Medical Time Series
Existing time-series backbones usually output length-scaling feature sequences rather than patient-level phenotypes optimized for interpretability. Subject-wise splits were used to avoid leakage in evaluation. Medical time series are descr…
1 sources - 5 claims
Existing time-series backbones usually output length-scaling feature sequences rather than patient-level phenotypes optimized for interpretability. Subject-wise splits were used to avoid leakage in evaluation. Medical time series are described as noisy, high-dimensional, variable-length signals. The experiments included ECG, EEG, and smartphone IMU datasets. The article argues that medical representations should be compact, sufficient, and disentangled.