Learning to Make Coherent Predictions in Domains with Discontinuities

Part of Advances in Neural Information Processing Systems 4 (NIPS 1991)

Bibtex Metadata Paper


Suzanna Becker, Geoffrey E. Hinton


We have previously described an unsupervised learning procedure that discovers spatially coherent propertit>_<; of the world by maximizing the in(cid:173) formation that parameters extracted from different parts of the sensory input convey about some common underlying cause. When given random dot stereograms of curved surfaces, this procedure learns to extract sur(cid:173) face depth because that is the property that is coherent across space. It also learns how to interpolate the depth at one location from the depths at nearby locations (Becker and Hint.oll. 1992). 1n this paper, we pro(cid:173) pose two new models which handle surfaces with discontinuities. The first model attempts to detect cases of discontinuities and reject them. The second model develops a mixture of expert interpolators. It learns to de(cid:173) tect the locations of discontinuities and to invoke specialized, asymmetric interpolators that do not cross the discontinuities.