Understanding how biological structure gives rise to cortical function remains a core challenge in systems neuroscience. We address this by building a large-scale, biologically realistic model of mouse primary visual cortex (V1) that integrates recent anatomical and physiological datasets, including the MICrONS electron microscopy data and the Allen Institute’s synaptic physiology data. The model captures realistic connectivity, synaptic dynamics, and stable spontaneous activity without tuning.
We then use backpropagation through time to refine synaptic weights while maintaining empirical weight distributions. The optimization aligns model activity with in vivo Neuropixels population statistics, reproducing firing-rate distributions, orientation and direction selectivity, and synchronization. Training uncovers emergent structure–function relationships such as like-to-like connectivity and laminar specialization, consistent with experimental data and offering predictions where data are lacking.
This work shows how biologically grounded modeling can unify multimodal datasets into mechanistic frameworks that explain and predict cortical computation—bridging anatomical realism with functional alignment and offering a foundation for both neuroscience and biologically inspired AI.
This Talk will be broadcasted in the following zoom link: https://us06web.zoom.us/j/89027654460?pwd=Wg9TYMPqqP2ipfj2JVvEagmzaTw29c.1
Contact details:
Claudio Mirasso Contact form