ParSPIKE-a parallel DSP-accelerator for dynamic simulation of large spiking neural networks

Wolff, Carsten; Hartmann, Georg; Rückert, Ulrich

doi:10.1109/mn.1999.758882

Cited by 13 publications

(10 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…decaying of all dendritic potentials in one step or processing of synapses and neurons with parallel units. A further parallel approach is the distribution of neurons or synapses into several equal processing units 22,23,33]. The communication between the units is based on spikes or spike list exchange.…”

Section: Basic Ideasmentioning

confidence: 99%

“…The resubmission has to be processed if the neuron receives input or is able to generate a new spike. Otherwise the neuron can be removed from the resubmission list after a given time period 33,34]. The resubmission time can be calculated from the time when -after an increment caused by a spike emission -the decaying dynamic threshold reaches again the decaying membrane potential of the neuron.…”

Section: Reducing the Neuron Calculationsmentioning

confidence: 99%

See 1 more Smart Citation

Simulation of spiking neural networks — architectures and implementations

et al. 2002

View full text Add to dashboard Cite

The fast simulation of large networks of spiking neurons is a major task for the examination of biology-inspired vision systems. Networks of this type label features by synchronization of spikes and there is strong demand to simulate these e ects in real world environments. As the calculations for one model neuron are complex, the digital simulation of large networks is not e cient using existing simulation systems. Consequently, it is necessary to develop special simulation techniques. This article introduces a wide range of concepts for the di erent parts of digital simulator systems for large vision networks and presents accelerators based on these foundations. Pulse-Coded Neural Vision NetworksThe communication in PCNNs is based on spike e x c hange. In contrast to conventional model neurons, e.g. McCulloch & Pitts neurons, the generation of a spike requires high computational e ort in connection with the time behaviour in the biological example. The computational e ort for individual neuron calculations compared to whole network processing is much higher in PCNNs than in conventional ANNs.Common simulation techniques for neural networks make use of vector representations for the neurons and matrix representations for the connection network 11]. These techniques are not suitable for PCNNs because the actual activity of one neuron is not representable by o n l y o n e v alue. Hence, common simulation techniques based on the acceleration of matrix-vector-calculations are not su cient for PCNNs. A new simulation paradigm is required with

show abstract

Section: Basic Ideasmentioning

confidence: 99%

Section: Reducing the Neuron Calculationsmentioning

confidence: 99%

Simulation of spiking neural networks — architectures and implementations

et al. 2002

View full text Add to dashboard Cite

show abstract

“…In this context, real time means an execution time of 1 ms for one simulation timeslot. The ParSPIKE design [17] extends the architecture by a parallel approach to networks with more than one million neurons. In both systems the neurons are represented by the values of their dendritic potentials and of the dynamic threshold.…”

Section: Adaptation To the Parspikementioning

confidence: 99%

“…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.…”

Section: Introductionmentioning

confidence: 99%

Biology-Inspired Early Vision System for a Spike Processing Neurocomputer

Thiem

Wolff

Hartmann

2000

Biologically Motivated Computer Vision

View full text Add to dashboard Cite

Abstract. Examinations of the early vision system of mammals have shown Gabor-like behaviour of the simple cell responses. Furthermore, the advantages of sc Gabor-like image filtering are evident in computer vision. This causes strong demand to achieve Gabor-like responses in biology inspired spiking neural networks and so we propose a neural vision network based on spiking neurons with Gabor-like simple cell responses. The Gabor behaviour is theoretically derived and demonstrated with simulation results. Our network consists of a cascaded structure of photoreceptors, ganglion and horizontal neurons and simple cells. The receptors are arranged on a hexagonal grid. One main advantage of our approach compared to direct Gabor filtering is the availability of valuable intersignals. The network is designed as the preprocessing stage for pulse-coupled neural vision networks simulated on the SPIKE neurocomputer architecture. Hence, the simple cells are implemented as Eckhorn neurons.

show abstract

“…The most recent developments for the simulation of I pulsed neural networks are parSpike [5], which is based on DSP Processors, NESPINN [I] and MASPINN [4], which are application specific integrated circuits. All three architectures are optimized for the so called Eckhom model which is comparable t(> an integrate and fire neuron with additional linking input$ for the modulation of other inputs.…”

Section: Introductionmentioning

confidence: 99%

RACER - a rapid prototyping accelerator for pulsed neural networks

Grassmann

Anlauf²

Proceedings. 10th Annual IEEE Symposium on Field-Programmable Custom Computing Machines

View full text Add to dashboard Cite

In this extended abstract we sketch the employment of programmable logicfor the acceleration of the simulation of pulsed neural networks. We compare our approach to solutions which are based on DSPs and digital neuroprocessors. Our solution is a rapid prototyping accelerator board which is based on a data flow concept. The accelerator provides three module sockets with a rather simple 32Bit interface. The design is focused on a maximal data through-put to and from each module. Due to the architecture a very high parallelism between the modules can be achieved Two programmable devices on each module are supported by the on-board programming and test unit, which provides in-circuit programming by the host during operation. As a result the accelerator delivers a high performance andflexibility without introducing a complex interface or handling. Any programmable device, FPGA, CPLD or special architectures like KressArrays [2] may be used on a module of this accelerator boara: hence coarse and fine grain architectures can be used.

show abstract

ParSPIKE-a parallel DSP-accelerator for dynamic simulation of large spiking neural networks

Cited by 13 publications

References 14 publications

Simulation of spiking neural networks — architectures and implementations

Simulation of spiking neural networks — architectures and implementations

Biology-Inspired Early Vision System for a Spike Processing Neurocomputer

RACER - a rapid prototyping accelerator for pulsed neural networks

Contact Info

Product

Resources

About