A training data selection method for machine learning

Abstract

The invention discloses a data selection method for machine learning, comprising the following steps: step 1, selecting a machine learning model of the data to be selected, and obtaining a training data set corresponding to the machine learning model; step 2, randomly selecting from the training data set Select a data subset as the policy training data set, apply the policy training data set to the machine learning model for several rounds of training through deep reinforcement learning, and determine the data selection strategy that matches the machine learning model according to the training results; Step 3, by determining The data selection strategy selects the input data of the machine learning model in batches, and uses the selected data for the training of the machine learning model. The method can use the obtained optimal training data selection strategy for the current machine learning model to select the training data of the machine learning model and improve the performance of the machine learning model.

Description

Technical field [0001] The present invention relates to the field of machine learning, and more particularly to a training data selection method for machine learning. Background technique [0002] In recent years, machine learning, especially based on deep learning technology based on large-scale deep neural networks, has developed rapidly, and has been applied in various aspects of life. With the increasing popularity of deep learning, data selection in machine learning has become an increasingly concern. How to automatically select data, improve the performance of deep learning models, becoming a current urgent need. [0003] At present, there have been many methods in the field of machine learning data selection, such as in accordance with the "difficulty", "curriculum), which is sequentially trained from low to high, and is conducive to the training process of the model. In addition, the loss function size of self-study data is the measurement of "difficulty". In the self-lea...

AI Technical Summary

Problems solved by technology

[0004] However, the above-mentioned existing data selection strategies are artificially defined heuristic strategies, which are highly specific. Since different machine learning tasks usually have different data distribution and model characteristics, these rules are often used in different machine learning tasks. difficult to generalize

Images