to a number of photonic platforms, paving the way to all-optical information processing. We dedicate this book to Jan Van Campenhout, who first launched the idea of photonic reservoir computing in 2008.
The first chapter starts with a historical overview, as well as a future-minded motivation for photonic computing. Starting with the oldest known mechanical computer, the reader is guided from the first implementation of optical memory to the controversial topic of all-optical computing logic. This walk through the archives of optical computing naturally arrives at an abstract comparison between electrons and photons in terms of their usefulness within the framework of logic computing. While photonic hardware may have some deficiencies in the perspective of logic computing, it can support the transition from the Von Neumann architecture toward analogue and neuro-morphic computing schemes with parallel computations. In this context, this chapter also includes an introduction to the concept of neural networks. In a series of small but detailed examples, the reader becomes familiar with the basic concepts and relevant parameters which constitute either a feed-forward or a recurrent artificial neural network. The end of this chapter firmly grounds the newly introduced concepts of neural networks and Hopfield networks by discussing their physical counterparts, implemented using photonic hardware.
Chapter 2 starts by contrasting digital computing with analogue computing schemes. A very mature industry exists around digital computing, leading to highly optimized digital computers with wide availability and creating a high entry barrier for any alternative computing scheme.
Chapter 3 introduces the state of the art in reservoir integration, mostly focusing on the realization of passive architectures.
Chapter 4 elaborates on the huge potential of large scale photonic reservoirs. The authors discuss several optical configurations that can generate complex network couplings. Taking advantage of their inherent parallelism, these optical networks can easily be made of thousands of nonlinear nodes.
Chapter 5 gives a general introduction of reservoir computing based on delay systems. Delay-system RC are among the first successful RC’s hardware implementations, lending particular importance to this chapter.
Chapter 6 elaborates on the relevance of Ikeda delay dynamics for the control and development of photonic reservoir computers.
Chapter 7 deals with the implementation of photonic RC using semiconductor lasers as the physical substrate.
Chapter 8 focuses on how to build advanced photonic reservoir computers, using an optoelectronic system as their workhorse.
Скачать Photonic Reservoir Computing: Optical Recurrent Neural Networks
True PDF:
ePub:
|
Разделы сайта
Авторизация
Реклама
Photonic Reservoir Computing: Optical Recurrent Neural Networks 
