Intelligent control with hierarchical stacked neural networks

摘要

An intelligent control system based on an explicit model of cognitive development (Table 1) performs high-level functions. It comprises up to O hierarchically stacked neural networks, Nm, . . . , Nm+(O−1), where m denotes the stage/order tasks performed in the first neural network, Nm, and O denotes the highest stage/order tasks performed in the highest-level neural network. The type of processing actions performed in a network, Nm, corresponds to the complexity for stage/order m. Thus N1 performs tasks at the level corresponding to stage/order 1. N5 processes information at the level corresponding to stage/order 5. Stacked neural networks begin and end at any stage/order, but information must be processed by each stage in ascending order sequence. Stages/orders cannot be skipped. Each neural network in a stack may use different architectures, interconnections, algorithms, and training methods, depending on the stage/order of the neural network and the type of intelligent control system implemented.

主权项

1. An artificial neural network system configured to receive input data and produce an abstracted output in dependence on the received input data, comprising:
a plurality of successive artificial neural network layers, each respective successive artificial neural network layer being implemented by at least one automated processor and comprising an array of hidden layer neurons and a respective set of weights in a stacked architecture, the array of hidden layer neurons of a respective artificial neural network layer having a state dependent on at least a state of a preceding artificial neural network layer, and a respective set of connection weights to the preceding artificial neural network layer; at least one artificial neural network layer further automatically receiving feedback from at least one succeeding artificial neural network layer; each respective set of connection weights being dependent on at least training information, wherein the training information comprises a relationship of the received input data and abstract information represented in the respective received input data, wherein the feedback received from the at least one succeeding artificial neural network layer acts to modify at least one connection weight; the artificial neural network system being implemented to achieve a predefined level of abstraction based on at least the training information modified based on the feedback; wherein the arrangement of at least a respective array of hidden layer neurons, and the respective sets of connection weights define an architecture of the artificial neural network layer; and the plurality of artificial neural network layers each having a different respective architecture and operating sequentially to achieve a plurality of levels of abstraction between the received input data and an artificial neural network output.