Mo Shakiba


Anatomically and Functionally Constrained Bio-Inspired Recurrent Neural Networks Outperform Traditional RNN Models.

Shakiba M, Rokni R, Mohammadi M, Dehghani N

Kempner Institute Frontiers in NeuroAI Symposium, 2025. Poster

CITATIONS:

Abstract

“We introduce and train a bio-inspired RNN that integrates detailed anatomical connectivity and functional data from the MICrONS dataset. Using neuronal positions, synaptic connections, functional correlations, and Spike Time Tiling Coefficients (STTC)-a robust metric that eliminates firing rate biases- from a subset of these neurons, we constrain our model using biologically informed weight initialization, communicability metrics, and regularizer that favors short, communicative connections to reflect realistic network properties.”

  • Key: frnt-rnn

Tags

neurosciencemachine learningrecurrent neural networksbio-inspired modelsneural connectivitycomputational neuroscienceMICrONS