A paper titled “Deep Semantic Dictionary Learning for Multi-label Image Classification”, with Zhou Fengtao, a postgraduate student of 2019 in the research group led by Associate Professor Huang Sheng from the School of Big Data and Software Engineering of Chongqing University, as the first author, was accepted by Thirty-Fifth AAAI Conference on Artificial Intelligence recently. This Conference is a type-A conference recommended by China Computer Federation. Associate Professor Huang Sheng is the corresponding author of the paper. Xing Yun, a postgraduate student of 2018, is the co-author.
AAAI (The AAAI Conference on Artificial Intelligence) is an annual top international academic conference on artificial intelligence, and is a type-A conference recommended by China Computer Federation. Despite the outbreak of COVID-19, the number of papers submitted to AAAI this year is not less than that in previous years. A total of 9,034 papers have been received by AAAI-2021. Of these papers, 7,911 were reviewed and 1,692 accepted, with an acceptance rate of 21%. To cope with the pandemic, AAAI 2021 will be held virtually on line in February 2021.
Fig.1 Deep semantic dictionary learning model
Fig.2 Alternately parameter updating
The purpose of this project is to solve problems with task of multi-label image classification. In this paper, the problems with multi label image classification were considered as a dictionary learning task for the first time. Based on this, a new end-to-end deep semantic dictionary learning model (as shown in Fig. 1) was designed. The model can better mine the discriminant information of sample multi-label classification from label and semantic space. In addition, inspired by the iterative optimization of traditional dictionary learning, an alternative parameters update strategy (as shown in Fig. 2) was proposed for deep dictionary learning model. The experimental results showed that the algorithm proposed in this paper achieved exciting results in multi-label image classification task.
Chongqing University is the first completing organization of the paper. The research project has been supported by the National Natural Science Foundation of China, National Key R&D Program and funds allocated by the central government to colleges and universities.