Researchers have developed a new method for filling in gaps in physiological time-series data, such as blood pressure or glucose levels. The method, called Curriculum-Aware Interpolate-then-Refine (CAIR), uses two stages to learn and correct the imputed values. In the first stage, it uses a bidirectional-GRU interpolator to create a coarse base curve. Then, in three successive passes, it refines this estimate using a Transformer-based model trained on a broad range of signal-agnostic random-gap curricula. CAIR outperforms other methods under various missingness mechanisms and gap lengths, particularly when the gaps are clinically extreme or span large ranges. The researchers also found that simply reducing reconstruction error does not necessarily improve clinical metrics. --- Why it matters: This matters to engineers working on medical AI because it provides a more accurate method for filling in gaps in physiological time-series data, which is essential for making informed decisions in healthcare. Source: https://arxiv.org/abs/2608.21207