Journal Article

FZJ-2022-00386

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png Dynamical Characteristics of Recurrent Neuronal Networks Are Robust Against Low Synaptic Weight Resolution

Dasbach, S. (Corresponding author)FZJ* ; Tetzlaff, T.FZJ* ; Diesmann, M.FZJ* ; Senk, J.FZJ*

2021
Frontiers Research Foundation Lausanne

This record in other databases:    

Please use a persistent id in citations: http://hdl.handle.net/2128/30125  doi:10.3389/fnins.2021.757790

Abstract: The representation of the natural-density, heterogeneous connectivity of neuronalnetwork models at relevant spatial scales remains a challenge for ComputationalNeuroscience and Neuromorphic Computing. In particular, the memory demandsimposed by the vast number of synapses in brain-scale network simulations constitutea major obstacle. Limiting the number resolution of synaptic weights appears to bea natural strategy to reduce memory and compute load. In this study, we investigatethe effects of a limited synaptic-weight resolution on the dynamics of recurrent spikingneuronal networks resembling local cortical circuits and develop strategies for minimizingdeviations from the dynamics of networks with high-resolution synaptic weights. Wemimic the effect of a limited synaptic weight resolution by replacing normally distributedsynaptic weights with weights drawn from a discrete distribution, and compare theresulting statistics characterizing firing rates, spike-train irregularity, and correlationcoefficients with the reference solution. We show that a naive discretization of synapticweights generally leads to a distortion of the spike-train statistics. If the weights arediscretized such that the mean and the variance of the total synaptic input currents arepreserved, the firing statistics remain unaffected for the types of networks considered inthis study. For networks with sufficiently heterogeneous in-degrees, the firing statisticscan be preserved even if all synaptic weights are replaced by the mean of the weightdistribution. We conclude that even for simple networks with non-plastic neurons andsynapses, a discretization of synaptic weights can lead to substantial deviations in thefiring statistics unless the discretization is performed with care and guided by a rigorousvalidation process. For the network model used in this study, the synaptic weightscan be replaced by low-resolution weights without affecting its macroscopic dynamicalcharacteristics, thereby saving substantial amounts of memory.

---

Contributing Institute(s):

1. Computational and Systems Neuroscience (INM-6)
2. Theoretical Neuroscience (IAS-6)
3. Jara-Institut Brain structure-function relationships (INM-10)

Research Program(s):

1. 5231 - Neuroscientific Foundations (POF4-523) (POF4-523)
2. HBP SGA2 - Human Brain Project Specific Grant Agreement 2 (785907) (785907)
3. HBP SGA3 - Human Brain Project Specific Grant Agreement 3 (945539) (945539)
4. ACA - Advanced Computing Architectures (SO-092) (SO-092)
5. Brain-Scale Simulations (jinb33_20191101) (jinb33_20191101)

Appears in the scientific report 2021

---

Database coverage:
; ; ; ; Article Processing Charges ; BIOSIS Previews ; Biological Abstracts ; Clarivate Analytics Master Journal List ; Current Contents - Clinical Medicine ; DOAJ Seal ; Essential Science Indicators ; Fees ; IF < 5 ; JCR ; PubMed Central ; SCOPUS ; Science Citation Index Expanded ; Web of Science Core Collection

---