[ad_1]
The diagram shows how new information is learned each time a data distribution is entered while preserving information learned in the past.Credit: Naoki Masuyama, Osaka Metropolitan University Tokyo
Advances in IoT (Internet of Things) technology have made it possible for us to acquire large amounts of diverse data easily and continuously. Artificial intelligence technology is attracting attention as a tool for utilizing this big data.
Conventional machine learning mainly deals with single-label classification problems in which data and corresponding phenomena or objects (label information) have a one-to-one relationship. However, in the real world, there is rarely a one-to-one relationship between data and label information.
Therefore, in recent years, multi-label classification problems that deal with data that has a one-to-many relationship between data and label information have attracted attention. For example, a single landscape photo can contain multiple labels for elements such as the sky, mountains, and clouds. In addition, in order to learn efficiently from continuously obtained big data, the ability to learn over time without destroying what has been learned in the past is also required.
A research group led by Associate Professor Naoki Masuyama and Professor Yusuke Nojima at the Graduate School of Informatics, Tokyo Metropolitan University, has developed a new method that combines the ability to classify data with multiple labels and the ability to continuously learn from the data. bottom. Numerical experiments on real-world multilabel datasets showed that the proposed method outperforms conventional methods.
The simplicity of this new algorithm makes it easy to devise advanced versions that can be integrated with other algorithms. The underlying clustering method groups data based on similarities between data entries and is therefore expected to be a useful tool for continuous big data preprocessing.
Additionally, the label information assigned to each cluster is continuously learned using a method based on a Bayesian approach. By continuously performing data learning and learning of label information corresponding to data separately, it achieves both high classification performance and continuous learning ability.
“Our method enables continuous learning from multi-labeled data, and we believe it has the functions necessary for artificial intelligence in the future big data society,” concludes Professor Masuyama.
published the research results. IEEE Transactions on Pattern Analysis and Machine Intelligence December 19, 2022.
For more information:
Naoki Masuyama et al., Multi-label classification by clustering based on adaptive resonance theory, IEEE Transactions on Pattern Analysis and Machine Intelligence (2022). DOI: 10.1109/TPAMI.2022.3230414
Provided by Tokyo Metropolitan University
Quote: using multi-label classification data (02/01/2023) obtained from https://techxplore.com/news/2023-02-method-ai-multilabel-classification.html on 02/01/2023 How to train AI
This document is subject to copyright. No part may be reproduced without written permission, except in fair trade for personal research or research purposes. Content is provided for informational purposes only.
[ad_2]
Source link
