An accelerator for neural networks with pulse-coded model neurons

Abstract

The labeling of features by synchronization of spikes seems to be a very efficient encoding scheme for a visual system. Simulation of a vision system with millions of pulse-coded model neurons, however, is almost impossible on the base of available processors including parallel processors and neurocomputers. A "one-to-one" silicon implementation of pulse-coded model neurons suffers from communication problems and low flexibility. On the other hand, acceleration of the simulation algorithm of pulse-coded leaky integrator neurons has proved to be straightforward, flexible, and very efficient. Thus we decided to develop an accelerator for a special version of the French and Stein neurons with modulatory inputs which are advantageous for simulation of synchronization mechanisms. Moreover, our accelerator also provides a Hebbian-like learning rule and supports adaptivity. Up to 128 K neurons with a total number of 16 M freely allocatable synapses are simulated within one system. The size of networks, however, is not at all limited by these numbers as the system may be arbitrarily expanded. Simulation speed obviously depends on the number of interconnections and on the average activity within the network. In the case of locally interconnected networks for simulation of vision mechanisms there is only a very low percentage of simultaneously active neurons: stimuli are not simultaneously presented in all orientations and at all positions of the visual field. In these cases our accelerator provides close to real-time behavior if one second of a biological neuron is simulated by 1000 time slots.

Neural mechanisms of scene segmentation: recordings from the visual cortex suggest basic circuits for linking field models

“…We will discuss these effects with respect to an extension of the previous linking field hypothesis in which phase coupling among rhythmic spike patterns at different frequencies can define feature relations [24]- [26]. We began to implement our linking field model for technical applications in collaboration with four other groups which forced the development of hardware accelerators capable of modeling large PCNN's of linking field neurons in real time by two of the collaborating groups (see [69]). …”

Implementing Spiking Neural Networks for Real-Time Signal-Processing and Control Applications: A Model-Validated FPGA Approach

“…This can lead to unacceptable performance in real-time applications and has, therefore, not been adopted in this design. In other hardware NN processors, the network interconnection information is stored using a target orientated, i.e., postsynaptic address, rather than one based on storing source information [31], [39], [40]. A target-orientated approach is favored because of the sparse interneural connectivity and low average activity rates of neuronal networks.…”

Biology-Inspired Early Vision System for a Spike Processing Neurocomputer

“…One implementation is the SPIKE128K system [5,6] which allows the close to real time simulation of networks with 131072 neurons and up to 16 million connections. In this context, real time means an execution time of 1 ms for one simulation timeslot.…”

“…Several neurocomputer architectures have been designed to offer the required simulation power. The presented network is designed for the SPIKE [5,6,17] neurocomputer architecture, especially for the ParSPIKE design [17]. ParSPIKE is based on a parallel DSP array and the presented network is adapted to this architecture.…”