Real Time Multi Chip Neural Network For Cognitive Systems

Simulation of brain neurons in real-time using biophysically-meaningful models is a pre-requisite for comprehensive understanding of how neurons process information and communicate with each other, in effect efficiently complementing in-vivo experiments. In spiking neural networks (SNNs), propagated information is not just encoded by the firing rate of each neuron in the network, as in artificial neural networks (ANNs), but, in addition, by amplitude, spike-train patterns, and the transfer rate. The high level of realism of SNNs and more significant computational and analytic capabilities in comparison with ANNs, however, limit the size of the realized networks. Consequently, the main challenge in building complex and biophysically-accurate SNNs is largely posed by the high computational and data transfer demands.Real-Time Multi-Chip Neural Network for Cognitive Systems presents novel real-time, reconfigurable, multi-chip SNN system architecture based on localized communication, which effectively reduces the communication cost to a linear growth. The system use double floating-point arithmetic for the most biologically accurate cell behavior simulation, and is flexible enough to offer an easy implementation of various neuron network topologies, cell communication schemes, as well as models and kinds of cells. The system offers a high run-time configurability, which reduces the need for resynthesizing the system. In addition, the simulator features configurable on- and off-chip communication latencies as well as neuron calculation latencies. All parts of the system are generated automatically based on the neuron interconnection scheme in use. The simulator allows exploration of different system configurations, e.g. the interconnection scheme between the neurons, the intracellular concentration of different chemical compounds (ions), which affect how action potentials are initiated and propagate.

This book presents neuromorphic cognitive systems from a learning and memory-centered perspective. It illustrates how to build a system network of neurons to perform spike-based information processing, computing, and high-level cognitive tasks. It is beneficial to a wide spectrum of readers, including undergraduate and postgraduate students and researchers who are interested in neuromorphic computing and neuromorphic engineering, as well as engineers and professionals in industry who are involved in the design and applications of neuromorphic cognitive systems, neuromorphic sensors and processors, and cognitive robotics. The book formulates a systematic framework, from the basic mathematical and computational methods in spike-based neural encoding, learning in both single and multi-layered networks, to a near cognitive level composed of memory and cognition. Since the mechanisms for integrating spiking neurons integrate to formulate cognitive functions as in the brain are little understood, studies of neuromorphic cognitive systems are urgently needed. The topics covered in this book range from the neuronal level to the system level. In the neuronal level, synaptic adaptation plays an important role in learning patterns. In order to perform higher-level cognitive functions such as recognition and memory, spiking neurons with learning abilities are consistently integrated, building a system with encoding, learning and memory functionalities. The book describes these aspects in detail.

A significant amount of effort in neural modeling is directed towards understanding the representation of information in various parts of the brain, such as cortical maps [6], and the paths along which sensory information is processed. Though the time domain is integral an integral aspect of the functioning of biological systems, it has proven very challenging to incorporate the time domain effectively in neural network models. A promising path that is being explored is to study the importance of synchronization in biological systems. Synchronization plays a critical role in the interactions between neurons in the brain, giving rise to perceptual phenomena, and explaining multiple effects such as visual contour integration, and the separation of superposed inputs. The purpose of this book is to provide a unified view of how the time domain can be effectively employed in neural network models. A first direction to consider is to deploy oscillators that model temporal firing patterns of a neuron or a group of neurons. There is a growing body of research on the use of oscillatory neural networks, and their ability to synchronize under the right conditions. Such networks of synchronizing elements have been shown to be effective in image processing and segmentation tasks, and also in solving the binding problem, which is of great significance in the field of neuroscience. The oscillatory neural models can be employed at multiple scales of abstraction, ranging from individual neurons, to groups of neurons using Wilson-Cowan modeling techniques and eventually to the behavior of entire brain regions as revealed in oscillations observed in EEG recordings. A second interesting direction to consider is to understand the effect of different neural network topologies on their ability to create the desired synchronization. A third direction of interest is the extraction of temporal signaling patterns from brain imaging data such as EEG and fMRI. Hence this Special Session is of emerging interest in the brain sciences, as imaging techniques are able to resolve sufficient temporal detail to provide an insight into how the time domain is deployed in cognitive function. The following broad topics will be covered in the book: Synchronization, phase-locking behavior, image processing, image segmentation, temporal pattern analysis, EEG analysis, fMRI analyis, network topology and synchronizability, cortical interactions involving synchronization, and oscillatory neural networks. This book will benefit readers interested in the topics of computational neuroscience, applying neural network models to understand brain function, extracting temporal information from brain imaging data, and emerging techniques for image segmentation using oscillatory networks

Spiking neural networks (SNN) are biologically inspired computational models that represent and process information internally as trains of spikes. This monograph book presents the classical theory and applications of SNN, including original author’s contribution to the area. The book introduces for the first time not only deep learning and deep knowledge representation in the human brain and in brain-inspired SNN, but takes that further to develop new types of AI systems, called in the book brain-inspired AI (BI-AI). BI-AI systems are illustrated on: cognitive brain data, including EEG, fMRI and DTI; audio-visual data; brain-computer interfaces; personalized modelling in bio-neuroinformatics; multisensory streaming data modelling in finance, environment and ecology; data compression; neuromorphic hardware implementation. Future directions, such as the integration of multiple modalities, such as quantum-, molecular- and brain information processing, is presented in the last chapter. The book is a research book for postgraduate students, researchers and practitioners across wider areas, including computer and information sciences, engineering, applied mathematics, bio- and neurosciences.

Understanding and implementing the brain's computational paradigm is the one true grand challenge facing computer researchers. Not only are the brain's computational capabilities far beyond those of conventional computers, its energy efficiency is truly remarkable. This book, written from the perspective of a computer designer and targeted at computer researchers, is intended to give both background and lay out a course of action for studying the brain's computational paradigm. It contains a mix of concepts and ideas drawn from computational neuroscience, combined with those of the author. As background, relevant biological features are described in terms of their computational and communication properties. The brain's neocortex is constructed of massively interconnected neurons that compute and communicate via voltage spikes, and a strong argument can be made that precise spike timing is an essential element of the paradigm. Drawing from the biological features, a mathematics-based computational paradigm is constructed. The key feature is spiking neurons that perform communication and processing in space-time, with emphasis on time. In these paradigms, time is used as a freely available resource for both communication and computation. Neuron models are first discussed in general, and one is chosen for detailed development. Using the model, single-neuron computation is first explored. Neuron inputs are encoded as spike patterns, and the neuron is trained to identify input pattern similarities. Individual neurons are building blocks for constructing larger ensembles, referred to as "columns". These columns are trained in an unsupervised manner and operate collectively to perform the basic cognitive function of pattern clustering. Similar input patterns are mapped to a much smaller set of similar output patterns, thereby dividing the input patterns into identifiable clusters. Larger cognitive systems are formed by combining columns into a hierarchical architecture. These higher level architectures are the subject of ongoing study, and progress to date is described in detail in later chapters. Simulation plays a major role in model development, and the simulation infrastructure developed by the author is described.

An Introduction to Neural Networks falls into a new ecological niche for texts. Based on notes that have been class-tested for more than a decade, it is aimed at cognitive science and neuroscience students who need to understand brain function in terms of computational modeling, and at engineers who want to go beyond formal algorithms to applications and computing strategies. It is the only current text to approach networks from a broad neuroscience and cognitive science perspective, with an emphasis on the biology and psychology behind the assumptions of the models, as well as on what the models might be used for. It describes the mathematical and computational tools needed and provides an account of the author's own ideas. Students learn how to teach arithmetic to a neural network and get a short course on linear associative memory and adaptive maps. They are introduced to the author's brain-state-in-a-box (BSB) model and are provided with some of the neurobiological background necessary for a firm grasp of the general subject. The field now known as neural networks has split in recent years into two major groups, mirrored in the texts that are currently available: the engineers who are primarily interested in practical applications of the new adaptive, parallel computing technology, and the cognitive scientists and neuroscientists who are interested in scientific applications. As the gap between these two groups widens, Anderson notes that the academics have tended to drift off into irrelevant, often excessively abstract research while the engineers have lost contact with the source of ideas in the field. Neuroscience, he points out, provides a rich and valuable source of ideas about data representation and setting up the data representation is the major part of neural network programming. Both cognitive science and neuroscience give insights into how this can be done effectively: cognitive science suggests what to compute and neuroscience suggests how to compute it.

Artificial Intelligence (AI) has found many applications in the past decade due to the ever increasing computing power. Artificial Neural Networks are inspired in the brain structure and consist in the interconnection of artificial neurons through artificial synapses. Training these systems requires huge amounts of data and, after the network is trained, it can recognize unforeseen data and provide useful information. The so-called Spiking Neural Networks behave similarly to how the brain functions and are very energy efficient. Up to this moment, both spiking and conventional neural networks have been implemented in software programs running on conventional computing units. However, this approach requires high computing power, a large physical space and is energy inefficient. Thus, there is an increasing interest in developing AI tools directly implemented in hardware. The first hardware demonstrations have been based on CMOS circuits for neurons and specific communication protocols for synapses. However, to further increase training speed and energy efficiency while decreasing system size, the combination of CMOS neurons with memristor synapses is being explored. The memristor is a resistor with memory which behaves similarly to biological synapses. This book explores the state-of-the-art of neuromorphic circuits implementing neural networks with memristors for AI applications.