SUPR
Spatio-temporal covariance properties for neuromorphic neural networks
Dnr:

NAISS 2024/22-1334

Type:

NAISS Small Compute

Principal Investigator:

Jens Egholm Pedersen

Affiliation:

Kungliga Tekniska högskolan

Start Date:

2024-11-01

End Date:

2025-11-01

Primary Classification:

10203: Bioinformatics (Computational Biology) (applications to be 10610)

Allocation

Abstract

Without temporal averaging, such as rate codes, it remains a challenge to precisely train spiking neural networks for temporal regression tasks. Due to the temporal dynamics in biologically plausible neural systems, training a spiking network to provide precise numerical predictions implies a trade-off between dynamics that are active and adaptive versus inactive and sluggish. Here, we present a novel method to accurately predict translation-invariant spatial coordinates from sparse, event-based vision (EBV) signals using a fully spiking convolutional neural network (SCNN) without temporal averaging.