06 May 2026
Tianyi and Mingcheng at ICASSP2026
On 6 May 2026, Tianyi and Mingcheng gave a poster presentation on their joint research at ICASSP 2026 (the IEEE International Conference on Acoustics, Speech, and Signal Processing), held in Barcelona, Spain. The conference brought together international experts in artificial intelligence, machine learning, and signal processing to discuss cutting-edge advancements and the fundamental applications of deep learning methodologies.
Their work, titled “Cross-representation benchmarking in time-series electronic health records for clinical outcome prediction”, explored the critical challenge of patient data representation choices for deep learning-based clinical predictions. Their work introduces the first systematic benchmark for rigorously comparing various Electronic Health Record (EHR) representation methods, including multivariate time-series, event streams, and textual streams for large language models (LLMs). By evaluating an array of modelling families across distinct clinical settings, such as ICU care and longitudinal tracking, their analysis reveals that event stream models consistently deliver the strongest performance. This unified, reproducible pipeline has the potential to provide highly practical guidance for clinical AI researchers, ensuring the selection of the most effective and reliable data representations based on specific medical contexts and data availability.
