├── .gitignore ├── Dockerfile ├── README.md ├── deep_nets ├── data.py ├── eval_loss_surface.py ├── eval_sharpness.py ├── loss_surfaces.ipynb ├── losses.py ├── models.py ├── sam.py ├── train.py ├── utils.py ├── utils_eval.py └── utils_train.py ├── diagonal_linear_nets ├── diag_nets.py └── diag_nets_sam.ipynb ├── images ├── sam_diag_nets.png ├── sam_fig1.png ├── sam_finetuning.png └── sam_noisy_labels.png └── one_layer_relu_nets └── 1d_nn.ipynb /.gitignore: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/.gitignore -------------------------------------------------------------------------------- /Dockerfile: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/Dockerfile -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/README.md -------------------------------------------------------------------------------- /deep_nets/data.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/deep_nets/data.py -------------------------------------------------------------------------------- /deep_nets/eval_loss_surface.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/deep_nets/eval_loss_surface.py -------------------------------------------------------------------------------- /deep_nets/eval_sharpness.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/deep_nets/eval_sharpness.py -------------------------------------------------------------------------------- /deep_nets/loss_surfaces.ipynb: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/deep_nets/loss_surfaces.ipynb -------------------------------------------------------------------------------- /deep_nets/losses.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/deep_nets/losses.py -------------------------------------------------------------------------------- /deep_nets/models.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/deep_nets/models.py -------------------------------------------------------------------------------- /deep_nets/sam.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/deep_nets/sam.py -------------------------------------------------------------------------------- /deep_nets/train.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/deep_nets/train.py -------------------------------------------------------------------------------- /deep_nets/utils.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/deep_nets/utils.py -------------------------------------------------------------------------------- /deep_nets/utils_eval.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/deep_nets/utils_eval.py -------------------------------------------------------------------------------- /deep_nets/utils_train.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/deep_nets/utils_train.py -------------------------------------------------------------------------------- /diagonal_linear_nets/diag_nets.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/diagonal_linear_nets/diag_nets.py -------------------------------------------------------------------------------- /diagonal_linear_nets/diag_nets_sam.ipynb: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/diagonal_linear_nets/diag_nets_sam.ipynb -------------------------------------------------------------------------------- /images/sam_diag_nets.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/images/sam_diag_nets.png -------------------------------------------------------------------------------- /images/sam_fig1.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/images/sam_fig1.png -------------------------------------------------------------------------------- /images/sam_finetuning.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/images/sam_finetuning.png -------------------------------------------------------------------------------- /images/sam_noisy_labels.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/images/sam_noisy_labels.png -------------------------------------------------------------------------------- /one_layer_relu_nets/1d_nn.ipynb: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/tml-epfl/understanding-sam/HEAD/one_layer_relu_nets/1d_nn.ipynb --------------------------------------------------------------------------------