Prof. Yu Qi
Zhejiang University, China
Yu Qi is a tenure-track professor and doctoral supervisor at the MOE Frontier Science Center for Brain Science and Brain-machine Integration, and the Affiliated Mental Health Center & Hangzhou Seventh People's Hospital, Zhejiang University School of Medicine, Hangzhou, China. Her research interests include brain-computer interface, artificial intelligence, and brain-like computing. She has proposed a series of innovative methods and systems for the dynamic modeling of brain information. She has published more than 30 academic papers and representative studies are published in NeurIPS, IJCAI, Trans. on BME, TNSRE. She is the young editor of Cyborg and Bionic Systems, the topic editor of Frontiers in Neuroscience, the program committee member of top artificial intelligence conferences NeurIPS, IJCAI, and ICML, and the reviewer of top journals such as TNNLS, TBME, and TNSRE.
Speech Title: "Dynamic Ensemble Bayesian Filter for Robust Control of a Human Brain-machine Interface"
Abstract: Brain-machine interfaces (BMIs) aim to provide direct brain control of devices such as prostheses and computer cursors, which have demonstrated great potential for motor restoration. One major limitation of current BMIs lies in the unstable performance due to the variability of neural signals, especially in online control, which seriously hinders the clinical availability of BMIs. We propose a dynamic ensemble Bayesian filter (DyEnsemble) to deal with the neural variability in online BMI control. Unlike most existing approaches using fixed models, DyEnsemble learns a pool of models that contains diverse abilities in describing the neural functions. In each time slot, it dynamically weights and assembles the models according to the neural signals in a Bayesian framework. In this way, DyEnsemble copes with variability in signals and improves the robustness of online control. Online BMI experiments with a human participant demonstrate that, compared with the velocity Kalman filter, DyEnsemble significantly improves the control accuracy (increases the success rate by 13.9% in the random target pursuit task) and robustness (performs more stably over different experiment days). DyEnsemble frames a novel and flexible dynamic decoding framework for robust BMIs, beneficial to various neural decoding applications.
Prof. Hongtao Lu
Shanghai Jiao Tong University, China
Hongtao Lu is a professor at the department of computer science and engineering, Shanghai Jiao Tong University. His research interests include machine learning, deep learning, computer vision and pattern recognition. He has authored or co-authored more than 100 research papers in journals and conference such as IEEE Transactions, Pattern Recognition, CVPR, ECCV, AAAI etc. His published papers have gotten more than 1300 citations from Web of Science, and more than 4000 citations from Google Scholar, his H-index is 37. He had served as PIs for dozens of projects from NSFC, Ministry of Science and Technology, Ministry of Education, Municipal Government of Shanghai, and industries. He has been continuously listed among the most cited researchers in computer science in China by Elsevier from years of 2014 to 2018. He also got several research awards from the government.
Speech Title: "Human Action Recognition from Skeleton and Video"
Abstract: In this talk, I will talk about two of our recent works about human action recognition. The first is about skeleton-based human action recognition. We proposed a novel Enhanced Discriminative Graph Convolutional Network (ED-GCN) based on the attention mechanism for skeleton-based action recognition. Discriminative channel-wise features are obtained by fusing the Squeeze and Excitation (SE) module to the GCN to selectively enhance the significant features and suppress the non-significant ones. The second is an efficient dual attention SlowFast networks for video action recognition. We proposed a cross-modality dual attention fusion module named CMDA to explicitly exchange spatial–temporal information between two pathways in two-stream SlowFast networks. We also proposed several two-stream efficient SlowFast networks based on well designed efficient 2D networks, such as GhostNet, ShuffleNetV2.
Assoc. Prof. Feng Liu
Shenzhen University, China
Feng Liu is currently an Associate Professor at School of Computer Science & Software Engineering, Shenzhen University. She obtained her B.Sc. and M.Sc. degrees both from Xidian University, Xi’an, Shaanxi, China. She received her Ph.D. degree in computer science from the Department of Computing at the Hong Kong Polytechnic University in 2014. Her research interests include pattern recognition and image processing, especially focus on their applications to fingerprints. Dr. Liu has published more than 40 papers at some international journals and conferences, including IEEE TIP, TFIS, TCYB, PR, ECCV, AAAI, etc., and participated in many research projects either as principal investigators or as primary researchers. She is a reviewer for many renowned field journals and conferences and a member of the IEEE.
Speech Title: "Research on Optical Coherence Tomography (OCT) –based Fingerprints"
Abstract: Optical coherence tomography (OCT) is a new non-invasive, high-resolution and three-dimensional imaging technology developed in the 1990s. It has been maturely applied in Ophthalmology, dermatology and other fields. However, OCT-based fingerprint research began in 2006. Compared with traditional fingerprints (two-dimensional surface fingerprints, high-resolution fingerprints, and three-dimensional fingerprints), OCT-based fingerprints: 1) can overcome the low-quality image problems caused by the skin status of surface fingerprints (such as dry, wet, stains, etc.), which benefits in improving the recognition accuracy; 2) It can capture sweat glands / pores and other features that can be extracted only on high-resolution fingerprint images, so as to realize alive detection; 3) In essence, it is a kind of three-dimensional fingerprint data containing rich multi-level fingerprint information, which can easily identify false fingerprints without internal fingerprint structure, so as to achieve the purpose of anti-spoofing. Therefore, OCT opens up a new field for fingerprint recognition. This talk will mainly focus on the anti-spoofing technology in OCT-based fingerprints. Firstly, we will review current researches of anti-spoofing technology on OCT-based fingerprint, analyze the existing problems, and report some progress and achievements we have made; Finally, the further direction and expectation of OCT-based fingerprint image is summarized.
Kok Chin Chai
Dr. Kok Chin Chai is a currently a technopreneur who received his PhD degree and completed his postdoctoral research in 2016 and 2017, respectively. He was recognized as professional technologist by Malaysia Board of Technology in 2021. He has developed novel AI models over 8 years, resulting in over 10 scientific publications in collaboration with team members at Deakin University and Universiti Malaysia Sarawak. He managed several industrial-university research projects in Malaysia and Taiwan. By aggregating his research skills and business experiences, he has founded FLINKEN (Automated Manufacturing Factory) and NEUON (AI consultancy and solutions provider). Currently, he is the CEO of NEUON.
Speech Title: "Research To Commercialisation: Challenges and Opportunities"
Abstract: With the background in both AI research and business experiences, Dr Chai founded NEUON in 2018 specialised in AI Computer Vision Solutions. Via the sharing session, he will share his experiences while designing and deploying the AI computer vision technology in industry. The real implementation is always challenging with much more design factors that may not be focused and addressed in academic research. Dr Chai will share his experiences when dealing with their current use cases on road asset management, manufacturing as well as plant science. Furthermore, he will share on their existing initiative and collaboration with research universities and institutes to bring their researches to meet the market needs. It is a great opportunity for all researchers ,academicians, professionals and also entrepreneurs to meet and explore the potentials together.
Prof. Fan Liu
Hohai University, China
Fan Liu is currently a professor of Hohai University. He received his B.S. degree of networking and Ph.D. degree of technology for computer applications from Nanjing University of Science and Technology (NUST) in 2009 and 2015. From September 2008 to December 2008, he studied at Ajou University in South Korea. From February 2014 to May 2014, he worked at Microsoft Research Asia. His research interests include computer vision, pattern recognition, and machine learning. Dr. Liu also serves as a reviewer of IEEE TNNLS, IEEE TKDE, ACM TIST, Information Sciences, Neurocomputing, Pattern Analysis and Application.
Speech Title: "Music-driven Conducting Motion Generation"
Abstract: Although recent studies have successfully generated motion for singers, dancers, and musicians, few have explored motion generation for orchestral conductors. In order to verify the feasibility of applying deep learning models to this task, we construct a large-scale dataset ConductorMotion100, which consists of an unprecedented 100 hours of conducting motion data. Based on the constructed ConductorMotion100 dataset, we designed a novel Music Motion Synchronized Generative Adversarial Network (M2S-GAN). The M2S-GAN is a cross-modal generative network comprising four components: 1) a music encoder that encodes the music signal; 2) a generator that generates conducting motion from the music codes; 3) a motion encoder that encodes the motion; 4) a discriminator that differentiates the real and generated motions. These four components respectively imitate four key aspects of human conductors: understanding music, interpreting music, precision and elegance. The music and motion encoders are first jointly trained by a self-supervised contrastive loss, and can thus help to facilitate the music motion synchronization during the following adversarial learning process. Through visualization, we show that our approach can generate plausible, diverse, and music-synchronized conducting motion. To facilitate future research, the dataset is made public via the 1st Prospective Cup Meta-Intelligent Data Challenge.
Maejo University, Thailand
Grienggrai Rajchakit was born in 1981. He received the B.S. (Mathematics), Thammasat University, Bangkok, Thailand, in 2003. He received the M.S. (Applied Mathematics), Chiangmai University, Chiangmai, Thailand, in 2005. He was awarded a Ph.D. (Applied Mathematics), King Mongkut’s University of Technology Thonburi, Bangkok, Thailand, in the field of mathematics with specialized area of stability and control of neural networks. Currently, he is working as a lecturer at Department of Mathematics, Faculty of Science, Maejo University, Chiang Mai, Thailand. He was the recipient of Thailand Frontier author Award by Thomson Reuters Web of Science in the year 2016 and TRF-OHEC-Scopus Researcher Awards by The Thailand Research Fund (TRF), Office of the Higher Education Commission (OHEC) and Scopus in the year 2016. His research interests are complex-valued NNs, complex dynamical networks, control theory, stability analysis, sampled-data control, multi-agent systems, T-S fuzzy theory, and cryptography, etc. Dr. Grienggrai Rajchakit serves as a reviewer for various SCI journals. He has authored and co-authored more than 149 research articles in various SCI journals.
Speech Title: "An Application of Neural Networks for Traffic Problems"
Abstract: Traffic congestion is a thorny issue to many large and medium-sized cities, posing a serious threat to sustainable urban development. Recently, intelligent traffic systems (ITS) have emerged as an effective tool to mitigate urban congestion. The key to ITS lies in the accurate forecast of traffic flow. However, the existing forecast methods of traffic flow cannot adapt to the stochasticity and sheer length of traffic flow time series. To solve the problem, this paper relies on deep learning (DL) to forecast traffic flow through time series analysis. The authors developed a traffic flow forecast model based on the bidirectional Long Short-Term Memory Recurrent Neural Network (BLSTM-RNNs) with delay. The proposed model was compared with two classic forecast models, namely, the autoregressive integrated moving average (ARIMA) model and the Long Short-Term Memory Recurrent Neural Network (LSTM-RNNS) model, through long-term traffic flow forecast experiments, using an actual traffic flow time series from Open ITS. The experimental results show that the proposed BLSTM-RNNs network outperformed the classic models in prediction accuracy. Our research discloses the dynamic evolution law of traffic flow and facilitates the decision-making of traffic management.
Prof. Yaohua Wang
National University of Defense Technology, China
Yaohua Wang is a Professor and PhD advisor in the School of Computer Science of National University of Defense Technology. He was a Visiting Professor of ETHZ (2016-2017).His current research interests include computer architecture, compiler, and memory subsystem. He has published in prestigious conferences (such as ISCA(2018,2021), MICRO(2018, 2020), HPCA(2013), CVPR(2020), and DAC(2022)) and top-tier journals (such as IEEE TC(2018), IEEE TPDS(2021), and IEEE TCAD(2020)). His research received Best Paper Award at IEEE CAL 2012.
Speech Title: "Feeding Computation Units as Needed"
Abstract: FT-Matrix is a processor architecture proposed by National University of Defense Technology (NUDT). In this talk, we will mainly describe how to orchestrate instruction flows and suppoly data for computation units efficiently in FT-Matrix, and brief the corresponding mechanisms including instruction shuffle, multiple SIMD multiple data, charge level aware look ahead partial restoration. We hope our presentation can provide inspirations for computor architecture design and attract a little bit attention on FT-Matrix processor.
Dr. Van Thanh Huynh
Deakin University, Australia
Dr. Van Thanh Huynh is an early career researcher and lecturer at Deakin University. He received his PhD degree in robotics and control systems in 2019. Leveraging the nexus of applied mathematics and engineering, he is inspired to develop enabling technologies and engineering methodologies that assist in efficiently and effectively solving robotic, interdisciplinary, and science-driven problems. He has published in prestigious journals, including SIAM Journal on Control and Optimization, IEEE Transactions on Circuits and Systems, IEEE Systems Journal, etc. His research is recognised with Deakin University awards. He has successfully attracted competitive funding from both industry and not-for-profit organisations.
Speech Title: "A Hybrid Attention-Based CNN LSTM Model for Traffic Flow Prediction"
Abstract: Intelligent transports systems (ITS) of smart cities have been developed to address the traffic management problem caused by a growing number of urban vehicles and increasing traffic congestion. Giving accurate traffic flow prediction plays a pivotal role in a well-functioning ITS. However, such prediction and associated learning of nonlinear spatial-temporal dynamics are challenging due to existence of nonlinearity and stochasticity and has attracted attention of many researchers. In this talk, a discussion on a hybrid, deep-learning model based on attention CNN-LSTM is presented to solve the traffic flow prediction problem. Comparison of the proposed method with counterparts is also discussed, offering some insights into pros and cons of prevalent deep-learning methods in solving the traffic prediction issue.