John Platt, Nada Matic
This paper discusses a fairly general adaptation algorithm which augments a standard neural network to increase its recognition ac(cid:173) curacy for a specific user. The basis for the algorithm is that the output of a neural network is characteristic of the input, even when the output is incorrect. We exploit this characteristic output by using an Output Adaptation Module (OAM) which maps this out(cid:173) put into the correct user-dependent confidence vector. The OAM is a simplified Resource Allocating Network which constructs ra(cid:173) dial basis functions on-line. We applied the OAM to construct a writer-adaptive character recognition system for on-line hand(cid:173) printed characters. The OAM decreases the word error rate on a test set by an average of 45%, while creating only 3 to 25 basis functions for each writer in the test set.