WORKSHOPS
Workshop 1
Medical AI Workshop: Making AI Safe and Healthy (MASH)
Abstract
Over the past few years, the development of large-scale foundation models has significantly advanced the state-of-the-art in machine learning. Models trained on massive datasets with labelled images or billions of image-text pairs have become available and shown to be very precise and extremely capable. Moreover, even for huge datasets without labels or annotation, unsupervised learning technologies, such as MAE (Masked Autoencoders) [He et al., 2022], DINO [Caron et al., 2021], and CLIP [Radford et al., 2021], have demonstrated strong generalization capabilities to a variety of downstream tasks across domains. These approaches follow a prevailing trend known as the scaling law, where model performance improves predictably as model size, dataset scale, and computing power increase [Kaplan et al., 2020].
One of the most critical and promising AI application areas is the medical or medicare applications. However, the above-mentioned large-scale foundation model paradigm is practically challenging when applied to medical imaging. Unlike natural image domains that enjoy nearly unlimited access to labeled and unlabeled web-scale data, the medical domain is constrained by strict privacy regulations, heterogeneous data sources, limited patient populations, and the requirement for costly, expert-level annotation. High-quality medical datasets are scarce, and when available, they often suffer from domain shifts, label imbalance, and limited modality coverage. In such settings, bluntly applying scaling laws is not only impractical, but also potentially misleading.
In this workshop, we aim to bring together experts from both areas (AI and medicine) to share their experience on research activities on medical AI and to explore promising technologies. Some of the potential issues for discussion include:
Topics of the special session include (reliable/robustness/secure learning methods), including but not limited to:
- Foundation model for small scale medical data domains
- Multi-modality machine learning for medical data
- Intelligent medicare helpers and physician co-pilots
- Robotic supporting capabilities for medical applications
This workshop will invite both international and domestic speakers. We will also openly call for research papers for presentation in the workshop. A panel session is planned to facilitate face-to-face discussions between invited medical professionals and senior AI researchers.
By holding this initial workshop devoted to the area of “medical AI”, we hope to bring AI researchers and medical professionals together in order to explore important AI research issues for medicine.
Workshop Organizer
Jane YJ Hsu, Chang Gung University, Taiwan
Kwei-Jay Lin, Chang Gung University, Taiwan
Anna Kobusinska, Poznan University of Technology, Poland
Ying-Feng Chang, Chang Gung University, Taiwan
Workshop 2
Reliable and Trustworthy Artificial Intelligence Workshop
Abstract
The increasing adoption of artificial intelligence (AI) is driving massive transformations across many sectors, such as finance, robotics, manufacturing and healthcare. It is critical to design, develop and deploy reliable and robust AI models for building trustworthy systems that offer trusted services to users with high-stakes decision-making, including AI-assisted robotic surgery, automated financial trading, and autonomous driving. Nevertheless, AI applications are vulnerable to reliability issues, such as concept drifts, dataset shifts, misspecifications, misconfiguration of model parameters, perturbations, and adversarial attacks on human or even machine comprehension levels, thereby posing tangible threats to various stakeholders at different levels. This workshop aims to draw together state-of-the-art artificial intelligence advances to address challenges for ensuring reliability, security and privacy in trustworthy systems. The following topics are welcomed but not limited to (i) trustworthy agentic AI, (ii) bias and fairness, (iii) explainability, (iv) robust mitigation of adversarial attacks, (v) improved privacy and security in model development, (vi) scalability and (vii) resource efficiency.
Workshop Organizer
Harry Nguyen, University College Cork, Ireland
Duc-Trong Le, University of Engineering and Technology, Vietnam National University, Hanoi
Xuan-Son Vu, Umeå University, Sweden
Johanna Björklund, Umeå University, Sweden
