(+86) 18081079313; (+86)28-86256789
![]() |
Prof. Weisi Lin FIEEE, FIET, CEng, Hon.
FSIET
|
Keynote: "Deep-learnt Features to Facilitate Image Compression and Computer Vision with an Integrated Framework" Abstract: It has been long for image (or video) compression and computer vision to be two largely separate domains, and as a result, a computer vision task typically can only start after a whole image is decoded. This talk explores for intermediate deep-learnt visual features (rather than whole image/video) to be extracted and then coded, and this facilitates integration of signal compression and computer vision, accurate feature extraction, privacy preservation, flexible load distribution between edge and cloud, and green visual computing. It is hoped that the presentation can trigger more R&D in the related fields due to the nature of fundamental paradigm-shift in the proposed framework. Short bio: |
![]() |
Prof. Witold Pedrycz IEEE Life Fellow
|
Keynote: " Federated Learning, Knowledge Transfer, and Knowledge Distillation: Developments with Information Granules" Abstract: The visible trends of Machine Learning
addressing the emerging needs of coping with the
diversity of real-world motivated learning scenarios
involve federated learning, transfer learning, and
knowledge distillation. Short bio: |
![]() |
Prof. Nam Ling IEEE Fellow, IET Fellow
|
Keynote: "Visual Coding – From Traditional Approach to Deep Learning Approach" Abstract: In today’s Internet, visual data are everywhere. Eighty percent of all Internet traffic are from video data. The immensity of the size and amount of visual data dictates the need of efficient coding technology to effectively compress them. From the first video coding standard in the mid-1980s to the latest VVC/H.266 in 2020, coding efficiency has improved a lot. Traditional video codec has been based on a block-based hybrid codec structure. With the advancement of deep learning technology, video coding can be assisted by deep learning tools or/and use deep learning-based neural network as the backbone. The improvement in coding efficiency comes with huge computational complexity associated with deep approaches and the need of a more appropriate visual quality metric. Image coding is similar. From the early JPEG to BPG to VVC intra, coding efficiency has improved a lot. Deep learning approaches further improve this but often come with a high computational complexity. In this talk, we first discuss the key tools in the block-based hybrid codec structure, then discuss deep learning-based approaches, from autoencoder to the use of generative adversarial network (GAN). Finally, we highlight a couple of our on-going projects based on GANs, including the use of GANs in image coding and in video coding. Short bio: |
![]() |
Prof. Kezhi Mao Nanyang Technological
University,
|
Short bio: |
![]() |
Prof. Robert Minasian IEEE Fellow & OSA
Fellow
|
Keynote: "Advances in Integrated Photonic Signal Processing " Abstract: Photonic signal processing enables the realisation of functions that are difficult or not even possible to be achieved using electronic techniques. This exploits the inherent advantages of photonics including wide bandwidth and immunity to electromagnetic interference. Photonic signal processors can provide in-built signal conditioning for overcoming a range of challenging problems in the processing of high-speed signals. Recently, there has been a significant global drive to achieve integration of photonic signal processors on silicon platforms, especially since this leverages the CMOS fabrication technology to enable boosting the performance of future systems performing communications, sensing and deep learning with the potential for implementing high bandwidth, fast and complex functionalities. Advances in integrated photonic signal processing are presented. These include dense optical integration techniques for LIDAR on a chip systems, photonic neural networks for deep learning, multi-function signal processors, programmable integrated photonic processors and high-resolution integrated sensors for IoT. These photonic processors herald new capabilities for achieving high-performance signal processing. Short bio: |