PLENARY SPEAKERS

Ian R. Petersen (Fellow of the Australian Academy of Science, FIEEE, FIFAC)

School of Engineering, College of Engineering, Computer Science and Cybernetics, Australian National University, Canberra, Australia

Plenary Talk: Control Theory Applied to Accelerated Gradient Optimization Algorithms

This presentation will describe some recent work in which control theory methods are used to analyze some gradient based optimization methods. Gradient based optimization methods are widely used in many areas of optimization the area such as in machine learning and artificial intelligence approaches. The heavy ball optimization method is analyzed using the discrete time circle criterion for a class of non-convex cost functions and it is shown that convergence could only be guaranteed for cost functions with a condition number which less than a fixed maximum value. Then the coefficients of the original heavy ball algorithm are modified so that the circle criterion could be used to guarantee convergence for all cost functions in the class being considered, at the expense of a slightly slower convergence rate. Also, an additional term is added to the heavy ball algorithm to obtain an improved guaranteed convergence rate. This leads to an optimization algorithm with a similar form to the triple momentum method (TMM). However, with our choice of parameters, a faster convergence rate is obtained than the TMM method. In addition, a control theory result on optimal gain margin is used to prove that the original heavy ball method has the fastest convergence rate of any linear gradient base optimization method for strictly convex quadratic cost functions. In addition, the case of quadratic cost functions with moving optimal points is addressed. Such problems arise in the area of online convex optimization algorithms. By introducing integral action into the optimization algorithm, an algorithm can be obtained with zero steady state error. For the case of optimization algorithms having the same form as the heavy ball method, this leads to an algorithm with a convergence rate which is slower than the standard gradient algorithm. However, by applying a control theory method for optimal gain margin, a more complicated algorithm can be derived with the optimal convergence rate.


Biography


Ian R. Petersen (Fellow, IEEE) received a Ph.D. in Electrical Engineering in 1984 from the University of Rochester, USA. From 1983 to 1985, he was a Postdoctoral Fellow at the Australian National University. In 1985, he joined the University of New South Wales, Canberra, Australia. He moved to The Australian National University in 2017 where he is currently a Professor in the School of Engineering. He was the Australian Research Council Executive Director for Mathematics, Information and Communications in 2002 and 2003. He was Acting Deputy Vice-Chancellor Research for the University of New South Wales in 2004 and 2005. He held an Australian Research Council Professorial Fellowship from 2005 to 2007, an Australian Research Council Federation Fellowship from 2007 to 2012, and an Australian Research Council Laureate Fellowship from 2012 to 2017. He is a fellow of IFAC, the IEEE and the Australian Academy of Science.


His main research interests are in robust control theory, quantum control theory, and stochastic control theory. He served as an Associate Editor for the IEEE Transactions on Automatic Control, Systems and Control Letters, Automatica, and SIAM Journal on Control and Optimization. He was an Editor for Automatica in the area of optimization in systems and control. He was elected IFAC Council Member for the 2014–2017 Triennium and was also elected to be a member of the IEEE Control Systems Society Board of Governors for the periods 2011–2013 and 2015–2017. He was a Vice President of IEEE Control Systems Society 2023-2024. He is the Vice President for Technical Activities of IEEE Control Systems Society. He was General Chair of the 2015 IEEE Multi- Conference on Systems and Control.

Sam Tak-Wu Kwong (Fellow of Canadian Academy of Engineering, FIEEE, FNAI)

Department of Computing and Decision Science, Lingnan University, Hong Kong, China

Plenary Talk: High Dynamic Range video

High Dynamic Range (HDR) video is a technology that significantly enhances the visual experience by expanding the range of contrast and color in video content. Unlike standard dynamic range (SDR) video, HDR allows for brighter highlights, deeper shadows, and a wider color gamut. This results in more realistic and vibrant images that closely mimic the way the human eye perceives the real world. In this segment, we will delve into the fundamental principles of HDR, exploring how it works, the technical standards behind it (such as HDR10, Dolby Vision, and HLG), and the benefits it brings to various types of content, from movies and TV shows to video games and live broadcasts In this talk, I will talk about the following:

HDR Image Reconstruction High dynamic range (HDR) image reconstruction is a process that aims to create images with a greater range of luminance levels than what is achievable with standard digital imaging techniques. This allows for the capture of both very bright and very dark details in a scene, closely mimicking human vision. By combining multiple images taken at different exposure levels, HDR reconstruction techniques can produce visually stunning and highly detailed images that better represent the range of light present in real-world scenes.

EIN: Exposure Induced Network for Single Image HDR Reconstruction
The Exposure Induced Network (EIN) for Single Image HDR Reconstruction is a novel deep learning approach designed to generate HDR images from a single standard dynamic range (SDR) input. Unlike traditional methods that require multiple exposures, EIN leverages a neural network to predict and reconstruct HDR content by learning the relationships between different exposure levels. This enables the creation of high-quality HDR images even in situations where only a single exposure is available, making HDR imaging more practical and accessible for various applications.

LGFM: HDR Image Quality Assessment Based on Frequency Disparity
The Local and Global Frequency Modulation (LGFM) method for HDR image quality assessment is a sophisticated approach that evaluates the quality of HDR images based on the disparity in frequency components. By analyzing both local and global frequency information, LGFM can more accurately reflect the human visual system's sensitivity to different types of artifacts and distortions in HDR content. This results in a more reliable and comprehensive assessment of HDR image quality.


Biography


Professor KWONG Sam Tak Wu is the Associate Vice-President (Strategic Research), J.K. Lee Chair Professor of Computational Intelligence, the Dean of the School of Graduate Studies and the Acting Dean of the School of Data Science of Lingnan University. Professor Kwong is a distinguished scholar in evolutionary computation, artificial intelligence (AI) solutions, and image/video processing, with a strong record of scientific innovations and real-world impacts. Professor Kwong was listed as the World’s Top 2% Scientists by Stanford University since 2021 and one of the most highly cited researchers by Clarivate in 2022 and 2023. He has also been actively engaged in knowledge transfer between academia and industry. He was elevated to IEEE Fellow in 2014 for his contributions to optimization techniques in cybernetics and video coding. He was a Fellow of the Asia-Pacific Artificial Intelligence Association (AAIA) in 2022, and the President of the IEEE Systems, Man, and Cybernetics Society (SMCS) in 2021-23. He is a fellow of US National Academy of Inventors (NAI) and the Hong Kong Academy Awards of Engineering and Sciences (HKAES). Professor Kwong has a prolific publication record with over 350 journal articles, and 160 conference papers with an h-index of 90 based on Google Scholar. He is currently the associate editor of a number of leading IEEE transaction journals.

Jian CHU (Founder of SUPCON Tech Co)

Plenary Talk: TPT-Driven Industrial Intelligence: AI for Safer, Higher-Quality and Cheaper Process Manufacturing

In process industries, particularly petrochemicals, massive time-series production data serve as the fundamental 'fuel' for AI advancement. Leveraging this data, the pre-trained Time-series Transformer (TPT) model has been engineered to transcend conventional process optimization tools like APC and RTO. By establishing a closed-loop of perception, decision-making, and execution, the TPT model plays a pivotal role in enhancing operational safety through early anomaly detection, reducing production costs via energy and material optimization, and improving product quality through precise real-time control.


Biography


Jian CHU was the Professor in process control in Zhejiang University. He founded SUPCON in 1993 starting with DCS. In the past years, he focuses on industrial intelligence in process industries with massive time-series data. He was honored with three times second-class National Awards for Sci-Tech Progress and Inventions.