TRAINING SAMPLE DATA AUGMENTATION METHOD BASED ON VARIATIONAL AUTOENCODER, STORAGE MEDIUM AND COMPUTER DEVICE

摘要

A training sample data augmentation method based on a variational autoencoder, a storage medium and a computer device, related to the technical field of big data, the method comprising: obtaining an original sample (S102); inputting the original sample into an encoder of a variational autoencoder, the encoder of the variational autoencoder comprising two neural networks (S104), the two neural networks respectively outputting μ and σ, μ and σ each comprising a function of the original sample; in accordance with the square of μ and σ, i.e., σ², generating random numbers having a corresponding Gaussian distribution (S106); performing random sampling on a standard normal distribution, obtaining a sampled value ε, and, in accordance with the sampled value ε and the random numbers having a Gaussian distribution, determining a sampling variable Z (S108); inputting the sampling variable Z into a decoder of the variational autoencoder, the decoder of the variational autoencoder decoding same and then outputting a sample similar to the original sample, and using the similar sample as an augmentation sample (S110). The method is able to solve the problems in the prior art that manually augmenting sample data is time-intensive, laborious and low-efficiency.