Speakers and Speeches Information

Narendra Ahuja

Professor
University of Illinois at Urbana-Champaign

Title: Building an R&D Ecosystem for IT-Based Societal Problem Solving in India

Abstract: There is an experimental initiative under way in India, called Information Technology Research Academy (ITRA), aimed at advancing the quality and quantity of R&D in Information and Communications Technologies (IT) and its applications, at a steadily growing number of academic and research institutions, while strengthening the academic culture of IT based problem solving and societal development.
The objective is to build an R&D ecosystem, directed at finding solutions to important problems, by linking academic and research institutions, industry, government and other organizations into tightly knit teams. The ecosystem simultaneously enhances the quality of: R&D activity itself; Advanced IT and related education; Researchers’ ability, interest and satisfaction from directing their R&D prowess at real-world challenges; and Mechanisms/opportunities to convert these capabilities into impact by transferring technologies to industry and through start-ups. The ecosystem is designed to converge as a conglomerate of many coexisting centers of excellence, or think tanks, in specific areas, to simultaneously act as: an IT knowledge bank; a catalyst for raising generations of IT-equipped, technology-proficient, societally-sensitive researchers; and an engine of economic activity, driven by the vitality of action, intellectual prowess, emotional energy, and ultimately, the societal empathy of the researchers.
This talk will present an overview of the first three years of ITRA, including specific measures of quality targeted, challenges faced, models and mechanisms used to address these challenges, types of problems undertaken, solutions developed, their impact, an analysis of the ITRA performance, and where the entire experiment is headed.

Bio: Narendra Ahuja is Research Professor in the Dept. of Electrical and Computer Engineering, Beckman Institute, and Coordinated Science Laboratory, University of Illinois at Urbana-Champaign (UIUC), and Founding Director of Information Technology Research Academy, Ministry of Electronics and Information Technology, Government of India. He received B.E. with honors in electronics engineering from BITS, Pilani, India, M.E. with distinction in electrical communication engineering from IISc, Bangalore, India, and Ph.D. in computer science from University of Maryland, College Park, USA. In 1979, he joined UIUC where he was Donald Biggar Willet Professor of Engineering until 2012. During 1999-2002, he served as the Founding Director of International Institute of Information Technology, Hyderabad.
He has co-authored three books, several hundred papers, and received 4 patents. His algorithms/prototype systems have been used by about a dozen companies/organizations, including industrial systems at General Electric, Westinghouse, Lockheed and Honeywell. He is a fellow of IEEE, American Association for Artificial Intelligence, International Association for Pattern Recognition, Association for Computing Machinery, American Association for the Advancement of Science, and International Society for Optical Engineering. He received the Emanuel R. Piore award of the IEEE, and the Technology Achievement Award of the International Society for Optical Engineering, and TA Stewart-Dyer/Frederick Harvey Trevithick Prize of the Institution of Mechanical Engineers. With his students, he shared Best Paper Awards given by: International Conference on Pattern Recognition (Piero Zamperoni Award), Symposium on Eye Tracking Research and Applications, First IEEE International Workshop on Computer Vision in Sports, International Conference on Pattern Recognition, and IEEE Transaction on Multimedia.

Xilin Chen

Professor
University of Chinese Academy of Science

Title: Object Understanding in Natural Scene

Abstract: Object recognition is one of the basic and important tasks in computer vision. In the past two decades, computer vision has made some progresses in major tasks, such as face recognition, car recognition, etc. However, it’s still a big challenge to understanding objects and their relationship in real world. As an object exists in an ecosystem related to its property from many aspects, we model object recognition as a similarity measurement in high dimensional semantic space. This representation provides a natural hierarchical way to describe object, and it’s easy to support further tasks, such as zero shot object understanding, image caption and visual QA. To embed both appearance and semantic similarities, we propose to learn the binary code which encodes identity and semantic attributes simultaneously. The representation also provides a chance to share (latent) attributes a common semantic space. Therefore, it can be naturally used to inference unseen classes. We use this representation for multitask recognition, Zero-shot learning, and other vision tasks.

Bio: Dr. Xilin Chen is a professor with Institute of Computing Technology, Chinese Academy of Sciences (CAS), Beijing. He is a Fellow of IEEE, a Fellow of IAPR and a Fellow of China Computer Federation (CCF).
Dr. Xilin Chen was an associate editor of IEEE Transactions on Image Processing from 2009 to 2014. He is now an associate editor of IEEE Transactions on Multimedia, a leading editor of Journal of Computer Science of Technology, an associate editor in chief of the Chinese Journal of Computer, and an associate editor in chief of the Chinese Journal of Pattern Recognition and Artificial Intelligence. He served as an organizing committee member / program committee member for more than 80 International Conferences in related areas, including ICCV, CVPR, ECCV, FG, ICMI, ACM MM, ICIP, ICPR, etc.
Dr. Xilin Chen’s research interests include Computer Vision, Pattern Recognition, Image Processing, Multimodal Interface. He is a recipient of one China's State Natural Science Award (2015) and four China's State Scientific and Technological Progress Awards (2012, 2005, 2003, and 2000). He co-authored more than 200 papers.

Yuxin Chen

Assistant Professor
Electrical Engineering of Princeton University

Title: Solving Random Quadratic Systems of Equations Is Nearly as Easy as Solving Linear Systems

Abstract: We consider the fundamental problem of solving quadratic systems of equations in n variables. We propose a novel method, which starting with an initial guess computed by means of a spectral method, proceeds by minimizing a nonconvex functional as in the Wirtinger flow approach. There are several key distinguishing features, most notably, a distinct objective functional and novel update rules, which operate in an adaptive fashion and drop terms bearing too much influence on the search direction. These careful selection rules provide a tighter initial guess, better descent directions, and thus enhanced practical performance. On the theoretical side, we prove that for certain unstructured models of quadratic systems, our algorithms return the correct solution in linear time, i.e. in time proportional to reading the data as soon as the ratio m/n between the number of equations and unknowns exceeds a fixed numerical constant. We complement our theoretical study with numerical examples showing that solving random quadratic systems is both computationally and statistically not much harder than solving linear systems of the same size.

Bio: Yuxin Chen is currently an assistant professor in the Department of Electrical Engineering at Princeton University. Prior to joining Princeton, he was a postdoctoral scholar in the Department of Statistics at Stanford University, and he completed his Ph.D. in Electrical Engineering at Stanford University. His research interests include high-dimensional structured estimation, convex and nonconvex optimization, statistical learning, and information theory.

Kyunghyun Cho

Assistant Professor of New York University

Title: Deep Learning, Where Are You Going?

Abstract: There are three axes along which advances in machine learning and deep learning happen. They are (1) network architectures, (2) learning algorithms and (3) spatio-temporal abstraction. In this talk, I will describe a set of research topics I've pursued in each of these axes. For network architectures, I will describe how recurrent neural networks, which were largely forgotten during 90s and early 2000s, have evolved over time and have finally become a de facto standard in machine translation. I continue on to discussing various learning paradigms, how they related to each other, and how they are combined in order to build a strong learning system. Along this line, I briefly discuss my latest research on designing a query-efficient imitation learning algorithm for autonomous driving. Lastly, I present my view on what it means to be a higher-level learning system. Under this view each and every end-to-end trainable neural network serves as a module, regardless of how they were trained, and interacts with each other in order to solve a higher-level task. I will describe my latest research on trainable decoding algorithm as a first step toward building such a framework.

Bio: Kyunghyun Cho is an assistant professor of computer science and data science at New York University. He was a postdoctoral fellow at University of Montreal until summer 2015, and received PhD and MSc degrees from Aalto University early 2014. He tries best to find a balance among machine learning, natural language processing and life, but often fails to do so.

Shenghua Gao

Assistant Professor
ShanghaiTech University

Title: Multi-Column CNN and Its Applications

Abstract: Recently, Convolutional Neural Networks (CNN) has demonstrated its great successes in many computer vision tasks. Other than building deeper and deeper CNN which encounters the difficulties in parameter optimization, we propose a Multi-Column CNN architecture, and the depth in each column is not so-deep. Compared with very deep CNN, our framework is easy to be optimized and to be paralleled. Then the proposed framework is applied in crowd counting and face alignment. Experimental results demonstrates the effectiveness of our framework.

Bio: Shenghua Gao is an assistant professor in ShanghaiTech University, China. He received the B.E. degree from the University of Science and Technology of China in 2008 (outstanding graduates), and received the Ph.D. degree from the Nanyang Technological University in 2012. From Jun 2012 to Aug 2014, he worked as a research scientist in Advanced Digital Sciences Center, Singapore. From Jan 2015 to June 2015, he visited UC Berkeley as a visiting scholar. His research interests include computer vision and machine learning. He has published more than 40 papers on image and video understanding related topics in peer-reviewed international conferences and journals, including IEEE T-PAMI,IJCV, IEEE TIP, IEEE TNNLS, CVPR, etc. He was awarded the Microsoft Research Fellowship in 2010, ACM Shanghai Young Research Scientist in 2015.

Bernd Girod

Professor
Electrical Engineering,Stanford University

Title: From Pixels to Information - Recent Advances in Visual Search

Abstract: With intelligent processing, cameras have great potential to link the real world and the virtual world. We review advances and opportunities for algorithms and applications that retrieve information from large databases using images as queries. For rate-constrained applications, remarkable improvements have been achieved over the course the MPEG-CDVS (Compact Descriptors for Visual Search) standardization. Beyond CDVS lie applications that query video databases with images, while others continually match video frames against image databases. Exploiting the temporal coherence of video for either case can yield large additional gains. We will look at implementations for example applications ranging from text recognition to augmented reality to understand the challenges of building databases for rapid search and scalability, as well as the tradeoffs between processing on a mobile device and in the cloud.

Bio: Bernd Girod is the Robert L. and Audrey S. Hancock Professor of Electrical Engineering at Stanford University, California. He received the Engineering Doctorate degree from University of Hannover, Germany, and the M.S. degree from the Georgia Institute of Technology. Until 1999, he was a Professor with the Electrical Engineering Department, University of Erlangen– Nuremberg. He has authored over 600 conference and journal papers and six books, receiving the EURASIP Signal Processing Best Paper Award in 2002, the IEEE Multimedia Communication Best Paper Award in 2007, the EURASIP Image Communication Best Paper Award in 2008, the EURASIP Signal Processing Most Cited Paper Award in 2008, the EURASIP Technical Achievement Award in 2004, and the Technical Achievement Award of the IEEE Signal Processing Society in 2011. His research interests are in the area of image, video, and multimedia systems. As an entrepreneur, he was involved in numerous startup ventures, among them Polycom, Vivo Software, 8x8, and RealNetworks. He is a Fellow of the IEEE, a EURASIP Fellow, a member of the the National Academy of Engineering, and a member of the German National Academy of Sciences (Leopoldina).

Yongdae Kim

Professor
Korea Advanced Institute of Science and Technology

Title: Hacking Sensors

Abstract: Sensors are designed to measure sensor inputs (e.g., physical quantities) and transfer sensor outputs (e.g. voltage signal) into the embedded devices. In addition, sensor-equipped embedded systems (called sensing-and-actuation systems) decide their actuations according to these sensor outputs, and the systems have no doubt whether the sensor outputs are legitimate or not. Sensors are essential components for safety-critical systems such as self-driving cars, drones and medical devices. Breaking safety in these systems may cause loss of life or disasters. Because of these safety reasons, sensors are often designed to be robust against failure or faults.
However, can they maintain safety under adversarial conditions? In this talk, I detail how sensors can be spoofed or prevented from providing correct operation through regular and side-channels. Attacks on various devices such as medical devices, drones, smart wearables, and automobiles will be shown. I'll complete the talk with a few directions and guides to prevent these attacks with a few open problems.

Bio: Yongdae Kim is a Professor in the Department of Electrical Engineering and an affiliate professor in the GSIS at KAIST. He received PhD degree from the computer science department at the University of Southern California. Between 2002 and 2012, he was an associate/assistant professor in the Department of Computer Science and Engineering at the University of Minnesota - Twin Cities. Before joining U of Minnesota, he worked as a research staff for two years in Sconce Group in UC Irvine. Before coming to the US, he worked 6 years in ETRI for securing Korean cyberinfrastructure. Between 2013 and 2016, he served as a KAIST Chair Professor. He received NSF career award on storage security and McKnight Land-Grant Professorship Award from University of Minnesota in 2005. Currently, he is serving as a steering committee member of NDSS and associate editor for ACM TISSEC. His current research interests include security issues for various systems such as cyber physical systems, social networks, cellular networks, P2P systems, medical devices, storage systems, mobile/ad hoc/sensor networks, and anonymous communication systems.

Yann LeCun

Director
AI Research, Facebook Inc.

Title: Obstacles to Progress in AI and Deep Learning

Abstract: Deep learning is at the root of revolutionary progress in visual and auditory perception by computers, and is pushing the state of the art in natural language understanding, dialog systems and language translation. Deep learning systems are deployed everywhere from self-driving cars to content filtering, search, and medical image analysis. But almost all real-world applications of deep learning use supervised learning in which the machine is trained with inputs labeled by humans. But humans and animals learn vast amounts of knowledge about the world by observation, with very little feedback from intelligent teachers. Humans construct complex predictive models of the world that allows them to interpret percepts, predict future events, and plan a course of actions. Enabling machines to learn predictive models of the world is a major obstacle towards significant progress in AI. I will describe a number of promising approaches towards unsupervised and predictive learning.

Bio:

Cheng-Lin Liu

Director
National Laboratory of Pattern Recognition (NLPR)

Title: Document Image Analysis: New Drives and Frontiers

Abstract: Document image analysis (DIA, including OCR, online and offline handwriting recognition) has been studied intensively since 1960s and has gained success in many applications. In recent years, the development of Internet, Internet of Things, and the popular use of smart phones have triggered a new round of boom in research and application needs. Deep learning approaches, while pushing forward the performance of DIA, cannot fill in the gap between research and application need. In this talk, I will retrospect some traditional applications of DIA that are still alive or even growing, such as financial forms processing and postal mail sorting. Emerging application needs include Web document retrieval, historical document digitization, applications in education and medicine. I will show some technical challenges behind these applications and give a list of frontier research directions. Deep learning will be an optional component in future DIA systems but there are many important issues in addition to deep learning.

Bio: Cheng-Lin Liu is a Professor at the National Laboratory of Pattern Recognition (NLPR), Institute of Automation of Chinese Academy of Sciences, Beijing, China, and is now the director of the laboratory. He received the B.S. degree in electronic engineering from Wuhan University, Wuhan, China, the M.E. degree in electronic engineering from Beijing Polytechnic University, Beijing, China, the Ph.D. degree in pattern recognition and intelligent control from the Chinese Academy of Sciences, Beijing, China, in 1989, 1992 and 1995, respectively. He was a postdoctoral fellow at Korea Advanced Institute of Science and Technology (KAIST) and later at Tokyo University of Agriculture and Technology from March 1996 to March 1999. From 1999 to 2004, he was a research staff member and later a senior researcher at the Central Research Laboratory, Hitachi, Ltd., Tokyo, Japan. His research interests include pattern recognition, image processing, neural networks, machine learning, and especially the applications to character recognition and document analysis. He has published over 200 technical papers at prestigious international journals and conferences. He won the IAPR/ICDAR Young Investigator Award of 2005. He is on the editorial board of Pattern Recognition Journal, Image and Vision and Computing, International Journal on Document Analysis and Recognition, and Cognitive Computation. He is a Fellow of the IAPR and the IEEE.

Tie-Yan Liu

Principal Researcher
Microsoft Research Asia

Title: What You Might Not Know About Distributed Deep Learning

Abstract: Due to the big data and big model, distributed training has become the de facto setting for many practical deep learning tasks. In the literature of distributed deep learning, people have mainly focused on the design of different synchronization mechanisms (and proved their convergence properties), but have largely ignored the impacts of other components. For example, local shuffling of training data is widely used in distributed deep learning, which violates the i.i.d. assumption and poses challenges to the convergence analysis; the aggregation methods like model average may have no theoretical guarantee when the learning problem is nonconvex (which is exactly the case for deep learning). In this talk, we first discuss the probabilistic nature of random shuffling and analyze its impacts on the convergence of distributed deep learning under different settings (e.g., convex vs. nonconvex problems, local shuffling vs. global shuffling, efficient shuffling vs. inefficient shuffling). Second, we propose a new model aggregation method specially designed for nonconvex problems, and demonstrate its superior performance to model average in the context of deep learning. At the end of the talk, we will discuss some other important and usually ignored issues related to distributed deep learning.

Bio: Tie-Yan Liu is a principal researcher of Microsoft Research Asia, leading the research on artificial intelligence and machine learning. He is very well known for his pioneer work on learning to rank and computational advertising, and his recent research interests include deep learning, reinforcement learning, and distributed machine learning. Many of his technologies have been transferred to Microsoft’s products and online services (such as Bing, Microsoft Advertising, and Azure), and open-sourced through Microsoft Cognitive Toolkit (CNTK), Microsoft Distributed Machine Learning Toolkit (DMTK), and Microsoft Graph Engine. He has also been actively contributing to academic communities. He is an adjunct/honorary professor at Carnegie Mellon University (CMU), University of Nottingham, and several other universities in China. His papers have been cited for tens of thousands of times in refereed conferences and journals. He has won quite a few awards, including the best student paper award at SIGIR (2008), the most cited paper award at Journal of Visual Communications and Image Representation (2004-2006), the research break-through award at Microsoft Research (2012), and Top-10 Springer Computer Science books by Chinese authors (2015). He has been invited to serve as general chair, program committee chair, local chair, or area chair for a dozen of top conferences including SIGIR, WWW, KDD, ICML, NIPS, IJCAI, AAAI, ACL, ICTIR, as well as associate editor/editorial board member of ACM Transactions on Information Systems, ACM Transactions on the Web, Neurocomputing, Information Retrieval Journal, and Foundations and Trends in Information Retrieval. Tie-Yan Liu is a fellow of the IEEE, a distinguished member of the ACM, an academic committee member of the CCF, and a vice chair of the CIPS information retrieval technical committee.

Shiqian Ma

Assistant Professor
The Chinese University of Hong Kong

Title: Geometric Descent Method for Convex Composite Minimization

Abstract: We extend the geometric descent method recently proposed by Bubeck, Lee and Singh to tackle nonsmooth and strongly convex composite problems. We prove that our proposed algorithm, dubbed geometric proximal gradient method (GeoPG), converges with a linear rate $(1-1/\sqrt{\kappa})$, and thus achieves the optimal rate among first-order methods, where $\kappa$ is the condition number of the problem. Numerical results on linear regression and logistic regression with elastic net regularization show that GeoPG compares favorably with Nesterov's accelerated proximal gradient method, especially when the problem is ill-conditioned.

Bio: Shiqian Ma received his B.S. from Peking University in 2003, M.S. from Chinese Academy of Sciences in 2006 and Ph.D. in Industrial Engineering and Operations Research from Columbia University in 2011. He then spent one and half years in the Institute for Mathematics and Its Applications at University of Minnesota as an NSF postdoctoral fellow. Shiqian Ma joined the Department of Systems Engineering and Engineering Management, The Chinese University of Hong Kong in December 2012. His current research interests include theory and algorithms for large-scale optimization and its applications in big data analytics, statistics, machine learning, bioinformatics, signal processing and image processing. Shiqian Ma received the INFORMS Optimization Society best student paper prize in 2010, honorable mention of INFORMS George Nicholson student paper competition in 2011. He was one of the finalists of the 2011 IBM Herman Goldstine fellowship. He received the Journal of the Operations Research Society of China Excellent Paper Award in 2016.

Jitendra Malik

Arthur J. Chick Professor
EECS,University of California, Berkeley

Title: Deep Visual Understanding from Deep Learning

Abstract: Deep learning and neural networks coupled with high-performance computing and big data have led to remarkable advances in computer vision. For example, we now have a good capability to detect and localize people or objects. But we are still quite short of “visual understanding”. I’ll sketch some of our recent progress towards this grand goal. One is to explore the role of feedback or recurrence in visual processing. Another is to unify geometric and semantic reasoning for understanding the 3D structure of a scene. Most importantly, vision in a biological setting, and for many robotics applications, is not an end in itself but to guide manipulation and locomotion. I will show results on learning to perform manipulation tasks by experimentation, as well as on a cognitive mapping and planning architecture for mobile robotics.

Bio: Jitendra Malik is Arthur J. Chick Professor and Department Chair of Electrical Engineering and Computer Science at UC Berkeley. Over the past 30 years, Prof. Malik's research group has worked on many different topics in computer vision. Several well-known concepts and algorithms arose in this research, such as anisotropic diffusion, normalized cuts, high dynamic range imaging, shape contexts and R-CNN. Prof. Malik received the Distinguished Researcher in Computer Vision Award from IEEE PAMI-TC, the K.S. Fu Prize from the International Association of Pattern Recognition, and the Allen Newell award from ACM and AAAI. He has been elected to the National Academy of Sciences, the National Academy of Engineering and the American Academy of Arts and Sciences. He earned a B.Tech in Electrical Engineering from Indian Institute of Technology, Kanpur in 1980 and a PhD in Computer Science from Stanford University in 1985.

Jianbo Shi

Professor
University of Pennsylvania

Title: Connecting the Dots: Embodied Visual Memory with First-person Vision

Abstract: A computer has a complete photographical memory. It creates massive but isolated sensory moments. Unlike such fragmented photographic memory, human memories are highly connected through episodes that allow us to relate past experiences and predict future actions. How to computationally model a human like episodic memory system that connects photographically accurate sensory moments? Our insight is that an active interaction is a key to link between episodes because sensory moments are fundamentally centered on an active person-self. Our experiences are created by and shared through our social and physical interactions, i.e., we connect episodes driven by similar actions and, in turn, recall these past connected episodes to take a future actions. Therefore, connecting the dotted moments to create an episodic memory requires understanding the purposeful interaction between human (person-self) and world. In this talk we focus on our work in 1) visual attention, 2) action prediction, 3) visual control and 4) skill assessment.

Bio: Jianbo Shi studied Computer Science and Mathematics as an undergraduate at Cornell University where he received his B.A. in 1994. He received his Ph.D. degree in Computer Science from University of California at Berkeley in 1998. He joined The Robotics Institute at Carnegie Mellon University in 1999 as a research faculty, and in 2003 University of Pennsylvania where he is currently a Professor of Computer and Information Science. In 2007, he was awarded the Longuet-Higgins Prize for his work on Normalized Cuts. His current research focuses on first person vision, human behavior analysis and image recognition-segmentation. His other research interests include image/video retrieval, 3D vision, and vision based desktop computing. His long-term interests center around a broader area of machine intelligence, he wishes to develop a "visual thinking" module that allows computers not only to understand the environment around us, but also to achieve cognitive abilities such as machine memory and learning.

Harry Shum

Executive Vice President
Microsoft Inc.

Title: Artificial Intelligence: From the Labs to the Mainstream

Abstract: Three are three big forces that are making AI possible: huge amounts of data with the Internet and sensors everywhere; massive computing power especially and breakthrough algorithms. These forces are enabling computers to accomplish more and more sophisticated tasks on their own with deep learning. In this talk, Dr. Shum will discuss Microsoft’s overall efforts and progress with AI research and product development, with a particular focus on computer vision. Microsoft has long been committed to developing new computer vision technologies, making them available to developers, and incorporating them into many products. He will briefly review more than 25 years of computer vision research at Microsoft Research (MSR), highlighting MSR’s contributions to the vision community and emphasizing the importance of long-term commitment to funding successful industrial research labs.

Bio: Dr. Harry Shum is Executive Vice President of Microsoft Corporation, in charge of Artificial Intelligence and Microsoft Research. He joined Microsoft Research in Redmond, Washington as a researcher in 1996, and has since taken on various research and development roles including Managing Director of Microsoft Research Asia, Corporate Vice President for Bing Product Development, and Executive Vice President of Technology and Research. He is an ACM Fellow and IEEE Fellow for his research contribution in computer vision and computer graphics. In 2017, he was elected to the National Academy of Engineering of the United States.

Dawn Song

Professor
University of California, Berkeley

Title: Towards Secure AI: Lessons, Challenges and Future Directions

Abstract: In this talk, I will first present recent results in the area of secure deep learning, in particular, adversarial deep learning. To make neural networks more resilient, I will also describe our recent work on how to make neural programs generalize better and provide provable guarantees of generalization in certain application domains. Finally, I will conclude with key challenges and future directions at the intersection of AI and Security: how AI and deep learning can enable better security, and how Security can enable better AI.

Bio: Dawn Song is a Professor in the Department of Electrical Engineering and Computer Science at UC Berkeley. Her research interest lies in deep learning and security. She has studied diverse security and privacy issues in computer systems and networks, including areas ranging from software security, networking security, database security, distributed systems security, applied cryptography, to the intersection of machine learning and security. She is the recipient of various awards including the MacArthur Fellowship, the Guggenheim Fellowship, the NSF CAREER Award, the Alfred P. Sloan Research Fellowship, the MIT Technology Review TR-35 Award, the George Tallman Ladd Research Award, the Okawa Foundation Research Award, the Li Ka Shing Foundation Women in Science Distinguished Lecture Series Award, the Faculty Research Award from IBM, Google and other major tech companies, and Best Paper Awards from top conferences. She obtained her Ph.D. degree from UC Berkeley. Prior to joining UC Berkeley as a faculty, she was an Assistant Professor at Carnegie Mellon University from 2002 to 2007.

Yuandong Tian

Research Scientist
Facebook Inc.

Title: AI in Games: Achievements and Challenges

Abstract: Recently, substantial progress of AI has been made in applications that require advanced pattern reading, including computer vision, speech recognition and natural language processing. However, it remains an open problem whether AI will make the same level of progress in tasks that require sophisticated reasoning, planning and decision making, e.g., when exploring complicated game environments that bear a resemblance to real-world applications. In this talk, I present our recent contributions on these directions, in terms of novel algorithms and new environments. We will also discuss issues and challenges.

Bio: Yuandong Tian is a Research Scientist in Facebook AI Research, working on reasoning with deep learning in games and theoretical analysis of deep non-convex models. He is the leader researcher and engineer for DarkForest (Facebook Computer Go project). Prior to that, he was a Software Engineer/Researcher in Google Self-driving Car team during 2013-2014. He received Ph.D in Robotics Institute, Carnegie Mellon University on 2013, Bachelor and Master degree of Computer Science in Shanghai Jiao Tong University. He is the recipient of 2013 ICCV Marr Prize Honorable Mentions for his work on global optimal solution to nonconvex optimization in image alignment.

Kewei Tu

Assistant Professor
ShanghaiTech University

Title: Representation and Learning of Grammars Beyond the Language Domain

Abstract: I will introduce how stochastic grammars that are widely used in NLP can be extended to become stochastic And-Or grammars (AOGs) that are capable of modeling many other types of data such as images and vectors. To demonstrate the utility of the AOG framework, I will discuss novel models and algorithms in probabilistic modeling that are related to or inspired by AOGs. Supervised learning of AOGs is not possible in most scenarios because gold parses are not available. Therefore, I will discuss a few unsupervised approaches to learning the structures and parameters of AOGs.

Bio: Dr. Kewei Tu is an Assistant Professor with the School of Information Science and Technology at ShanghaiTech University, China. He received BS and MS degrees in Computer Science and Technology from Shanghai Jiaotong University, China in 2002 and 2005 respectively and received a PhD degree in Computer Science from Iowa State University, USA in 2012. During 2012-2014, he worked as a postdoctoral researcher at the Center for Vision, Cognition, Learning and Autonomy, Departments of Statistics and Computer Science of the University of California, Los Angeles, USA. His research lies in the areas of natural language processing, machine learning, and artificial intelligence in general, with a focus on the representation, learning and application of stochastic grammars.

Xiaofeng Wang

Professor
Indiana University

Title: Innovations in Data-Centric Security

Abstract: The rapid progress in computing has produced a huge amount of data, which will continue to grow in the years to come. In this big-data era, we envision that tomorrow’s security technologies will be data-centric: new defense will become smart and proactive by using the data to understand what the attackers have already done, what they are about to do, what their strategies and infrastructures are; effective protection will be provided for dissemination and analysis of the data involving sensitive information on an unprecedented scale. In this talk, I report our first step toward this future of secure computing. We show that through effective analysis of over a million Android apps, previously unknown malware can be detected within a few seconds, without resorting to conventional Anti-Virus means such as signatures and behavior patterns. Also, by leveraging trillions of web pages indexed by search engines, we can capture tens of thousands of compromised websites (including those of government agencies like NIH, NSF and leading education institutions world-wide) by simply asking Google and Bing right questions and automatically analyzing their answers through Natural Language Processing. Our findings indicate that by unlocking the great value of data, we can revolutionize the security landscape, making tomorrow security technologies more intelligent and effective.

Bio: Dr. XiaoFeng Wang is a professor in the School of Informatics and Computing at Indiana University, Bloomington. He received his Ph.D. in Electrical and Computer Engineering from Carnegie Mellon University in 2004, and has since been a faculty member at IU. Dr. Wang is a well-recognized researcher on system and network security. His work focuses on data-centric security, mobile and cloud security, and healthcare privacy. He is a recipient of 2011 Award for Outstanding Research in Privacy Enhancing Technologies (the PET Award) and the Best Practical Paper Award at the 32nd IEEE Symposium on Security and Privacy. His work frequently receives attention from media, including CNN, MSNBC, Slashdot, CNet, PC World, etc. Examples include his discovery of security-critical vulnerabilities in API integrations (http://money.cnn.com/2011/04/13/technology/ecommerce_security_flaw/) and his study of the security flaws on the Apple platform (http://money.cnn.com/2015/06/18/technology/apple-keychain-passwords/). His research is supported by the NIH, NSF, Department of Homeland Security, the Air Force, Samsung Research and Microsoft Research. He is the director of IU’s Center for Security Informatics.

Wenyuan Xu

Professor
Zhejiang University

Title: Analog Cybersecurity of Cyber-Physical Systems—from 0101 to Mixed Signals

Abstract: Much security research focuses on protecting the digitalized information, e.g., securing communication via cryptographic methods. Nevertheless, hardware implementation and its internal signal conditioning path could undermine the otherwise secure mechanisms, e.g., attackers can extract secret keys via side channels. As the emerging cyber-physical systems depend on sensors to make automated decisions, it is critical to examine analog cybersecurity, i.e., analyzing the integrity and dependability of information prior to its digitalization. Such a problem is especially important in cyber-physical systems because they depend on sensors to make automated decisions. In this talk, we illustrate a few analog signal injection attacks that utilize the build-in hardware vulnerabilities of various commodity sensing systems as well as proposing the defense strategies. Our work calls into questioning the wisdom of allowing microprocessors and embedded systems to blindly trust that hardware abstractions alone will ensure the integrity of sensor outputs.

Bio: Wenyuan Xu is currently a professor in the College of Electrical Engineering at Zhejiang University. She received her B.S. degree in Electrical Engineering from Zhejiang University in 1998, an M.S. degree in Computer Science and Engineering from Zhejiang University in 2001, and the Ph.D. degree in Electrical and Computer Engineering from Rutgers University in 2007. Her research interests include wireless networking, embedded systems security, and IoT security. Dr. Xu received the NSF Career Award in 2009. She was granted tenure (an associate professor) in the Department of Computer Science and Engineering at the University of South Carolina in the U.S. She has served on the technical program committees for several IEEE/ACM conferences on wireless networking and security, and she is an associated editor of EURASIP Journal on Information Security. She has published over 60 papers and her papers have been cited over 3000 times (Google Scholar).

Wotao Yin

Professor
University of California, Los Angeles

Title: Theory and Applications of Asynchronous Parallel Computation

Abstract: Many problems reduce to the fixed-point problem of solving x=T(x). This talk discusses a coordinate-friendly structure in the operator T that prevails in many optimization problems and beyond. This structure enables highly efficient asynchronous algorithms. By “asynchronous”, we mean that the algorithm runs on multiple threads/processes and each computes with possibly delayed information from the others. The threads/processes do not coordinate their iterations. On modern computer architectures and distributed networks, asynchronous algorithms are order-of-magnitude faster that the standard (synchronous) parallel algorithms! We demonstrate how to solve large-scale applications in machine learning, image processing, portfolio optimization, second-order cone programs, and decentralized computing with asynchronous algorithms.
We also present theoretical results. In the most basic form, as long as T has a fixed point and is nonexpansive, the asynchronous coordinate-update algorithm converges to a fixed point under either a bounded delay or certain kinds of unbounded delays. The operator does not have to be contractive. We also discuss how to develop and tune asynchronous algorithms in practice.

Bio: Wotao Yin is a professor in the Department of Mathematics of UCLA. His research interests lie in computational optimization and its applications in image processing, machine learning, and other inverse problems. He received his B.S. in mathematics from Nanjing University in 2001, and then M.S. and Ph.D. in operations research from Columbia University in 2003 and 2006, respectively. During 2006 - 2013, he was with Rice University. He won NSF CAREER award in 2008, Alfred P. Sloan Research Fellowship in 2009, and Morningside Medal in 2016.

Yizhou Yu

Professor
The University of Hong Kong

Title: Deep Learning for Bioinformatics, Vision and Graphics

Abstract: In this talk, I present a few deep learning methods for bioinformatics, computer vision and graphics. In the first part, I briefly discuss an end-to-end deep neural network for protein secondary structure prediction, which is an important problem in bioinformatics. Considering long-range dependencies existing in amino acid sequences, we set up a bidirectional neural network consisting of gated recurrent units to capture global contextual features. Multi-task learning is utilized to further improve the performance. Our proposed deep network achieves state-of-the-art performance on both CASP10 and CASP11 datasets.
In the second part, I introduce a deep transfer learning scheme, called selective joint fine-tuning, for boosting the performance of deep learning tasks with insufficient training data. In this scheme, a target learning task with insufficient training data is carried out simultaneously with another source learning task with abundant training data. However, the source learning task does not use all existing training data. Our core idea is to identify and use a subset of training images from the original source learning task whose low-level characteristics are similar to those from the target learning task, and jointly fine-tune shared convolutional layers for both tasks. Experiments demonstrate that our deep transfer learning scheme achieves state-of-the-art performance on multiple visual classification tasks with insufficient training data for deep learning. Such tasks include Caltech 256, MIT Indoor 67, and fine-grained image classification problems (Oxford Flowers 102 and Stanford Dogs 120).
In the third part, I present a deep learning based sketching system for 3D face and caricature modeling. This system has a labor-efficient sketching interface, that allows the user to draw freehand imprecise yet expressive 2D lines representing the contours of facial features. A CNN based deep regression network is designed for inferring 3D face models from 2D sketches. Our system also supports gesture based interactions for users to further manipulate initial face models. Both user studies and numerical results indicate that our sketching system can help users create face models quickly and effectively.

Bio: Yizhou Yu is currently a full professor in the Department of Computer Science at the University of Hong Kong, and an adjunct professor in the Department of Computer Science at University of Illinois, Urbana-Champaign (UIUC). He was first a tenure-track and then a tenured professor at UIUC for more than 10 years. He has also collaborated with eBay Research, Google Brain and Microsoft Research in the past. He received his PhD degree in computer science from the computer vision group at University of California, Berkeley. He also holds a MS degree in applied mathematics and a BE degree in computer science and engineering from Zhejiang University. Prof Yu has made many important contributions to AI and visual computing, including computer vision, images and graphics, and VR/AR. He is a recipient of 2011 and 2005 ACM SIGGRAPH/EG SCA Best Paper Awards, 2007 NNSF China Overseas Distinguished Young Investigator Award, 2002 US National Science Foundation CAREER Award, and 1998 Microsoft Graduate Fellowship. Innovative technologies co-invented by him has been frequently adopted by the film and special effects industry. He has more than 100 publications in international conferences and journals. His current research interests include deep learning methods for machine intelligence, computational visual media, geometric computing, intelligent video surveillance, and biomedical data analysis.

Junsong Yuan

Associate Professor Program Director
Nanyang Technological University

Title: Towards More Intelligent Machines: Understanding Human Gestures and Actions Using RGB-D Sensors

Abstract: To make machines as smart as human beings, it is important for machines to understand human behaviors, e.g., their gestures and actions, so that machines can better sense the intentions of humans and communicate with them in a more natural way. Recently, the availability of commodity depth cameras, such as Microsoft Kinect and Intel RealSense, has brought a new level of excitement to this field. With these new sensors, rapid progresses have been made to enable new applications. In this talk, I will introduce our recent work for human behaviour understanding using RGB-D cameras. Applications in video surveillance, human-robot interaction, virtual reality, gaming, and tele-presence will also be discussed.

Bio: Junsong Yuan is currently an Associate Professor at School of Electrical and Electronics Engineering (EEE), Nanyang Technological University (NTU), Singapore. He received Ph.D. from Northwestern University in 2009. He is an Associate Editor of IEEE Trans. on Image Processing (T-IP), IEEE Trans. on Circuits and Systems for Video Technology (T-CSVT) and The Visual Computer journal (TVC). He is Program Co-Chair of ICME’18 and Area Chair of CVPR’17, ICIP’17, ICPR’16, ACCV’14, etc. He received 2016 Best Paper Award from IEEE Trans. on Multimedia (T-MM), Doctoral Spotlight Award from IEEE Conf. on Computer Vision and Pattern Recognition (CVPR'09), Nanyang Assistant Professorship from NTU, and Outstanding EECS Ph.D. Thesis award from Northwestern University.

Ming Zhou

Assistant Managing Director
Microsoft Research Asia

Title: The New Progress of Neural Machine Translation, Chatbot and Reading Comprehension

Abstract: NLP, as one of the most important technology for AI, has made unprecedented progress in recent three years with the rapid and wide use of deep learning approaches. In this presentation, I would like to present the new progress of NLP illustrated by three typical NLP tasks including machine translation (MT), chatbot and reading comprehension. I will share my observations on the challenges of current approaches and discuss the future directions.

Bio: Dr. Ming Zhou is assistant managing director of MSRA in charge of research areas of natural language processing, knowledge mining and enterprise AI, as well as social computing. He is the chair of the Chinese Computer Federation’s (CCF) Chinese Information Technology Committee and an executive member of the Chinese Information Processing Society (CIPS). He is also president-elect of Association of Computational Linguistic (ACL), the most prestigious NLP research association in the world. He developed the first Chinese-English machine translation system in 1989 at Harbin Institute of Technology, and the most famous Chinese-Japanese machine translation product in 1998 at Kodensha Ltd of Japan. He has published over 120 papers at top conferences and journals including 50+ papers at ACL. He obtained his PhD from Harbin Institute of Technology in 1991, and then worked at Tsinghua University as post-doc and associate professor until 1999. He joined Microsoft in 1999 as researcher and became the research manager of its Natural Language Computing Group in 2000.

Distinguished Speakers