A modular massively parallel learning framework for brain-like computers

It is generally believed that a brain-like computer should possess the following essential capabilities: (a) massively parallel and distributed information processing; (b) real-time information processing; (c) flexible information processing; and (d) be able to solve large-scale problems. However, it seems that there are few existing neural network models which can satisfy the above basic requirements currently. We present a massively parallel and modular learning framework for brain-like computers. We narrow our sights to consider only pattern recognition problems and discuss the characteristics of the framework from the aspects of modularity and parallelism, responsiveness, plasticity, and scalability. We demonstrate that the framework may provide us with a simple model for implementing specific brain-like computers for pattern recognition