Extreme sparsity by Input Convex Neural Networks.
- FC + MNIST
- U-NET + LungCT
- ResNET + Cifar10
- ConvAE + CelebA(reduced to 40k images)
- AlexNET + CATS/DOGS
- https://youtu.be/QrcHkKbUOPg - video presentation
- https://www.overleaf.com/read/hcvrgydykdtw – report source
- https://www.overleaf.com/read/snxwfrszvtys – presentation source
- https://skoltech.instructure.com/courses/2361/discussion_topics/10733 – general guidlines
- https://skoltech.instructure.com/courses/2361/files/158268/download?wrap=1 – report guidlines
- https://github.com/locuslab/icnn - repo with TF implementation of input-convex-nns
- https://arxiv.org/pdf/1609.07152.pdf - paper by Amos, describing a general approach
- https://arxiv.org/pdf/1909.13082v2.pdf - paper by Korotin, describing a couple of architectures in appendix
- https://docs.google.com/spreadsheets/d/1zNm02EqpN855odlE-SOPIRvJwtfs1ZzQt_Q59nES4ho/edit#gid=0 - spreadsheet with all projects, our goes by the number 29
- https://github.com/bhpfelix/Variational-Autoencoder-PyTorch – potential code for VAE (not used in final solution)
- https://drive.google.com/drive/folders/0B7EVK8r0v71pTUZsaXdaSnZBZzg - img_align_celeba.zip