Target task training method and system

Abstract

The invention discloses a target task training method and system. The method comprises the following steps of: obtaining a distinguishing threshold value corresponding to a target task; determining sample correctness of a plurality of attribute samples in the target task according to the distinguishing threshold value; determining relevance between the target task and other tasks by utilizing thesample correctness, wherein the other tasks are to-be-trained tasks or trained tasks except the target task; and training the target task according to the relevance. According to the method and system, a plurality of tasks can be grouped to realize joint training, so that the workloads and algorithm consumed time of training and application are decreased, and the training effect is improved.

Description

technical field [0001] The invention relates to the field of computer application technology, in particular to a target task training method and system. Background technique [0002] With the development of technology, deep learning has gradually become the mainstream method in the field of machine learning and computer vision. It is widely used in many fields such as image classification, target detection, face detection, face recognition, voice recognition and natural language processing. Compared with traditional methods, it has been greatly improved. [0003] The most intuitive way to adopt the deep learning method is to train each task separately. The training workload, training time, and algorithm time consumption in actual application are all related to the number of tasks, which increases the training cycle. In addition, tasks are related to each other, and separate training cannot effectively utilize the association between tasks of each attribute. For example, th...

AI Technical Summary

Problems solved by technology

However, this depends on the experience and preference of the implementer. If there is not enough relevant experience in advance, when the number of tasks is large (such as 40 tasks), the combination space between tasks will be very large, and it is impossible to traverse and try to find optimal combination

Therefore, if each task is trained separately, in actual application, not only the algorithm prediction time is linearly related to the number of tasks, but also increases the training time and preprocessing workload

Images