Using Non-oscillatory Dynamics to Disambiguate Pattern Mixtures

This chapter describes a model which takes advantage of the time domain through feedback iterations that improve recognition and help disambiguate pattern mixtures. This model belongs to a class of models called generative models. It is based on the hypothesis that recognition centers of the brain reconstruct an internal copy of inputs using knowledge the brain has previously accumulated. Subsequently, it minimizes the error between the internal copy and the input from the environment. This model is distinguished from other generative models which are unsupervised and represent early visual processing. This approach utilizes generative reconstruction in a supervised paradigm, implementing later sensory recognition processing. This chapter shows how this paradigm takes advantage of the time domain, provides a convenient way to store recognition information, and avoids some of the difficulties associated with traditional feedforward classification models.