Meeting w/ Julia, Josh, Juyoung, Dongsoo 100818

[jbmelander]

Shared Components

• New data -> trained model: random shuffling, allocation of validation, cross-correlation datasets from arbitrary datasets

  • Train input
  • Train output
  • Test input
  • Test output

• Dataset documentation and storage

• Model saving / loading

• Visualization Toolbox

• Benchmarking

• Nice way to inspect and manipulate weights and activations

  • Being able to virtually a unit

• Stimuli generation

  • Currently 50x50, want arbitrary size

Experiments to Do

• High resolution models
• Deep retina models with single bin stimulus input
  • Recurrent:
    • Reservoir computing
    • RNN
• Constrain + and - cells
• Temporal convolutions in inner layers

Questions

• Why recurrence?
• Longer time-scales