Keynotes
Optimized Selection and Compression of Discriminative Features
Recently deep learning has achieved great success, but the parameter number and computational complexity have increased greatly, so it is difficult to realize the lightweight deployment in some scenarios, such as edge applications. Feature selection and compression can not only reduce the amount of calculation, but also can reduce the model weight, template size. This speech introduces a variety of advanced optimized selection and compression of discriminative features, which provides many ideas for the lightweight designs of the related models and algorithms.
Abstract
How can we create technologies to help us reflect on and potentially change our behavior, as well as improve our health and overall wellbeing both at work and at home? In this talk, I will briefly describe the last several years of work our research team has been doing in this area. We have developed wearable technology to help families manage tense situations with their children, mobile phone-based applications for handling stress and depression, as well as automatic stress sensing systems plus interventions to help users just in time. The overarching goal in all of this research is to develop intelligent systems that work with and adapt to the user so that they can maximize their personal health goals and improve their wellbeing.
Prof. Lu LENG
School of Software, Nanchang Hangkong University, China
Wong Wei Kitt
Curtin University Malaysia
Priority-Based Multi-Objective Stochastic Optimisation for Watermarking and Encryption: A New Perspective
Digital watermarking and cryptographic encryption both require balancing multiple, often conflicting objectives that vary in importance depending on application context. In watermarking, imperceptibility, robustness, and security must coexist, while in cryptographic design, substitution boxes (S-boxes) must satisfy criteria such as nonlinearity, differential uniformity, avalanche, and independence. Existing optimisation approaches—typically weighted-sum or Pareto-based—treat these objectives as fixed trade-offs or equally negotiable, failing to capture real deployment scenarios where priorities shift across domains. This work introduces a priority-based lexicographic multi-objective framework, implemented through a Self-Adaptive Differential Evolution algorithm, to explicitly encode hierarchical preferences and adapt thresholds dynamically once requirements are satisfied. The framework ensures that high-priority criteria are never compromised for lower ones, while retaining the exploration capability needed to handle complex search spaces. By applying a unified, priority-driven stochastic optimisation strategy to both watermarking and encryption, this perspective highlights a flexible pathway towards security solutions that are aligned with policy, application demands, and real-world constraints.
His current research interests include biomedical engineering, biomedical informatics, ehealth, AAL, personalised health, biosignal analysis, medical imaging, and neurosciences. He has published more than 500 papers in peer-reviewed international journals, books and conference proceedings out of which over 160 as full peer review papers in indexed international journals. He has developed graduate and undergraduate courses in the areas of (bio)medical informatics, biomedical signal processing, personal health systems, physiology and biological systems simulation.
He has served as a Reviewer in CEC AIM, ICT and DGRT D-HEALTH technical reviews and as reviewer, associate editor and editorial board member in more than 20 international journals, and participated as Coordinator or Core Partner in over 45 national and EU and US funded competitive research projects attracting more than 16 MEUROs in funding. He has served as president of the EAMBES in 2008-2010. Dr. Maglaveras has been a member of the IEEE, AMIA, the Greek Technical Chamber, the New York Academy of Sciences, the CEN/TC251, Eta Kappa Nu and an EAMBES Fellow.
Abstract
The last years saw a steep increase in the number of wearable sensors and systems, mhealth and uhealth apps both in the clinical settings and in everyday life. Further large amounts of data both in the clinical settings (imaging, biochemical, medication, electronic health records, -omics), in the community (behavioral, social media, mental state, genetic tests, wearable driven bio-parameters and biosignals) as well as environmental stressors and data (air quality, water pollution etc.) have been produced, and made available to the scientific and medical community, powering the new AI/DL/ML based analytics for the identification of new digital biomarkers leading to new diagnostic pathways, updated clinical and treatment guidelines, and a better and more intuitive interaction medium between the citizen and the health care system.
Thus, the concept of connected and translational health has started evolving steadily, connecting pervasive health systems, using new predictive models, new approaches in biological systems modeling and simulation, as well as fusing data and information from different pipelines for more efficient diagnosis and disease management.
In this talk, we will present the current state-of-the-art in personalized health care by presenting cases from COVID-19 and COPD patients using advanced wearable vests and new technology sensors including lung sound and EIT, new outcome prediction models in COVID-19 ICU patients fusing X-Rays, lung sounds, and ICU parameters transformed via AI/ML/DL pipelines, new approaches fusing environmental stressors with -omics analytics for chronic disease management, and finally new ML/AI-driven methodologies for predicting mental health diseases including suicidality, anxiety, and depression.
A data-driven deterministic algorithm for multimodal global optimization
In many fields of science, economics, engineering and medicine, practical applications can be formulated as global optimization problems, such as various cost functions to be minimized. Global optimization algorithms are generally divided into deterministic and stochastic types, and the difference lies in whether the method provides theoretical guarantees. Although in literature, stochastic methods offer more satisfactory precision and acceptable efficiency than deterministic methods in many cases, the disadvantage is poor repeatability. Here, we introduce a practical and efficient deterministic method for general global optimization. It provides fast adaptive estimates for the upper bound of the change rate of an objective function and a fast exhaustive search method. It is applicable to objective functions of all dimensions and theoretically guarantees that the global minimum together with the complete set of minimizers can be found. Since global optimization is an NPHard problem, for high-dimensional data, parallel processing can provide support for finding global extreme values ​​more effectively.
Abstract
How can we create technologies to help us reflect on and potentially change our behavior, as well as improve our health and overall wellbeing both at work and at home? In this talk, I will briefly describe the last several years of work our research team has been doing in this area. We have developed wearable technology to help families manage tense situations with their children, mobile phone-based applications for handling stress and depression, as well as automatic stress sensing systems plus interventions to help users just in time. The overarching goal in all of this research is to develop intelligent systems that work with and adapt to the user so that they can maximize their personal health goals and improve their wellbeing.
Dr. Liming Zhang (張立明)
Faculty of Science and Technology, University of Macau, China
