KIP-Veröffentlichungen

Jahr 2020
Autor(en) Sebastian Billaudelle, Benjamin Cramer, Mihai A. Petrovici, Johannes Schemmel
Titel Structural plasticity on spiking neuromorphic hardware
KIP-Nummer HD-KIP 20-29
KIP-Gruppe(n) F9
Dokumentart Paper
Quelle Neuro-Inspired Computational Elements Workshop (NICE) 2020
Abstract (en)

The nervous system adheres to its spatial and energetic footprint by constraining the number of expressed synapses and enforcing a sparse connectome. To still be able to adapt to its inputs and learn new tasks, synaptic plasticity can not be limited to only changing synaptic weights: the connectome itself undergoes continuous structural modifications during the lifetime of an individual. Since neuromorphic systems impose similar limits on the implementable networks, one may again draw inspiration from the biological archetype.

 

We propose an efficient implementation of structural plasticity and demonstrate its functionality on the BrainScaleS-2 system. The plasticity rule enables neurons to dynamically select a set of suitable synapses out of a pool of potential connections. This policy optimizes performance in a chosen task and at the same time maintains a sparse connectome. It is evaluated in a supervised learning experiment: A two-layer network is tasked with the classification of the Iris data set. It reaches near-perfect performance for various sparsity levels.

bibtex
@conference{billaudelle2020structuralabstract,
  author   = {Sebastian Billaudelle, Benjamin Cramer, Mihai A. Petrovici, Johannes Schemmel},
  title    = {Structural plasticity on spiking neuromorphic hardware},
  booktitle = {},
  year     = {2020},
  volume   = {},
  pages    = {}
}
KIP - Bibliothek
Im Neuenheimer Feld 227
Raum 3.402
69120 Heidelberg