year | 2012 |
author(s) | Ioannis Kokkinos |
title | Feasibility Study On Declarative Routing For Neuromorphic Hardware |
KIP-Nummer | HD-KIP 12-114 |
KIP-Gruppe(n) | F9 |
document type | Diplomarbeit |
Abstract (de) | Eine passende Konfiguration für ein neuromorphes Hardware System zu finden, die einem gegebenen neuronalen Netzwerk entspricht, ist eine komplexe Aufgabe. Dies ist äquivalent zur Bestimmung von Graphenisomorphismen, was zu den |
Abstract (en) | Finding a suitable configuration for neuromorphic hardware systems, closely resembling a given neural network, is a complex task. It is equivalent to finding graph isomorphisms, which is known to be an NP-hard problem. The suitability of two neural network topologies for an accelerated parallel configuration procedure has been analyzed. Furthermore, a feasibility study has been carried out covering a declarative routing approach. Therefore, boolean clauses have been derived to describe structural features of a neuromorphic hardware and neural networks. In consequence, any satisfiability solver can be used to find an assignment satisfying these clauses. Lastly, a neuromorphic hardware simulator has been extended by a leaky integrate-and-fire neuron model with adaptation. The implementation has been verified against an estabilished neural network simulator. |
Datei | Diplomarbeit-Ioannis_Kokkinos |