Realistic models of cortical pyramidal neurons based on accurate whole-cell synaptic mapping: Implications for biologically-inspired AI models
Abstract
In this study, the whole-neuron synaptic mapping data we generate, enabled by newly developed synaptic labeling and imaging technologies, will be incorporated into anatomically and biophysically detailed models of L2/3 pyramidal neurons. These realistic models will provide new insights regarding the principles governing the processing of bottom-up and top-down visual information, highlighting the functional role of domain-specific synaptic placement. Further, incorporation of realistic neuronal models that include synaptic structural plasticity into artificial systems, would enable the development of a more human-like AI that can solve problems intractable to current AI systems. The experiments and modelling, of the type that we propose here are a critical step towards achieving this goal. Detailed characterization of excitatory and inhibitory synapse spatial organization across individual pyramidal cells, the placement of subcortical versus cortical excitatory inputs within the dendritic synaptic map, and the structural plasticity of these different inputs would be a substantial advance over the present-knowledge of pyramidal neuronsynaptic anatomy. This kind of information is essential for constructing cortically-inspired AI models that are based on computational principles underlying fast and robust learning.Our objective is to use experimentally generated synaptic maps of individual pyramidal neurons as a basis for computational analyses and modeling of local dendritic integration and its contribution to cellular information processing.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 07, 2019
- Source ID
- N000141912036
Entities
People
- Elly Nedivi
Organizations
- Massachusetts Institute of Technology
- Office of Naval Research
- United States Navy