Advances in AI technology have allowed us to easily and continuously acquire large amounts of diverse data. Artificial intelligence technology is gaining attention as a tool to harness this big data. Conventional machine learning mainly deals with single-label classification problems, in which data and corresponding phenomena or objects (label information) are in a one-to-one relationship. However, in the real world, the data and information on the label rarely have a one-to-one relationship.
Therefore, in recent years, attention has been focused on the multi-label classification problem, which deals with data that has a one-to-many relationship between label data and information. For example, a single landscape photo may contain several tags for features such as sky, mountains, and clouds. In addition, learning effectively from big data that is constantly being acquired also requires the ability to learn over time without destroying what we have learned before.
A research group led by Associate Professor Naoki Masuyama and Professor Yusuke Nojima of the Osaka Metropolitan University Graduate School of Informatics has developed a new method that combines classification performance for multi-label data with the ability to continuously learn with the data. Numerical experiments on real-world multi-label datasets have shown that the proposed method outperforms conventional methods.
The simplicity of this new algorithm makes it easy to design an evolved version that can be integrated with other algorithms. Since the underlying clustering method groups data based on the similarity between data records, it is expected to be a useful tool for continuous preprocessing of big data. In addition, the label information assigned to each cluster is continuously learned using a method based on a Bayesian approach. By learning the data and learning the label information corresponding to the data separately and continuously, both high classification performance and continuous learning ability are achieved.
“We believe our method is able to continuously learn from multi-label data and has the capabilities needed for artificial intelligence in the future big data society,” Professor Masuyama concluded.
