Duan, Lingyu is a professor in the School of Computer Science, and has served as the Deputy Director of the Rapid-Rich Object Search Laboratory (ROSE), a joint lab between Nanyang Technological University (NTU), Singapore and Peking University (PKU), China since 2012. He is also the Deputy Director of PKU-NTU Joint Research Institute (JRI) since 2015. He was a Visiting Associate Professor with the school of EEE, NTU, from 30 Nov. 2013 - 30 Nov. 2015.He obtained his B.Sc. from Dalian University of Technology (DUT) in 1992 ,his M.Sc. degree in automation from the University of Science and Technology of China, China, in 1999, his M.Sc. degree in computer science from the National University of Singapore, Singapore, in 2002, and the Ph.D. degree in information technology from The University of Newcastle, Australia, in 2008, respectively. His research interests include mobile visual search, visual feature coding, and video analytics.
Dr. DUAN has published more than 100 research papers in international journals and conferences, such as IEEE T-PAMI, IJCV, IEEE T-IP, IEEE T-MM, IEEE MM, IEEE ICCV, IEEE CVPR, IJCAI, AAAI, ACM MM, and DCC. He has served in the Technical Program Committee of various international conferences including ACM Multimedia, ACM ICMR, IEEE ICME, etc. He is serving as an Ad-Hoc Group Co-Chair of MPEG Compact Descriptor for Video Analytics (CDVA) ,and was a Co-Editor of MPEG Compact Descriptor for Visual Search (CDVS) Standard (ISO/IEC 15938-13). He was awarded Ministry of Education Technology Invention Award (First Prize) (2016), and National Information Technology Standardization Technical Committee "Standardization Work Outstanding Person" (2015).
Dr. DUAN has more than ten research grants including NSFC, 863 project, etc. His recent research achievements are summarized as follows:
1) MPEG CDVS Standardization: As a Co-Editor, he completed the standardization of CDVS (ISO/IEC 15938-13). The technical contribution to the standard is more than half. The primary goal of CDVS is to provide a standardized bitstream syntax of high performance compact visual descriptors to enable interoperability in visual search applications. Remarkable improvements were achieved in reducing the size of feature data and the computation cost in the feature extraction process.
2) Approximate Nearest Neighbor (ANN) Search: ANN search is an important topic in machine learning, computer vision and information retrieval. This research work presented a group of novel hashing approaches Minimal Reconstruction Bias Hashing (MRH) , Hamming Compatible Quantization (HCQ), and Affinity Preserving Quantization (APQ). Through optimizing quantization and adaptively balancing the information loss between projection and quantization, these approaches have significantly advanced the state-of-the-art hashing performance.
3) Representation Learning: The performance of machine learning methods is heavily dependent on the choice of data representation (or features). The research works have leveraged Convolution Neural Networks (CNNs) and triplet network learning to derive a variety of deep learning based image and video representation for fine-grained object recognition and retrieval, compact deep invariant representation for video matching and retrieval, and object detection, Those work has contributed to the ongoing MPEG CDVA standardization, and significantly improved the state-of-the-art performance over a variety of benchmarks.
