Learning by choice of internal representations-feed forward nets with binary weights

The authors have shown (Complex syst. vol.4, p.107 of 1990) a method for deriving the CHIR algorithm, whereby the internal representations as well as the weights are allowed to be modified, via energy minimization consideration. Computer simulations show a fast training process for this algorithm in comparison with the Back-Propagation and CHIR algorithms, both used in conjunction with a feed-forward net with continuous weights. These simulations include the restricted instances of parity, symmetry and parity-symmetry problems.<
>