1. Uncertanty estimation for deep learning model (eyegaze).
2. Debug Deep Learning model with gradient analysis..
3. Improve consitency in deep learning model (eyestate classification)
4. Full pipeline for refining label/model training/spotting edgecase cycles.
5. Train multitask model.
6. Knowledge Distiallation for Deep Learning. Example with heatmap landmarker model.
7. Simple domain adaptation from RGB to IR camera.