Interpretable ML
Affective computing and health-adjacent applications.
Temporal Contrastive Learning for Interpretable Facial Emotion Recognition
B.S. thesis. Two-stage approach: NT-Xent temporal contrastive pretraining over short- and long-range views of 38-dimensional OpenFace features, then a four-layer Transformer encoder with attention-based temporal aggregation.
Real-Time Depression-Risk Monitoring via Micro-Expressions
Privacy-aware, on-device analysis of facial dynamics (Action Unit asymmetry, blink rate, head motion) for depression-risk signal monitoring. Explicitly not a diagnostic tool.
This is a research-stage risk-monitoring tool. It is explicitly not a diagnostic instrument and makes no clinical claims. It does not detect, diagnose, or treat depression.
capstone · 2023Smart City IoT Services
REST microservices with MQTT-based IoT communication for a smart-city platform, developed as part of the Tsinghua University IEDE international program.