MedFlow: Multi-Scale Flow Matching for Medical Time-Series Synthesis
September 7, 2026
MedFlow uses a vector-quantized multi-scale tokenizer and Token Marginal Guidance to generate synthetic medical time-series data. The framework captures both coarse clinical trends and fine-grained temporal dynamics to improve downstream utility on imbalanced medical datasets.
HOW THIS AFFECTS YOU
●
researcherThe multi-scale approach addresses temporal resolution challenges in medical sequence modeling.
●
healthHigher fidelity synthetic data could accelerate clinical prediction model training where real data is scarce.