Slide 2 Recognition in neural networks Hebb rule Activation rule Output

Abstract:  alse positives - Red = false negatives - As predicted, the network collapses at 0.046N. - Issues to consider: 1. This data behaves REALLY nicely. This does not happen in real life. 2. There is a tradeoff between recognition of novelty and recognition of familiarity. - Most stimuli are familiar. Therefore, it is important for familiarity to work. On the other hand, novelty detection may be important for survival. Recognition in neural networks Investigation of the perirhinal cortex .4 - Then I did some other studies looking at variants of threshold rates. In this example, I optimized it based on the total error. * I propose basing recognition on self-organized feature maps -- they are clearly more robust and perhaps can lead to a better model. I hypothesize that it is a better model, but definitely lacking. Recognition 1. Identification. 2. Judgement concerning prior occurrence. a. when it was encountered b. how many times c. context of occurence Investigation of the perirhinal cortex