Multi-task network model training method and system based on adaptive task weight

Abstract

The invention relates to a multi-task network model training method and system based on adaptive task weights. According to the method, a sharing mode is learned through a strategy specific to tasks, the strategy autonomously selects which layers are executed in a multi-task network, and weights matched with the tasks can be searched at the same time, so that the model is better trained. According to the method, the multi-task network model is reconstructed based on ResNet, the learning strategy is effectively optimized according to the image in the data set in the training process, and the oneness of the multi-task model is overcome while the task index is improved. According to the method, a multi-task loss function suitable for regression and classification tasks is deduced based on probability theory maximum likelihood estimation, the task weight can be automatically adjusted in the training process so as to better improve the model performance, and the problem that the task weight is not flexible is solved.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a multi-task network model training method and system based on adaptive task weights. Background technique [0002] The task of scene understanding is an important problem in the field of computer vision, involving the joint learning of various regression and classification tasks at different units and scales, including semantic labels that describe the identities of different scene parts and surface normals that describe the physical geometry. These are often represented as pixmaps, which contain values ​​or labels for each pixel, such as a map containing the semantic labels of objects at each pixel or the vector coordinates of surface normals. [0003] Multi-task learning (MTL) is a method of learning multiple tasks at the same time. It can use the shared knowledge learned from each task to help other tasks learn. Compared with single-task learning, it is mo...

AI Technical Summary

Problems solved by technology

[0004] There are two serious problems in the current multi-task learning research: (1) Most of the existing multi-task learning methods use hard parameter sharing and soft parameter sharing mechanisms to achieve feature sharing. In these two mechanisms, The model of each task is fixed. However, in social life, there are various tasks, and of course different sharing models are required. It is difficult to meet the actual needs only by the fixed mechanism; (2) due to the existence of multiple Therefore, the balance between tasks is particularly important. Most of the existing research uses weighted linear sums for simple processing, that is, manually selecting the weight of each task. Obviously, this is very inflexible. If an inappropriate weight is selected, it will lead to different The learning effect of the task is quite different

Images