Online learning chip based on stacked width learning model

Abstract

The invention provides an online learning chip based on a stacked width learning model. The online learning chip comprises a main controller, a memory and a reconfigurable processing unit array, the operation mode is as follows: the main controller reads a control code and controls the configuration information memory to output initialized configuration information to the reconfigurable processing unit array; the reconfigurable processing unit array configures each feature node and each enhancement node to a processing unit in a one-to-one manner according to the initialized configuration information; the input memory inputs the identification training sample features into the model circuit; a processing result is output to an output memory, and then performance value information is fed back to the main controller; the calculated performance value is judged so as to update the width configuration information or the depth configuration information in the configuration information memory; and when the calculated performance value is greater than or equal to the set performance threshold value, online learning is stopped, and the online learning chip is solidified. The chip has the advantages of low computing resource, low power consumption, strong real-time performance and online learning capability.

Description

technical field [0001] The present invention relates to the technical field of a stacked width learning model, and more specifically, relates to an online learning chip based on a stacked width learning model. Background technique [0002] The Broad Learning System (BLS) is an efficient and shallow incremental neural network learning model. It maps the input into a set of feature nodes, which in turn maps the feature nodes into a set of augmentation nodes. The output of the width model can be expressed as a weighted combination of feature nodes and enhancement nodes. The width learning system can use existing nodes to obtain new feature nodes or enhanced nodes through some kind of mapping such as random mapping, and dynamically add new feature nodes or enhanced nodes to achieve better learning effects. [0003] The width learning system has similar performance to the deep neural network, but compared with the deep neural network, the width learning system has low complexit...

AI Technical Summary

Problems solved by technology

The information content of a single configuration information is small, and it has a real-time switching effect. Therefore, it belongs to a dynamic reconstruction processor and is an ideal chip for realizing the operation of online learning neural networks. However, it is still blank to use CGRA to realize the operation of online learning neural networks.

Images