Neural network learning such as the error backpropagation training is considered biologically implausible. The generalized recirculation (O'Reilly, 1996) or the deterministic Boltz-mann machine is biologically plausible, which is usually evoked by a local learning rule that does noe require another error propagating mechanism. This paper presents a more bioloigcally plausible network, which, besides the locality of learning requirement, can overcome some defects of the previous algorithms sustained by inefficient learning, and can provide a practically powerful algorithm.
展开▼
