ZHANG Panyu

Research Projects

Reproducible Mobile Sensing for Stress PredictionPublished / Completed
Built a reproducible stress prediction pipeline using mobile sensing data, focusing on transparent preprocessing, feature extraction, modeling, and evaluation. The project examined how design choices in mobile sensing pipelines affect stress prediction performance and reproducibility.
Mobile sensing; stress prediction; reproducibility; affective computing; mental health computing

Cross-User Generalization and Distribution Shift in Mobile Mental Health SensingResearch Project
Studied why mobile and wearable sensing models often fail to generalize across users. The project analyzes cross-user generalization through distribution shift, including covariate shift, label shift, and concept shift, and investigates how these shifts explain performance degradation in mental-health-related mobile sensing tasks.
Distribution shift; cross-user generalization; mobile sensing; wearable sensing; mental health; domain generalization

Multi-Wave Mobile and Wearable Affect Sensing BenchmarkResearch Project
Developed a multi-wave benchmark for in-the-wild affect sensing using smartphone sensing, wearable sensing, and dense self-report labels. The project evaluates affect prediction under multiple generalization settings, including within-user temporal prediction, cross-user generalization, and cross-wave generalization.
Benchmark; affect sensing; mobile sensing; wearable sensing; ESM; generalization

Temporal Shift and Adaptation in Mobile Stress SensingResearch Project
Investigated whether temporal shift can hurt mobile stress sensing models as much as cross-user shift. The project compares random, cross-user, and temporal evaluation settings and studies whether unsupervised domain adaptation methods can mitigate longitudinal performance degradation.
Temporal shift; domain adaptation; mobile stress sensing; longitudinal data; model robustness

Multimodal Sensor Fusion for Mental Health DetectionCollaborative Research
Contributed to research on mental health detection using multimodal sensing signals from mobile, wearable, and IoT sensors. The project focuses on combining heterogeneous sensing modalities to improve robustness and predictive performance in mental health computing.
Multimodal sensing; sensor fusion; mobile sensing; wearable sensing; IoT; mental health

Routine Extraction from Mobile, Wearable, and IoT Sensing DataSurvey / Ongoing Research
Worked on a systematic survey of routine extraction from mobile, wearable, and IoT sensing data. The project focuses on reusable per-user routine artifacts, routine extraction methods, and how visual analytics can support interpretation of behavioral routines.
Routine extraction; behavioral sensing; mobile sensing; wearable sensing; IoT; visual analytics; survey