A Dendritic Brunel Network for Studying NMDA-Driven Working Memory

Zenodo (CERN European Organization for Nuclear Research)

Abstract

Introduction Maintaining a working memory of sensory inputs is a fundamental neural computation, widely thought to rely on synaptic plasticity or dedicated attractor dynamics [1,2]. A complementary substrate is provided by dendritic subunits equipped with NMDA receptors, whose slowly decaying, voltage-gated conductances allow information to persist within individual neurons beyond the membrane timescale. To study how NMDA-driven dendritic dynamics contribute to working memory in a recurrent network, we introduce a dendritic extension of the classical Brunel network [3], yielding a minimal system with realistic NMDA kinetics that preserves the well-characterized asynchronous irregular dynamics of the original model. Methods Excitatory neurons are modeled with one somatic and five dendritic compartments incorporating AMPA/NMDA receptor kinetics. In this neuron model, NMDA receptors provide voltage-gated, slowly decaying conductances. Excitatory neurons are organized into clusters with tunable inter- and intra-cluster connection probabilities; inhibitory neurons provide global inhibition targeting random dendritic compartments. MNIST digit images are presented by converting pixel intensities to Poisson spike train firing rates. Network activity serves as a reservoir, and a linear readout trained on spike rates classifies digit identity. Decoding accuracy is evaluated across post-stimulus time windows to quantify the temporal persistence of stimulus information. Results Simulations are ongoing; early observations suggest that excitatory clustering modulates the emergence of NMDA-driven activity in the dendritic compartments and extends the temporal persistence of stimulus-specific population dynamics beyond the timescales seen in the standard Brunel network. Preliminary decoding results indicate that stimulus information is retained over longer post-stimulus intervals, with above-chance classification accuracy persisting at delays where the standard network falls to chance. Discussion These results show that spiking networks can maintain working memory without synaptic plasticity, positioning our dendritic extension of the Brunel network as a minimal, well-grounded reference model for studying how NMDA-driven dendritic subunits can contribute to working memory in recurrent spiking networks. Future work will formally characterize the asynchronous irregular regime and map how connectivity structure and NMDA kinetics jointly determine the network's capacity for sustained stimulus retention. Acknowledgments NeuroSys as part of the initiative "Clusters4Future" is funded by the Federal Ministry of Education and Research BMBF (03ZU2106CB). References [1] Mongillo, G., Barak, O., & Tsodyks, M. (2008). Synaptic Theory of Working Memory. Science, 319(5869), 1543-1546. doi: 10.1126/science.1150769[2] Amit, D. (1997). Model of global spontaneous activity and local structured activity during delay periods in the cerebral cortex. Cerebral Cortex, 7(3), 237-252. doi: 10.1093/cercor/7.3.237[3] Brunel, N. (2000). Dynamics of Sparsely Connected Networks of Excitatory and Inhibitory Spiking Neurons. Journal of Computational Neuroscience, 8(3), 183-208. doi: 10.1023/A:1008925309027

Authors 4

1. Noah Ostendorf Aachen Faculty of Computer Science 3--Software Engineering

   Forschungszentrum Jülich · RWTH Aachen University

   Affiliation as printed

   Institute for Advanced Simulations (IAS-6), Research Center Jülich, Jülich, Germany; Faculty of Computer Science 3--Software Engineering, RWTH Aachen University, Aachen, Germany

2. Joshua Böttcher Aachen Faculty of Computer Science 3--Software Engineering

   Forschungszentrum Jülich · RWTH Aachen University

   Affiliation as printed

   Institute for Advanced Simulations (IAS-6), Research Center Jülich, Jülich, Germany; Faculty of Computer Science 3--Software Engineering, RWTH Aachen University, Aachen, Germany

3. Abigail Morrison Aachen Faculty of Computer Science 3--Software Engineering

   Forschungszentrum Jülich · RWTH Aachen University

   Affiliation as printed

   Institute for Advanced Simulations (IAS-6), Research Center Jülich, Jülich, Germany; Faculty of Computer Science 3--Software Engineering, RWTH Aachen University, Aachen, Germany

4. Willem A. M. Wybo

   Forschungszentrum Jülich

   Affiliation as printed

   Peter Grünberg Institute (PGI 15), Research Center Jülich, Jülich, Germany