Recently, learning based hashing techniques have attracted broad research interests because they can support efficient storage and retrieval for high-dimensional data such as images, videos, documents, etc. However, a major difficulty of learning to hash lies in handling the discrete constraints imposed on the pursued hash codes, which typically makes hash optimizations very challenging (NPhard in general). The widely adopted relaxation scheme simplifies the optimization, however, tends to produce less effective codes.
In this project, we investigate how to directly learn binary codes via discrete optimization, namely discrete hashing. Discrete hashing has been shown to significantly boost the performance of hashing algorithms (e.g., Supervised Discrete Hashing) in large-scale similarity search.
Most binary coding and hashing algorithms were designed to deal with Approximate Nearest Neighbor (ANN) search. Studies have rarely been dedicated to Maximum Inner Product Search (MIPS), which actually plays a critical role in various vision and learning applications.
In this project, we investigate learning binary codes to exclusively handle the MIPS problem. The core problem is, the similaritis between two sets of data should be revealed by the inner products of their corresponding binary codes. We study this problem by proposing the Asymmetric Inner-product Bianry Coding algorithm for large-scale MIPS search, and the Discrete Collaborative Filtering algorithm for recommendation systems.
| Robust regression for practical visual recognition | Youth Project, National Science Fundation of China | Chief investigator | 2016~2018 |
| Robust regression for practical face recognition | Innovation Funds, Key Laboratory of Image and Video Understanding for Social Safety, NJUST | Chief investigator | 2015~2016 |
| Robust regression for pedestrian gait recognition | Fundamental Research Funds for the Central Universities, UESTC | Chief investigator | 2015~2016 |