Response to the letter by Li and Cao

Editor: In a previous work (Esposito, Rampone, & Tagliaferri, 1994), a neural net model which can be used to decode codewords belonging to a binary linear code and to correct noisy patterns which it receives from a transmission channel has been described. The net reduces the error probability to zero in the range of the error correcting capacity of the code, e, when applied to codewords whose maximum Hamming weight is not larger than 2e + 3. In a Letter to the Editor, Li and Cao (1997) proved theoretically that the network can be extended for all binary linear codes without limitations on maximum Hamming weight by means of a learning rule modiication. On the other hand, in a previous article (Caianiello, Marinaro, Rampone, & Tagliaferri, 1996) we achieved the same result with a diierent network model, having the further advantages of binary connection weights and integer thresholds. We suggest that the two approaches be united by a further learning rule modiication, as follows.