Frontier of Artificial Network
A Series of Invited Talks @ FAN Group, CityU

Biography: Huiyu Zhou received a Bachelor of Engineering degree in Radio Technology from Huazhong University of Science and Technology of China, and a Master of Science degree in Biomedical Engineering from University of Dundee of United Kingdom, respectively. He was awarded a Doctor of Philosophy degree in Computer Vision from Heriot-Watt University, Edinburgh, United Kingdom. Dr. Zhou currently is a full Professor at School of Computing and Mathematical Sciences, University of Leicester, United Kingdom. He has published over 600 peer-reviewed papers in the field. His research work has been or is being supported by UK EPSRC, ESRC, AHRC, MRC, EU, Innovate UK, Royal Society, British Heart Foundation, Leverhulme Trust, Puffin Trust, Alzheimer’s Research UK, Invest NI and industry. Homepage: https://le.ac.uk/people/huiyu-zhou

Abstract: Image completion is a challenging task, particularly when ensuring that generated content seamlessly integrates with existing parts of an image. While recent diffusion models have shown promise, they often struggle with maintaining coherence between known and unknown (missing) regions. This issue arises from the lack of explicit spatial and semantic alignment during the diffusion process, resulting in content that does not smoothly integrate with the original image. Additionally, diffusion models typically rely on global learned distributions rather than localized features, leading to inconsistencies between the generated and existing image parts. In this work, we propose ConFill, a novel framework that introduces a Context-Adaptive Discrepancy (CAD) model to ensure that intermediate distributions of known and unknown regions are closely aligned throughout the diffusion process. By incorporating CAD, our model progressively reduces discrepancies between generated and original images at each diffusion step, leading to contextually aligned completion. Moreover, ConFill uses a new Dynamic Sampling mechanism that adaptively increases the sampling rate in regions with high reconstruction complexity. This approach enables precise adjustments, enhancing detail and integration in restored areas. Extensive experiments demonstrate that ConFill outperforms current methods, setting a new benchmark in image completion.

Biography: Ren Wang is an Assistant Professor in the Department of Electrical and Computer Engineering at the Illinois Institute of Technology and a faculty member of the Institute for Data, Econometrics, Algorithms, and Learning. His research focuses on Trustworthy Machine Learning, Population-Based Machine Learning, and applications of ML in smart grids, biology, and healthcare. His long-term vision is to develop next-generation trustworthy and intelligent machine learning systems that accelerate progress in engineering and scientific discovery. Dr. Wang has published widely in top-tier conferences and journals spanning machine learning, signal processing, computer vision, control, power systems, and bioinformatics, and has served as an area chair for several premier conferences. He received the 2023 ORAU Ralph E. Powe Junior Faculty Enhancement Award and, as Principal Investigator, has led multiple research projects supported by U.S. federal agencies such as the NSF and DoE.

Abstract: Adversarial training typically views robustness as a zero-sum game, where securing a model against one attack vector often compromises generalization to others or degrades clean accuracy. This talk proposes a shift from monolithic defense to collective intelligence, introducing a population-based framework for building resilient deep models. I will first present Efficient Robust Mode Connectivity (ERMC), which demonstrates that models robust to disparate Lp attacks can be unified via continuous, low-loss paths in the weight space, allowing for a shared, multi-norm defense. I will then discuss the Dual-Model Mixture-of-Experts, an architecture that combines clean and robust experts within an architecture to balance accuracy and robustness. Together, these works illustrate that population diversity, spanning both attack types and architectural specializations, enables a level of systemic resilience that exceeds the capabilities of any single, isolated model.