Alumni Homecoming Workshop
Sharing by Prof. Xiaolong Jin
Event Knowledge Acquisition and Reasoning
In recent years, the continuous accumulation of massive data and the rapid development of technologies such as big data and deep learning have brought new opportunities to knowledge engineering. As an intuitive and flexible way of knowledge expression, the new generation of knowledge engineering technology represented by knowledge graph has attracted extensive attention from all walks of life. Knowledge graph automatically extracted and constructed from massive data has played an important role in the fields of vertical search, intelligent Q&A, automatic customer service and so on. However, knowledge graph usually contains binary and static factual knowledge. Because of the typical features of being n-ary, temporal and procedural, events have become a kind of special knowledge closer to business and more valuable in many fields such as finance, medical treatment and industrial control. This report will introduce our latest research results in event knowledge extraction, fusion, reasoning and prediction.
Xiaolong Jin is currently a full professor in the CAS Key Lab of Network Data Science and Technology, Institute of Computing Technology, Chinese Academy of Sciences (ICT-CAS). He is also a professor of the University of CAS. He is the deputy secretary-general of the Task Force on Big Data, China Computer Federation. He obtained his Ph.D. degree in Computer Science from Hong Kong Baptist University in 2005. His current research interests mainly include big data knowledge engineering, knowledge graph, knowledge computing, etc. Dr. Jin has co-authored four monographes, published by Springer and Tsinghua University Press, respectively. He has published more than 200 papers in prestigious international journals and conferences, including IEEE TKDE, ACM TWeb, IEEE/ACM TASLP, ACM TIST, IEEE TWC, IEEE TPDS, IJCAI, AAAI, WWW, WSDM, WI, ICBK, IAT, AINA, AAMAS etc. He received Best Student Paper Award at IEEE ICBK-2017, Best Paper Award at CIT-2015, Best Academic Paper Award at CCF Big Data 2015, Elsevier Top Cited Articles Award-2019.
Sharing by Prof. Mang Ye
Virtual-to-Real: Learning a Generalized Re-identification Model from All-weather Virtual Data
Person re-identification (Re-ID) is a task of retrieving a person of interest across multiple non-overlapping surveillance cameras, where the model generalization ability is important for practical large-scale surveillance camera networks. However, real training data annotation is costly and model generalization ability is hindered by the lack of large-scale and diverse training data. In this talk, I will introduce a virtual-to-real Weather Person pipeline, generating a large-scale and diverse synthesized Re-ID dataset with different weather, scenes and natural lighting conditions automatically. We evaluate our pipeline using direct transfer on 3 widely-used real-world benchmarks, achieving competitive performance without any real-world training data.
Mang Ye is currently a Full Professor at the School of Computer Science, Wuhan University. He received the PhD degree from Hong Kong Baptist University in 2019, supported by Hong Kong PhD Fellowship. He worked as a Research Scientist at Inception Institute of Artificial Intelligence from 2019-2020 and worked as a Visiting Scholar at Columbia University in 2018. He has published more than 50 papers, including 20+ CCF-A/IEEE Trans. papers as the first/corresponding author. He received 2100+ citations, including those from 2 Turing awardees (Geoffrey Hinton and Yann Lecun). Five papers are ESI Highly Cited. His research interests include open-world visual learning and its applications in multimedia analysis, person re-identification and etc.
Sharing by Dr. Shaohuai Shi
Towards Scalable Distributed Training of Deep Learning on Public Cloud Clusters
Distributed training techniques have been widely deployed in large-scale deep models training on dense-GPU clusters. However, on public cloud clusters, due to the moderate inter-connection bandwidth between instances, traditional state-of-the-art distributed training systems cannot scale well in training large-scale models. In this talk, I will introduce our recent techniques using new top-k computation and communication schemes for efficient distributed training. Experimental results on a 16-node Tencent Cloud cluster (each node with 8 Nvidia Tesla V100 GPUs) show that our system achieves 25%-40% faster than existing state-of-the-art systems on CNNs and Transformer. We finally break the record on DAWNBench on training ResNet-50 to 93% top-5 accuracy on ImageNet.
Shaohuai Shi is currently a Research Assistant Professor in the Department of Computer Science and Engineering of the Hong Kong University of Science and Technology. Shaohuai received his Ph.D. degree in Computer Science from Hong Kong Baptist University in 2020. His current research interests focus on distributed machine learning systems. He is an awardee of the Best Paper Award in IEEE INFOCOM 2021 and IEEE DataCom 2018.
Sharing by Dr. Qian Chen
Towards End-to-End Automatic Speech Recognition
Traditional automatic speech recognition (ASR) systems usually involve acoustic, pronunciation and language model components which are trained separately. Curating each component requires expert knowledge, and is time-consuming. The recent deep neural network has boosted the tremendous advances in end-to-end ASR technologies. The end-to-end models directly map a sequence of input acoustic features into a sequence of graphemes or words. These models are trained to optimize criteria that are related to the final evaluation metric that we are interested in (typically, word error rate). The models require less expert knowledge and are trained much faster than the traditional models while achieving high quality at the same time.
Dr. Chen obtained his Ph.D degree in Computer Science from the Hong Kong Baptist University (HKBU) in 2015. He received his BSc in Computer Science from the East China Normal University in 2011. He was awarded as a world finalist in ACM/ICPC 2020. His publications appear in SIGMOD, VLDB, ICDE, TKDE etc.
Currently Dr. Chen is a software engineer at Google focusing on automatic speech recognition (ASR) and machine translation areas. Prior to that, he was a senior software engineer at Grab and Netease Games.
Sharing by Dr. Xiaowei Chen
Cloud & Transformation, Always Day 1
Cloud is the new normal, which brings the prosperous growth in AI and data analytics. I will introduce the clients' cloud stories to showcase the practices of ABCDMI (AI, Blockchain, Cloud, Data, Mobile, IoT). Then I will introduce FinTech solutions powered by OneConnect and Ping An group, like digital identity verification, smart lending, auto insurance, as well as related user stories. These transformation journeys show the power of emerging technologies. More to come, and many more to be transformed. It is still Day 1.
Dr. Michael Chen is the General Manager of OneConnect Financial Technology (Hong Kong), associate of Ping An Group. Prior to joining OneConnect, he worked in Deloitte and AWS, where he led various large scale cloud transformations for enterprise and global financial clients. He obtained the PhD degree in Computer Science from Hong Kong Baptist University.
Sharing by Dr. Jinhua Ma
Background Robust Motion perception for Self-supervised Video Representation Learning
In this talk, I will present two recent works on background robust motion perception for self-supervised video representation learning. The first one is the development of a novel pretext task namely Channel Aliasing Video Perception (CAVP). The main idea of this method is to recognize the number of different motion flows within a channel aliasing video for perception of discriminative motion cues. In the second work, the objective is to mitigate the representation bias w.r.t. background. For an input video, a (static) frame is randomly selected and added to the input to construct a distracting video for contrastive learning. As plug-and-play techniques, the proposed methods can be integrated with other self-supervised learning methods to further improve the performance. Experimental results on publicly available action recognition benchmarks verify the effectiveness of these two methods for self-supervised pre-training.
Andy J. Ma received his Ph.D. degree in Computer Science at Hong Kong Baptist University. After that, he worked as a Post-Doctoral Fellow at Rutgers University and Johns Hopkins University. Now, he is an associate professor at Sun Yat-sen University. He has published more than 50 papers in top-tier journals and conferences including IEEE TPAMI, IJCV, IEEE TIP, Critical Care Medicine, Alimentary Pharmacology & Therapeutics, ICCV, CVPR, ECCV, AAAI, IJCAI, MICCAI, etc. He serves as an associate editor for the SPIE Journal of Electronic Imaging. His current research interests focus on developing machine learning algorithms for intelligent video surveillance and medical applications.
Sharing by Dr. Fei Liu
Application of Deep Learning in Business Research
The capturing and monetization of consumer attention is the key to success in the platformization of content creation. The recent advancement in deep learning enables new approaches to theory building and testing for decoding consumers’ attention mechanism. This talk highlights two exemplary studies that apply deep learning to circumvent pre-existing methodological constraints in business research. The first study utilized a state-of-art natural language model (i.e., BERT) to investigate how social media users react to misinformation during an infodemic. By fine-tuning the language model with a COVID-19 Twitter archive and a fact-check database, we are able to identify if a tweet entails debunked misinformation. It then analysed how Twitter users respond to misinformation. The second study employed latest computer vision models (e.g., AMNet) to predict how viewers allocate attention on each service marketing image on a service e-tailing platform. It then investigated how to make use of aesthetically pleasing and memorable images to boost service sales.
Dr. Fei Liu is currently an Assistant Professor in Management Information Systems (MIS) at Hong Kong Polytechnic University. He received his PhD in Digitalization from Copenhagen Business School. He also holds a PhD in Computer Science from Hong Kong Baptist University. His current research focuses on behavioral analytics and digital transformation. By combining predictive analytics with behavioral science, he aims to investigate new usage patterns and socioeconomic impacts of digital technologies. His research areas include but not exclusive to service platformization, crowdfunding, social media, and creator economy. His research has been presented at prestigious international conferences and published in reputable journals. In recognition of his engagement to the MIS community, he has been awarded AIS Doctoral Student Service Award and several best reviewer awards.
Sharing by Dr. Zhe Fan
A Brief Overview of Derivatives Market Maker
The derivatives market refers to the financial market for financial instruments such as futures contracts or options that are based on the values of their underlying assets. As such market may be lack of liquidity, the Market Makers (in short MM) involve. MM refers to a firm or individual who actively quotes two-sided orders (a.k.a., bids and asks) in a particular financial instrument, providing the market with liquidity. In this talk, we will see how MM works and earns profits, and what techniques MM uses, etc.
I am now a Quant Trader in China Merchants Securities (招商证券) since 2018-01, doing Derivatives Market Making. I was a big data researcher in Huawei from 2015-10 to 2017-12. I obtained my Ph.D degree in CS from HKBU in 2015, supervised by Dr. Byron Choi and Prof. Jianliang Xu. I received my BEng degree in CS from SCUT (华南理工大学) in 2011.
Sharing by Dr. Lei Chen
Enterprise Intelligence: Challenges and Opportunities
In this talk, I will introduce some representative industrial scenarios and some of our technical works in the Enterprise Intelligence Team, Huawei Noah's Ark Lab. The scenarios vary in different industry areas, including the supply chain management, the storage system and product, circuit design and our business solutions supported by Huawei Cloud. I'd like to share and discuss the challenges and our solutions via some real-world application examples in these areas. Efficient and intelligent algorithms would greatly improve the efficiency of industrial systems and bring numerous values. We expect to promote the joint research of both academia and industry in these essential practical problems.
Dr. Chen is a researcher at Huawei Noah's Ark Lab, Hong Kong. Prior to joining Huawei, Dr. Chen obtained his PhD degree in Computer Science from the Hong Kong Baptist University in 2016 and received his BEng degree in Computer Science from South China University of Technology in 2012. His publications appear in SIGKDD, SIGMOD, VLDB, ICDE, AAAI, TKDE, etc.
October 9 (Saturday), 2021
FSC501, Ho Sin Hang Campus
“AI for Social Good: Opportunities and Challenges” by Professor William Cheung
— Professor, Department of Computer Science
Workshop on AI and Big Data
WLB104, Shaw Campus
/ Virtual (via Zoom)
14:45 - 16:25
16:25 - 17:35
14:45 - 14:55
Welcoming Remarks by Professor Jiming Liu
— Dean of Science and Chair Professor, Department of Computer Science
14:55 - 15:10
“Event Knowledge Acquisition and Reasoning” by Professor Xiaolong Jin
— Professor, Institute of Computing Technology (ICT), Chinese Academy of Sciences (CAS), China
15:10 - 15:25
“Virtual-to-Real: Learning a Generalized Re-identification Model from All-weather Virtual Data” by Professor Mang Ye
—Professor, School of Computer Science, Wuhan University, China
15:25 - 15:40
“Towards Scalable Distributed Training of Deep Learning on Public Cloud Clusters” by Dr. Shaohuai Shi
— Research Assistant Professor, Department of Computer Science and Engineering, Hong Kong University of Science and Technology, Hong Kong
15:40 - 15:55
“Towards End-to-End Automatic Speech Recognition” by Dr. Qian Chen
— Software Engineer, Google
15:55 - 16:10
“Cloud & Transformation, Always Day 1” by Dr. Xiaowei Chen
— General Manager, OneConnect Financial Technology, Hong Kong
16:25 - 16:40
“Background Robust Motion perception for Self-supervised Video Representation Learning” by Dr. Jinhua Ma
— Associate Professor, Sun Yat-sen University, China
16:40 - 16:55
“Application of Deep Learning in Business Research” by Dr. Fei Liu
— Assistant Professor, Department of Management and Marketing, The Hong Kong Polytechnic University, Hong Kong
16:55 - 17:10
“A Brief Overview of Derivatives Market Maker” by Dr. Zhe Fan
— Quant Trader, China Merchants Securities
17:10 - 17:25
“Enterprise Intelligence: Challenges and Opportunities” by Dr. Lei Chen
—Researcher, Huawei Noah's Ark Lab
17:25 - 17:35
Conclusion by Professor Jianliang Xu
— Head, Department of Computer Science
Students / Staff
Alumni / External Guests
October 6, 2021 (Wednesday)
Other Homecoming Events:
HKBU Homecoming Day 2021
HKBU Science Homecoming Day 2021