Neural network model training data processing method, device, electronic equipment and storage medium

Abstract

The invention provides a neural network model training data processing method and device, equipment and a storage medium, and the method comprises the steps: carrying out the rotation self-supervision processing of a first training data set, so as to form a corresponding second training data set; determining an initial training data set corresponding to the second training data set; according to the second training data set and the initial training data set, gradient parameters matched with the initial training data set are determined through the first training data processing network and the second training data processing network; and updating the initial training data set, and determining a target training data set matched with the target neural network model. Therefore, on the premise of reducing the total amount of the training data, the accuracy of training the neural network model by the training data with a small scale can be stably improved, the generalization ability of the target neural network model is enhanced, the trained target neural network model can be conveniently deployed in the mobile terminal, and large-scale application of the target neural network model is realized.

Description

Technical field [0001] The present invention relates to machine learning techniques, and more particularly to training data processing, devices, electronic devices, and storage media for neural network models. Background technique [0002] In artificial intelligence technology, the depth neural network has achieved very good performance in many computer visual tasks. In general, the larger number of training data is, the better the performance of the trained neural network model. However, deploying such larger training data on resource-limited embedded systems is very difficult, and the performance of neural networks obtained directly through larger training data is much lower than the performance of larger networks. Related technologies can implement compression of data by data distillation when the training data is processed, but the distillation of the non-standard note data is not realized, so that the processing of training data is complex, and the number of training data is...

AI Technical Summary

Problems solved by technology

However, it is very difficult to deploy such large-scale training data on resource-constrained embedded systems, and the performance of neural networks obtained directly from small-scale training data is much lower than that of larger-scale networks.

Related technologies can achieve data compression through data distillation when processing training data, but distillation of unlabeled data cannot be achieved, which makes the processing of training data complicated, and at the same time, the large amount of training data makes the training of neural network models difficult. The cost of hardware and data storage is too high, which is not conducive to large-scale promotion and use

Images