PulseBound Ensures Causal Information Boundaries in PPG Forecasting

PulseBound introduces a representation learner for photoplethysmography (PPG) that uses a content-independent cutoff to prevent data leakage. By enforcing prefix-only normalization and suffix replacement, the model achieves stored-suffix invariance when predicting nine rhythm and morphology descriptors.