[2020-03-06] Dr. Chu-Song Chen, Academia Sinica, "Construction of compact features for efficient image retrieval and continual learning of deep models"

Poster:Post date:2020-02-27
Title: Construction of compact features for efficient image retrieval and continual learning of deep models
Date: 2020-03-06 2:20pm-3:30pm
Location: R103, CSIE
Speaker: Dr. Chu-Song Chen, Academia Sinica
Hosted by: Prof. Yung-Yu Chuang


This talk includes two parts. In the first part, I will introduce a simple yet effective hash codes construction approach for fast retrieval. Adaptive labeling hash is a hash function learning approach via neural networks. In this approach, class label representations are adaptable during the network training. We express the labels as hypercube vertices in a K-dimensional space, and both the network weights and class label representations are updated in the learning process. As the label representations are explored from data, semantically similar categories will be assigned with the label representations that are close to each other in terms of Hamming distance in the label space. In the second part, I will survey some important methodologies that address the continual lifelong learning problem on deep models, including the approaches of gradient regularization, memory replay, and dynamic architecture. I will analyze their pros and cons, and discuss why they suffer from some limitations such as gradual forgetting, long training time, and structural redundancy. Then, I will introduce a recent study leveraging the principle of deep model compression with weight pruning, which can yield the models that are un-forgetting, sustainable, and extensible with compactness. The knowledge accumulated through learning previous tasks is helpful to adapt to a better model for the new tasks in the approach.

Dr. Chu-Song Chen is a research fellow/professor of the Institute of Information Science (IIS), Academia Sinica, and an adjunct professor of the Graduate Institute of Networking and Multimedia (GINM), National Taiwan University. He is on the governing board of the Image Processing and Pattern Recognition (IPPR) Society. He is also with AINTU (Most Joint Research Center for AI Technology and All Vista Healthcare) since 2018. His research interests include deep learning, computer vision, multimedia and big data. Currently, he serves as an associate editor of the journals Pattern Recognition (Elsevier) and Machine Vision & Applications (Springer). His recent studies focus on merging and lifelong learning of deep models, medical image analysis, AI in emergency medicine, as well as large-scale multimedia retrieval.
Last modification time:2020-02-27 AM 9:01

cron web_use_log