Tsvi Achler

Making Sense of Simultaneous Patterns

The ability to look beyond what is learned and apply the learned information to new scenarios often distinguishes animals from computer artifacts. An important component of this ability is recognizing novel combinations of previously learned patterns, which typically form scenes. While animals can quickly evaluate scenes, artificial methods require segmenting and analyzing patterns one by one. Segmentation, however, is not trivial or often even possible. Thus artificial methods show impoverished performance disambiguating patterns composed of simultaneous components such as scenes, cocktail party conversations, and odorants in a mixture. A new type of multiclass classifier motivated by neuroscience is presented that better differentiates simultaneous patterns. It utilizes a gain control mechanism where each piece of information is evaluated by its contribution to the network. Based on evaluation of contribution, the value of information is re-adjusted until the network determines its solutions. It allows more efficient processing of novel combinations of previously learned patterns and can benefit AI applications where simultaneous processes may occur at the same time.

Official inquiries about AIIS should be directed to Alexandre Klementiev (klementi AT uiuc DOT edu)
Last update: 08/30/2007