A reconfigurable deep belief network implementation system

Abstract

The invention discloses a reconfigurable deep belief network implementation system. The system comprises a control unit used for controlling the operation process of a deep belief network algorithm; the data temporary storage unit is used for storing and calculating required input layer information, weights, offsets and output layer results; the restricted Boltzmann machine is used for calculatingthe activation probability of each layer of nodes in the deep belief network so as to determine the activation state; and the parameter updating unit is used for updating the weight and bias of eachlayer of nodes in the deep belief network, and is only activated in the training algorithm. The implementation system has the beneficial effects that two algorithms of training and reasoning can be supported by multiplexing the restricted Boltzmann machine and controlling the state jump of the algorithm, the hardware resource cost of the algorithm is reduced, the performance of the algorithm is ensured, and the implementation system is suitable for various artificial intelligence scenes.

Description

technical field [0001] The invention relates to the field of artificial intelligence algorithms, in particular to a reconfigurable deep belief network realization system. Background technique [0002] With the advancement of machine learning and the advent of deep learning, several tools and graphical representations are gradually used to correlate large amounts of data. Deep Belief Networks (Deep Belief Networks, DBN) was proposed by Hinton et al. in 2006. It is essentially a graph representation network with generative capabilities, that is, it generates all possible values ​​​​of the current example. As a fusion of probability and statistics with machine learning and neural networks, DBN consists of multiple layers with values, where there are relationships between layers but not between values. The main goal of the deep belief network is to help the system classify data into different categories, and it is widely used in application scenarios such as pattern recognition...

AI Technical Summary

Problems solved by technology

However, most of the existing deep belief network algorithms focus on the application level, and few efficient and flexible deep belief network implementation systems and architectures have been proposed, which cannot balance performance and flexibility, and cannot well meet the needs of practical applications.

Images