An Associative Generated Model for Multi-signals Based on Deep Learning

During exploring the emergence of language, we found that the brain can extract some common features from the same thing in different representations by pattern recognition and association. Consequently, the brain would establish a connection for identical concept from multi-signals. An associative generated model primarily based on Restricted Boltzmann Machine (RBM) and Deep Belief Network (DBN) is set up for multi-signals to simulate the brain´s ability. The first step is to use the DBNs for extracting features from multiple input signal sources. The second is using top-level RBM to achieve the goal of associating and generating mutually by fusing each feature. Finally, we verify the feasibility of the model through the realization of generating Arabic digital pictures and Chinese characters digital images reciprocally.