Neural network training method and device, and computer readable storage medium

Abstract

The embodiment of the invention provides a neural network training method and device and a computer readable storage medium, and the method comprises the steps of carrying out the iterative training of an initial network through employing a sample of each task in a first data set based on initial meta-knowledge, the initial task knowledge of each task, and a task loss function, and obtaining the task knowledge of each task, wherein the first data set is a source domain data set containing an initial task category; based on the task knowledge, the initial meta-knowledge and the meta-loss function of each task, performing iterative training on the initial network by adopting samples corresponding to various tasks in the first data set to obtain optimal meta-knowledge; and based on the optimal meta-knowledge, the task knowledge of each task and a task loss function, performing iterative training on a network corresponding to the optimal meta-knowledge for a sample of each task in the second data set to obtain the optimal task knowledge of each task and a target network corresponding to the optimal task knowledge. According to the invention, generalization of training of a small number of samples is improved.

Description

technical field [0001] The present disclosure relates to computer vision technology, in particular to a neural network training method and device, and a computer-readable storage medium. Background technique [0002] Few-shot learning is one of the important emerging research areas in computer vision. Currently, many CNN-based object detectors have achieved great success due to the rapid development of Convolutional Neural Networks (CNNs). Among them, few-sample learning aims to develop the ability of deep learning models in few-sample scenarios. The main few-shot learning methods include: a meta-learning model based on Episode training method and a transfer learning method based on Pre-train finetune. [0003] However, the current two types of few-shot learning methods are very similar in the optimization process, but they have their own corresponding implementation methods, which are not compatible with each other and have limitations. Contents of the invention [000...

AI Technical Summary

Problems solved by technology

[0003] However, the current two types of few-sample learning methods are very similar in the optimization process, but each has a corresponding implementation method, which is not compatible with each other and has limitations.

Images