Motivation
Many affective-computing models produce useful predictions but remain difficult to interpret. This work focuses on improving both representation learning and model explainability in a sensitive health-related context.
Master's Thesis Research
Temporal Contrastive Learning for Interpretable Facial Emotion Recognition
A research project on interpretable facial video analysis using temporal contrastive learning, sequence modeling, and explanation-oriented outputs.
2024 - 2026
Project overview
This graduate research explores how self-supervised temporal learning and interpretable modeling can support emotion-related facial video analysis in a more understandable way.
Method and architecture
- Temporal contrastive pretraining for facial sequence representation learning
- Sequence modeling with recurrent and attention-based components
- Interpretability work through saliency and attention-oriented analysis
My contribution
- Researched model structure and training direction for sequence-based facial analysis
- Connected contrastive learning ideas with interpretable downstream prediction
- Positioned the work around responsible and understandable health-AI use
Results
- Cross-validation findings are documented in the project data and should be treated as research-stage results rather than broad deployment claims.
Impact
- Strengthened the bridge between deep learning research and practical interpretability needs
- Contributed to a more careful approach to facial-video analysis for health-related applications