However, existing contrastive learning methods primarily focus on one single data level, which fails to fully exploit the intricate nature of medical time series.

Sequential decision-making problems typically require significant human supervision and data.

These authors contributed equally to this work.