GEM: Empowering MLLM for Grounded ECG Understanding with Time Series and Images

Xiang Lan, Feng Wu, Kai He, Qinghao Zhao, Shenda Hong, Mengling Feng

Project Contributions

• First Unified Multimodal ECG Model.
• First High-granularity ECG Grounding Dataset.
• Clinically Oriented Diagnostic System.

[Page][Code]

Towards Enhancing Time Series Contrastive Learning: A Dynamic Bad Pair Mining Approach

Xiang Lan, Hanshu Yan, Shenda Hong, Mengling Feng

Project Contributions

• First study to investigate the bad positive pair problem exists in time series contrastive learning.
• A simple yet effective algorithm designed as a lightweight plug-in.
• Enhancing the performance of existing state-of-the-art methods.

[Paper][Code]

Learning the Unlearned: Mitigating Feature Suppression in Contrastive Learning

Jihai Zhang*, Xiang Lan*, Xiaoye Qu, Yu Cheng, Mengling Feng, Bryan Hooi

Project Contributions

• First multistage contrastive learning framework.
• First work to discuss and address feature suppression in both unimodal and multimodal contrastive learning.
• Adaptable to various contrastive learning settings.

[Paper][Code]

Intra-Inter Subject Self-Supervised Learning for Multivariate Cardiac Signals

Xiang Lan, Dianwen Ng, Shenda Hong, Mengling Feng

Project Contributions

• First work that integrates medical knowledge into self-supervision to boost the performance of cardiac arrhythmias diagnosis.
• Novel intra and inter subject self-supervision mechanism.
• State-of-the-art performance.

[Paper][Code]