A collaboration page for work with Eric Shea-Brown's lab at the University of Washington.
Question: What is the character of network correlations in learned and NEF-generated networks of spiking LIF neurons.
Approach:
Simulate a few basic networks (i.e. communication channel and integrator), and examine the correlational structure. Compare and contrast the NEF weights with learned weights.
- We can do the simulations of the networks (i.e. generate files with spike times over populations of cells)
- Eric's group has good methods for doing the correlational analyses.
Points to discuss:
Parameters of the simulations
- How long should runs be? 1000s
- How many neurons in the network? 100/layer
- What kinds of cells and parameters? LIF; 20ms membrane; 2ms refractory; max firing 10-100hz;
- What kind of input? Bandlimited white noise, to 30hz (?)
Other
Extensions:
- If there is a mismatch, can we enforce correlational structure while generating desired weights in the NEF?