PROJECTS

Mixup implementation

Data augmentation via convex combination.

Hedged instance embedding implementation

Modeled uncertainty by hedging the location of each input in embedding space.

[low uncertainty in clean image]

[high uncertainty in occluded image]

Error encoding network implementation

Visualized that the distance of different latent variables measures the similarity of images.

**Mixture density network implementation **

Visualized explainable variance and unexplainable variance.

AWARDS AND HONORS