Yucong and Jinning Complete Summer Research Internships with AI4DH

Our summer students Yucong Cao and Jinning Liang have completed their research internships with the AI4DH Lab at the University of Oxford. Yucong's internship was supported by the Department of Engineering Science's Engineering Undergraduate Research Opportunities Programme (EUROP), while Jinning's placement was funded through the Engineering and Physical Sciences Research Council (EPSRC) Vacation Placement scheme. Both students contributed to ongoing research projects exploring the interpretability and reliability of artificial intelligence in healthcare.

Yucong worked alongside Chenqi Li and Tingting Zhu on understanding what electroencephalogram (EEG) foundation models learn internally. His research investigated the use of sparse autoencoders (SAEs) to identify physiologically meaningful features across different EEG foundation models. The study revealed that seemingly interpretable features may not reliably correspond to the physiological signals they appear to represent, highlighting the importance of rigorous validation when interpreting AI models. The findings are presented in the paper When Is an SAE Feature Interpretable? A Validation Ladder for EEG Foundation Models.

Jinning worked with Mingcheng Zhu and Tingting Zhu on investigating how vision-language models (VLMs) utilise electrocardiogram (ECG) images for clinical prediction. His research identified a phenomenon termed ECG Mirage, where models appear capable of processing multimodal clinical data but fail to make meaningful use of patient-specific ECG information. To address this issue, the team developed a visual prompt tuning approach that encourages models to better incorporate ECG information into their predictions. The work highlights the importance of ensuring that multimodal AI systems genuinely utilise relevant clinical information, as presented in ECG Mirage: Revealing and Mitigating the Underutilisation of ECGs in Vision-Language Models for Clinical Prediction

Both internships provided valuable opportunities for students to engage in academic research and contribute to the development of trustworthy AI for healthcare. We sincerely thank Yucong and Jinning for their hard work and contributions throughout the summer, and wish them every success in their future studies and research.