Search Papers | Poster Sessions | All Posters
Poster B112 in Poster Session B - Thursday, August 8, 2024, 1:30 – 3:30 pm, Johnson Ice Rink
Structural bottlenecks promote temporal coding in spiking neural networks
Zach Mobille1, Usama Bin Sikandar1, Simon Sponberg1, Hannah Choi1; 1Georgia Institute of Technology
Convergent and divergent structure in the networks that comprise biological brains is found universally across many species and brain regions at various scales. Given the frequency with which this structural motif is observed, we investigate what its functional role may be. While previous theories have neglected the role of neuronal spiking, our model and analysis places this aspect at the forefront. For a suite of stimuli with different timescales, we demonstrate that bottlenecks created by network convergence have a stronger preference for spike timing codes than expansion layers created by structural divergence. Our work makes quantitative predictions concerning the relationship between a network's convergent structure and the optimal timescale it can use to encode a dynamic stimulus. These predictions suggest a connection between network architecture and information-processing capabilities hitherto unexplored, which could be confirmed experimentally in future studies.
Keywords: bottleneck temporal coding spike