Allen Institute Modeling Workshop | Stefan Mihalas
Stefan Mihalas, Allen Institute MindScope Program, presents on: "
Is cortex a hierarchically organized set of canonical local circuits?" at the Allen Institute Modeling Workshop, a three-day virtual event hosted by the Allen Institute for Brain Science emphasizing open and publicly accessible models and associated tools.
•••
Attendees at the workshop were invited to share written questions for the speakers. The speakers addressed many of the questions live at the end of their talks. Stefan’s written responses to the questions remaining at the end of his talk are posted here:
Q: Stefan — Do you think that the brain’s canonical circuitry can be good at tasks that involves spatio-temporal processing because naturally the brain has to deal with dynamic inputs?
Stefan Mihalas: Yes, most definitely. We have preliminary work in this area, but it would have been way to much to include it in this talk.
Q: Stefan. Enjoyed your talk. Thank you. Since many aspects of cortical computing emerge thru activity dependent processes during early (and later ) development, have you considered how emergence of refined circuits , perhaps with behavior or recognition as the teacher, can influence these cortical modules. Second, the Felleman and Van Essen hierarchy was built based on the anatomical rule off laminar connectivity. It seems to follow some ideas about a physiological hierarchy but there are some areas that may be functionally 'misplaced'. In macaque, area MT is one synapse away from V1 but is located several anatomical hierarchical layers above V1. Are there similar 'misplacements' in mouse cortex? How would a shortest distance rule impact mouse hierarchy? Thank you
Stefan Mihalas:
1. Yes. In top-down models we look at how the structure emerges to provide a computation. Our published work is limited to local circuit computations: in https://pubmed.ncbi.nlm.nih.gov/32390... we show that to integrate context local connections need to follow observed statistics from experiments. We are interested in pursuing these questions using task driven models.
2. Yes. There are many such misplaced connections in the mouse cortex. We quantified how much a misplacement that is (e.g. V1 projecting to MT with a feedforward connection is much less a misplacement than projecting to prefrontal cortex with a feedback connection), and the hierarchy was constructed to minimize such misplacements. For details see https://www.nature.com/articles/s4158...