Task-Aware Latent Domain Discovery for Non-IID Sensor Data
Mining latent, task-aware domains inside non-i.i.d. sensor time-series so that per-domain models can be trained and selected at inference. The Prism approach formulates domain partitioning as an optimization problem (shown NP-hard by reduction from weighted set cover), then approximates it with an EM-style loop: an E-step clusters encoder feature embeddings into candidate domains and an M-step retrains per-domain classifiers, yielding flexible user perception on IMU data across unseen users and devices (IEEE INFOCOM 2025, arXiv:2501.01598).
T
Twiki
Text
Prism: Mining Task-aware Domains in Non-i.i.d. IMU Data
A wide range of user perception applications leverage inertial measurement unit (IMU) data for online prediction. However, restricted by the non-i.i.d. nature of IMU data collected from mobile device…