A Similarity-preserving Network Trained on Transformed Images Recapitulates Salient Features of the Fly Motion Detection Circuit

Yanis Bahroun, Dmitri Chklovskii, Anirvan Sengupta

Abstract: Learning to detect content-independent transformations from data is one of the central problems in biological and artificial intelligence. A well-studied example is the problem of unsupervised learning of a visual motion detector. Rao and Ruderman formulated this problem as learning generators of the Lie group of relevant transformations via minimizing image reconstruction error. Unfortunately, mapping their model onto a biologically plausible neural network (NN) with local learning rules proves to be difficult. Here, we propose a biologically plausible model of motion detection. We also adopt the Lie-group approach but start with a similarity-preserving objective function. An online algorithm that optimizes such an objective function naturally maps onto a NN with biologically plausible learning rules. The trained NN recapitulates major features of the well-studied motion detector in the fly. In particular, it is consistent with the experimental observation that local motion detectors combine information from at least three adjacent pixels, something that contradicts the celebrated Hassenstein-Reichardt model.

CMT Num: 7967