Some Projects

Generative models:

• Diffusion model Generate new MNIST digits by training a model to reverse a gradual noising process.
  • Patches creation for vision transformer Cut each training image into patches for a vision transformer, and make that step much faster.
• Gibbs sampling and decision trees for tabular data Draw new tabular samples with a Gibbs sampler whose conditionals come from decision trees.

Anomaly Detection:

• A note on precision and recall in continuous contexts How precision and recall change when anomalies are stretches of a signal, not single points.

Image Segmentation

• Simple image segmentation with Pytorch Label every pixel of a scene with a small PyTorch segmentation model.