Sequential neural network model
Hui Wang, David Bell · 2002
We consider the problem of sequential processing (many-to-one mapping) and present a sequential backpropagation model, which is a generalization of the BP model and is intended to deal with the time dependent sequentiality of input patterns existing in varieties of practical problems, such as word recognition, speech recognition, natural language understanding, and so on. This model can be used to train a network to learn the sequentiality of input patterns, in fixed order or random order. Compared with the original BP model, this model is suitable for both one-to-one mapping and many-to-one mapping, and characterised as "recognising while accumulating". Besides, this model is open and partial associative, which is more cognition oriented.>