Skip to content

Feature Discussion – STDP and Delays

terhorstd edited this page Jun 29, 2020 · 4 revisions

Thursday, June 25, 2020

Introduction

In the NEST Meeting on May 25th there was a very interesting discussion of handling of delays in NEST, with focus on STDP models. It became obvious, that this topic requires a more in-depth discussion and it was decided to do a separate VC for this discussion.

Reference Materials

Relevant background reading is the following article

  • Phenomenological models of synaptic plasticity based on spike timing, Abigail Morrison, Markus Diesmann & Wulfram Gerstner; Biological Cybernetics volume 98, pages 459–478 (May 2008), doi: 10.1007/s00422-008-0233-1
  • Slides t.b.a.?

Agenda

  • Brief introduction of key speakers (if necessary)
  • Introduction into the topic
  • Discussion
  • Conclusions

Introduction to the topic

Bruno Golosio presented an overview and current work on a prototype implementation of different handling of delays. The impact on STDP weight updates was investigated and different mechanisms for handling spike event were suggested.

Discussion

  • processing of information on pre- and post-synaptic side distributed to different compute nodes
  • effect of the timing of weight updates
    • difference of axonal/dendritic delays
    • mechanisms of weight normalization
  • implementation proposal of spike-buffers
  • effect of spike delivery times on the simulation speed
  • numeric precision of implementations, GPU vs CPU

Conclusions

  • the discussed ideas will be looked at further
  • possibilities to realize the implementation also with MPI will be investigated
  • we will schedule another meeting in a couple of weeks
Clone this wiki locally