Longitudinal data scale affects learning of changing behavior
Longer Records, Broader Invariance: The Hidden Scaling Problem in Longitudinal Contrastive Learning
Machine LearningArtificial Intelligence
Summary
When studying how people change over time using computer learning, having longer records might actually hide important changes instead of revealing them. The authors found that using data from the same person across very long time spans can confuse the learning process by making different behaviors seem the same. They figured out that controlling how far apart in time connected data points are helps keep the important changes visible. Their work shows that just having more data isn't always better unless the data is used carefully.
What this means in practice
- •For healthcare data teams: Design patient monitoring models that preserve temporal changes by controlling the time span used in model supervision on long-term health records.
- •For smart home developers: Improve smart home behavior recognition by adjusting positive pairing in learning algorithms to retain meaningful changes over long sensor data periods.
Authors
Rameen Mahmood, Xuhai "Orson" Xu, Zachary Beattie, Jeffrey Kaye, Danny Yuxing Huang
Abstract
Longitudinal data are valuable because people change. Yet the objectives used to learn from these data can inadvertently erase that change. In person-level contrastive learning, observations from the same person are treated as positives; as records grow, those positives can span increasingly distant---and increasingly different---behavioral states. More history can therefore produce not only more data, but broader invariance. We show that this distinction is fundamental. We separate \emph{record span}, how much history the learner sees, from \emph{supervision span}, how far across that history positive-pair supervision reaches. Across in-home sensing records spanning up to 2.7 years, broader supervision systematically suppresses recoverable changing-state information, even when the available history is held fixed. At the broadest span, less than 10\% of the information recoverable from an untrained encoder remains. Yet keeping positives local is not sufficient: as records grow, even distant states that are never paired become increasingly similar. Explicitly contrasting other observations from the same person reverses this loss without shortening the record, revealing a second route by which longitudinal scale can broaden invariance. Finally, we prospectively reproduce the supervision-span effect in 199 GLOBEM participants. Longitudinal scale therefore presents a choice: more history need not mean more invariance. By controlling what is held invariant as records grow, we can preserve the change that made the longitudinal data valuable in the first place.