Learning Recurrent Neural Models with Minimal Complexity from Neural Tuning Data

Martin A. Giese

Abstract

We present a method for the estimation of nonlinear recurrent neural models from neural tuning data. Our algorithm makes no specific assumptions about the structure of the recurrent neural network model, except for the smoothness of the lateral interaction profile and feed-forward input kernel. A new learning algorithm is devised that achieves optimal generalization from small amounts of neural data that combines support vector regression with stability constraints resulting from the recurrent network dynamics, resulting in a semi-definite programming problem. The method is illustrated by fitting models for orientation tuning in the visual cortex.
BibTeX
@inproceedings{giese2003learning,
  author = {Martin A. Giese},
  title = {{Learning Recurrent Neural Models with Minimal Complexity from Neural Tuning Data}},
  booktitle = {Neurocomputing},
  year = {2003},
  url = {https://openreview.net/forum?id=UhZ5Q0cNX8},
}

Loading related content...

Figure