CardioState-JEPA Learns Shared Cardiac Representations Across ECG, PPG, and PCG

CardioState-JEPA uses a joint-embedding predictive architecture to map heterogeneous waveforms from ECG, PPG, and PCG into a common token space. The model employs a single shared Transformer encoder to predict masked latent cardiac states, focusing on shared physiology rather than sensor-specific waveform patterns. This approach accounts for temporal offsets between different sensing modalities.