The field of cardiac health research is experiencing a significant shift towards multimodal learning approaches, which integrate diverse data sources such as medical images, structured records, and physiological time series to improve diagnosis and treatment outcomes. Nonlinear dimensionality reduction techniques, such as manifold learning, are being explored for their potential to identify medically relevant features in ECG signals without the need for training or prior information. Variational learning frameworks, like DISCoVeR, are also being developed to learn disentangled representations that separate factors of variation shared across experimental conditions from those that are condition-specific. Furthermore, causal reasoning and de-confounding techniques are being applied to multimodal data to identify at-risk individuals and predict cardiovascular adverse events. Noteworthy papers in this area include:

• Manifold Learning for Personalized and Label-Free Detection of Cardiac Arrhythmias, which demonstrates the potential of nonlinear dimensionality reduction for cardiac monitoring.
• MOSCARD, which proposes a novel predictive modeling framework that integrates multimodal causal reasoning and co-attention to align CXR and ECG data.
• Contrastive Cross-Modal Learning for Infusing Chest X-ray Knowledge into ECGs, which introduces a framework for learning clinically informative ECG representations using chest X-rays during training.
• Causal Representation Learning with Observational Grouping for CXR Classification, which presents a method for learning identifiable causal representations for disease classification in chest X-rays.