The CHIR Algorithm: A Generalization for Multiple-Output and Multilayered Networks

摘要: A new learning algorithm, lear ning by choice of int ern al repr esentations (CHIR), was recently roduc ed. Th e basic version thi s algorit hs developed for a two-layer, single-out put, feed­ forward network binary neurons . This paper presents gener alized t he CHIR algorithm th at is capable training mult iple­ output net works. way to ada pt the hm ilayered networks also presented. 'v Ve test on two typical learnin g tas ks: combined parity- symm et ry problem an d rand om (random associat ions). The dependence algori m performance size and pa­ ramet ers st udied.