|
Past Seminars and Events
|
| November 11, 2019 |
-
Title: Computer Vision of Refractive Media
Time: 11:00am
Venue: Room 328, Chow Yei Ching Building, The University of Hong Kong
Speaker(s): Professor Herb Yang, Department of Computing Science, University of Alberta
Remark(s): Abstract:
Computer vision of opaque objects has been extensively studied. However, there is signicantly less attention on refractive media. In this talk, I will give an overview of my recent research in the area of computer vision of refractive media, which includes solids and uids. My interest in this topic began several years ago when my group was asked to develop an undersea 3D vision system for Neptune Canada, which had been merged with Venus Canada to form Ocean Networks Canada. During our research, we discovered that simply applying land-based computer vision techniques to undersea appears trival but is, unfortunately, incorrect. Surprisingly, most photogrammetry methods at the time incorrectly adapted land-based methods to undersea applications with minor tweaking of parameters. By accommodating refraction in our methods, we have developed several physics based algorithms that outperform the accuracy of existing algorithms. Rather than a hindrance, we also discovered that refraction can be leveraged in underwater imaging. For example, we take advantage of dispersion to calibrate an underwater camera with improved accuracy. As well, dispersion can also be used to reconstruct an object in 3D with only one single view, i.e. one single camera, which is not possible for a typical land-based camera. More recently, motivated by our underwater results, we further explore developing new physics based methods to reconstruct shapes of transparent objects, which include transparent solids and dynamic water surfaces and underwater scenes.
About the Speaker:
Herb Yang (SM IEEE) received his B.Sc. (first honours) from the University of Hong Kong, his M.Sc. from Simon Fraser University, and his M.S.E.E. and Ph.D. from the University of Pittsburgh. He was a faculty member in the Department of Computer Science at the University of Saskatchewan from 1983 to 2001 and served as Graduate Chair from 1999 to 2001. Since July, 2001, he has been a Professor in the Department of Computing Science at the University of Alberta. He served as Associate Chair (Graduate Studies) in the same department from 2003 to 2005 and as Science Internship Director from 2016 to 2019. His research interests cover a wide range of topics from computer graphics to computer vision, which include physically based animation of Newtonian and non-Newtonian fluids, texture analysis and synthesis, human body motion analysis and synthesis, computational photography, stereo and multiple view computer vision, underwater imaging and medical imaging. He has published over 150 papers in international journals and conference proceedings in the areas of computer vision, computer graphics and medical imaging. He is a Senior Member of the IEEE and serves on the Editorial Board of the journal Pattern Recognition and IET-Computer Vision. In addition, he has served as reviewer and program committee member to many international conferences and reviewer to many international journals, and funding agencies. In 2007, he was invited to serve on the expert review panel to evaluate computer science research in Finland.
All are welcome!
For enquiries, please call 2859 2180 or emailenquiry@cs.hku.hk
PDF
|
| November 08, 2019 |
-
Title: A Mutual Information Maximization Perspective of Language Representation Learning
Time: 03:00pm
Venue: Room 308, Chow Yei Ching Building, The University of Hong Kong
Speaker(s): Dr Lingpeng Kong, Senior Research Scientist, Google DeepMind
Remark(s): Abstract:
In this talk, we show state-of-the-art word representation learning methods maximize an objective function that is a lower bound on the mutual information between different parts of a word sequence (i.e., a sentence). Our formulation provides an alternative perspective that unifies classical word embedding models (e.g., Skip-gram) and modern contextual embeddings (e.g., BERT, XLNet). In addition to enhancing our theoretical understanding of these methods, our derivation leads to a principled framework that can be used to construct new self-supervised tasks. We provide an example by drawing inspirations from related methods based on mutual information maximization that have been successful in computer vision, and introduce a simple self-supervised objective that maximizes the mutual information between a global sentence representation and n-grams in the sentence. Our analysis offers a holistic view of representation learning methods to transfer knowledge and translate progress across multiple domains (e.g., natural language processing, computer vision, audio processing).
About the Speaker:
Lingpeng Kong is Senior Research Scientist at Google DeepMind. His research focuses on the computational modeling of structures in natural language processing (NLP) with applications related to sequence labeling, syntactic parsing, and machine translation. He received his Ph.D. from Carnegie Mellon University where he was co-advised by Professor Noah Smith and Professor Chris Dyer.
All are Welcome!
Tel: 2859 2180 for enquiries
PDF
|
| November 06, 2019 |
|
| November 01, 2019 |
|
| October 28, 2019 |
|
| October 24, 2019 |
|
| September 27, 2019 |
-
Title: What is Virtual Bank? (Co-organized with HKUGA)
Time: 05:30pm
Venue: Room MBG07, Main Building, The University of Hong Kong
Speaker(s): Mr. Lawrence Li & Dr. S.M. Yiu
Remark(s):
-
Title: Using and reusing coherence to realize quantum processes
Time: 02:00pm
Venue: Rm 308, Chow Yei Ching Building, The University of Hong Kong
Speaker(s): Dr. Matteo Rosati, Universitat Autonoma de Barcelona
Remark(s): Abstract:
Using and reusing coherence to realize quantum processes Coherent superposition is a key feature of quantum mechanics that underlies the advantage of quantum technologies over their classical counterparts. Recently, coherence has been recast as a resource theory in an attempt to identify and quantify it in an operationally well-defined manner.Here we study how the coherence present in a state can be used to implement a quantum channel via incoherent operations and, in turn, to assess its degree of coherence. We introduce the robustness of coherence of a quantum channel-which reduces to the homonymous measure for states when computed on constant-output channels-and prove that: i) it
quantifies the minimal rank of a maximally coherent state required to implement the channel; ii) its logarithm quantifies the amortized cost of implementing the channel provided some coherence is recovered at the output; iii) its logarithm also quantifies the zero-error asymptotic cost of implementation of many independent copies of a channel. We also consider the generalized problem of imperfect implementation with arbitrary resource states. Using the robustness of coherence, we find that in general a quantum channel can be implemented without employing a maximally coherent resource state. In fact, we prove that every pure coherent state in dimension larger than 2, however weakly so, turns out to be a valuable resource to implement some coherent unitary channel. We illustrate our findings for the case of single-qubit unitary channels.
About the Speaker:
Matteo Rosati did his BSc and MSc studies in Physics (2009-2014) at Università La Sapienza, Rome, studying the modelling of disordered and complex systems under the supervision of Prof. Giorgio Parisi. He took his PhD in Theoretical Physics (2017) at Scuola Normale Superiore,
Pisa with Prof. Vittorio Giovannetti, with a thesis aimed at devising efficient and implementable decoders for classical communication on quantum guassian channels. Since then, he has been a postdoctoral fellow at the Universitat Autonoma de Barcelona, working with Profs. Andreas
Winter and John Calsamiglia on resource theories and quantum learning.
In 2019 he has been awarded a Marie Skłodowska-Curie Fellowship from the EU, starting in January 2020.
All are welcome!
For enquiries, please call 2859 2180 or email enquiry@cs.hku.hk
PDF
|
| September 06, 2019 |
-
Title: The Power of Data Analytics and AI Techniques in the Digital Sector
Time: 05:30pm
Venue: Lecture Theatre A, Ground Floor, Chow Yei Ching Building, Main Campus, HKU
Speaker(s): Mr Alan Chan
Remark(s): Speaker:
Mr Alan Chan
Executive Vice President Lazada (Alibaba's SE Asia Commerce Business)
Date: September 6, 2019 (Friday)
Time: 5:30 - 6:45pm (Refreshments will be served from 5:00pm)
Venue: Lecture Theatre A, Ground Floor, Chow Yei Ching Building, Main Campus, HKU
About the talk:
In this talk, Mr Alan Chan will introduce how data analytics and AI techniques are used in the digital sector, and the changes that the industry is facing now and in the future. He will also share some tips on starting a career in data and analytics.
About the speaker:
With a background in strategy and analytics, and having led several organisations through their digital transformations, Alan is the Executive Vice President in Lazada (Alibaba’s South East Asia Commerce Business) and also part of the Alibaba Management Council. Alan joined Alibaba Group in 2016 and took on management roles in marketplace policy setting, data analytics and platform governance.
Before joining Alibaba, he spent 13 years in consulting with Accenture and left in 2016 as the Managing Director and Partner of Accenture Digital team in China. Alan is passionate about leadership, digital marketplaces and data science.
Outside of work, Alan engages actively in university collaborations and serves on the ex-officio board of a few start-ups in Asia. He received his Honors Degree in Economics and Statistics from the National University of Singapore, and is currently residing in Singapore.
PDF
|
| August 28, 2019 |
-
Title: Learning Neural Character Controllers from Motion Capture Data
Time: 03:00pm
Venue: Room 328, Chow Yei Ching Building, The University of Hong Kong
Speaker(s): Prof. Taku Komura
Remark(s): Prof. Taku Komura
Institute of Perception, Action and Behaviour
School of Informatics
University of Edinburgh
Date: August 28, 2019 Wednesday
Time: 3:00pm
Venue: Room 328 Chow Yei Ching Building The University of Hong Kong
Abstract:
I will cover our recent development of neural network-based character controllers. Using neural networks for character controllers significantly increases the scalability of the system - the controller can be trained with a large amount of motion capture data while the run-time memory can be kept low. As a result, such controllers are suitable for real-time applications such as computer games and virtual reality systems. The main challenge is in designing an architecture that can produce movements in production-quality and also manage a wide variation of motion classes. Our development covers lowlevel locomotion controllers for bipeds and quadrupeds, which allow the characters to walk, run, sidestep and climb over uneven terrain, as well as a high level character controller for humanoid characters to interact with objects and the environment, which allows the character to sit on chairs, open doors and carry objects. In the end of the talk, I will discuss about the open problems and future directions of character animation.
About the speaker:
Taku Komura is a Professor at the Institute of Perception, Action and Behaviour, School of Informatics, University of Edinburgh. As the leader of the Computer Graphics and Visualization Unit his research has focused on data-driven character animation, physically-based character animation, crowd simulation, cloth animation, anatomy-based modelling, and robotics. Recently, his main research interests have been the application of machine learning techniques for animation synthesis. He received the Royal Society Industry Fellowship (2014) and the Google AR/VR Research Award (2017).
PDF
|
