|Conference Keynote Speakers|
Prof. Yanjiao Chen
Zhejiang University, China
Biography: Yanjiao Chen received her B.E. degree in Electronic Engineering from Tsinghua University in 2010 and Ph.D. degree in Computer Science and Engineering from Hong Kong University of Science and Technology in 2015. She is currently a Bairen Researcher in the College of Electrical Engineering, Zhejiang University, China. Her research interests include ML security, AI in networking, and mobile sensing. Yanjiao has published papers in ACM CCS, IEEE INFOCOM, ICDCS, etc. Yanjiao has served on the editorial board of IEEE WCL and served as TPC member in IEEE INFOCOM, NDSS, ICNP, etc.
Speech Title: Attention-based QoE-aware Evasive Backdoor Attacks
Abstract: Deep neural networks have achieved remarkable success on a variety of mission-critical tasks. However, recent studies show that deep neural networks are vulnerable to backdoor attacks, where the attacker releases backdoored models that behave normally on benign samples but misclassify any trigger-imposed samples to a target label. Unlike adversarial examples, backdoor attacks manipulate both the inputs and the model, perturbing samples with the trigger and injecting backdoors into the model. In this talk, we will introduce a novel attention-based evasive backdoor attack, which features an attention-based trigger mask determination framework and a Quality-of-Experience (QoE) aware loss function. The attack is shown to be evasive to state-of-the-art defense methods, including model pruning, NAD, STRIP, NC, and MNTD.
Assoc. Prof. Pavel Loskot
Zhejiang University-University of Illinois at Urbana-Champaign Institute (ZJUI), China
Biography: Pavel Loskot joined the ZJU-UIUC Institute in January 2021 as the Associate Professor after being nearly 14 years with Swansea University in the UK. He received his PhD degree in Wireless Communications from the University of Alberta in Canada, and the MSc and BSc degrees in Radioelectronics and Biomedical Electronics, respectively, from the Czech Technical University of Prague in the Czech Republic. He is the Senior Member of the IEEE, Fellow of the Higher Education Academy in the UK, and the Recognized Research Supervisor of the UK Council for Graduate Education. His current research interest focuses on problems involving statistical signal processing and importing methods from Telecommunication Engineering and Computer Science to other disciplines in order to improve the efficiency and the information power of system modeling and analysis.
Speech Title: Common Transformations and Decompositions in Complex Signal Models
Abstract: Signal processing becomes significantly more complicated when the assumed signal models are non-linear and/or high-dimensional. In such a case, the underlying model complexity can be resolved by the appropriate model transformations and/or model decompositions. In this talk, we will review most common transformation and decomposition methods used in analyzing linear and non-linear systems defined or represented in multiple dimensions including Taylor, Fourier and Sobol expansions and other factorizations, signal folding, kernel trick, Volterra series, Koopman operator. The objective is to bring a better understanding about coping with complexity in signal processing, and also point out an intimate relationship between model non-linearity and the increase in model dimensionality.
Assoc. Prof. Kuo-Kun TsengHarbin Institute of Technology, Shenzhen/ Computer Science and Technology
Biography: Kuo-Kun Tseng is a Shenzhen peacock B class talent, and graduated in 2006 China Taiwan National Chiao Tung University, computer information and engineering doctorate. He has many teaching experience. Since 2010 to now, he is an associate professor of Harbin Institute of Technology (Shenzhen). For professional, from 1994 to 2002, he has many years of research and development experiences, long-term engaged research in the research area of Deep Learning Innovative Thinking and Deep Learning Architecture. The current research results, published more than 75 articles, of which about 35 is a high impact factor SCI/ACM/IEEE series or famous journals, and more than 40 invention patents. Currently, Recently, he has opened two university courses "Introduction to Deep Learning Innovative Thinking" and "Deep Learning Architecture" for a long time.
Speech Title: Brain-Inspired Few-Shot Class-Incremental Learning with Multi-modal Biometric Application
Abstract:Not as good as the human brain, Deep Neural Networks (DNNs) tend to overfit when the training sample is insufficient, and forget the learned old knowledge when learning new. Inspired by the memory replay mechanism and the high degree of generalization of human brains, we propose a new method named BIFSCIL to address this Few-Shot Class-Incremental Learning (FSCIL) problem in this article. First, we introduce an external memory module to save the feature data of all the old classes that have been learned, so as to reduce forgetting by replaying the old knowledge when learning new. Meanwhile, this external memory module can also naturally act as a Nearest-Class-Mean classifier to replace the traditional Softmax classifier. Second, we let the model learn a set of generalized parameters through the meta-learning algorithm. We combine meta-learning with the memory replay mechanism to generate a large number of few-shot pseudo-tasks composed of different types of class-combinations to perform continuous meta-training, enabling the model to understand how to learn tasks rather than just learning a task.
Furthermore, a advanced applied architecture is designed to improve biometric systems as Multi-modal Biometric Few Shot Learning (MBFSL) model. This approach is an incremental learning strategy that uses similarity scores to make predictions on sparse updating data. The proposed application is implemented combining recognition of face, palmprint, voice and signature that can be easy acquired from mobile phone or pad. And it should be the first proposed multi-modal biometric application of incremental learning.
Assoc. Prof. Shiling Zhang
State Grid Chongqing Electric Power Company Chongqing Electric Power Research Institute
Biography: Zhang Shiling, senior engineer, doctor of engineering. He has been engaged in scientific research and production of high voltage and insulation technology and physical and chemical detection technology for a long time. The development of UHV dry-type converter transformer bushing and SF6 gas insulated through wall bushing has been applied to the construction of UHV AC and DC projects in China. Presided over and completed the GIS fault detection sensing technology and system, won the excellent innovation achievement award of the international innovation and entrepreneurship Expo, and was awarded the title of excellent scientific and technological worker by Chongqing Institute of electrical engineering.
As the first author, he has published more than 90 SCI/EI search papers in domestic and foreign journals and international academic conferences, 19 Chinese Core Journals of Peking University, won 9 provincial and ministerial awards such as the first prize of Chongqing scientific and technological progress and the special first prize of China Water Conservancy and power quality management Association, authorized 1 international invention patent, 20 national invention patents and utility models, 18 software copyrights, and more than 20 reports of international and domestic conferences, As the project leader, he presided over 2 provincial and ministerial projects at the basic frontier and 3 science and technology projects at the headquarters of State Grid Corporation of China.