Image classification method based on random calculation Bayesian neural network error injection

Abstract

The invention discloses an image classification method based on random calculation Bayesian neural network error injection. The method comprises the following steps: S1, scaling input data, weight parameters and bias parameters; S2, converting the scaled floating point input data, the scaled floating point weight parameter and the scaled floating point bias parameter into a random bit stream form through a forward conversion circuit; S3, building a random calculation neuron structure of the Bayesian neural network; S4, calculating the scaling of each neuron node, and performing forward reasoning; S5, converting the output bit stream into a floating point form, and obtaining an output result of single reasoning; and S6, repeating the steps S4-S5, taking a mean value, and taking the mean value as a classification result. According to the image classification method based on Bayesian neural network error injection, inherent noise characteristics are calculated randomly, an additional error injection circuit does not need to be introduced, and unification of calculation and error injection in the Bayesian neural network reasoning process is achieved.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to an image classification method based on random calculation Bayesian neural network error injection. Background technique [0002] The weights and thresholds of Bayesian Neural Networks (BNNs) are expressed in the form of random distribution, which is a random number subject to the posterior random distribution, that is, the Bayesian inference algorithm is introduced into the neural network model. The essence of its algorithm lies in: repeating the forward propagation with different random sampling parameters to determine the final output. Considering the implementation at the hardware level, additional error generation circuits are required to inject errors to satisfy the prediction distribution that conforms to the randomness of each calculation of the network. This is a huge challenge for traditional CMOS circuits. [0003] For the hardware implementation...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to solve the problem of image classification, and pro

Images