https://github.com/vita-group/lth-pass

[TMLR] "Can You Win Everything with Lottery Ticket?" by Tianlong Chen, Zhenyu Zhang, Jun Wu, Randy Huang, Sijia Liu, Shiyu Chang, Zhangyang Wang
https://github.com/vita-group/lth-pass

adversarial-robustness explanability flatness generalization interpretability loss-landscape lottery-ticket-hypothesis out-of-distribution-detection pac-bayes robustness uncertainty winning-tickets

Last synced: about 1 month ago
JSON representation

[TMLR] "Can You Win Everything with Lottery Ticket?" by Tianlong Chen, Zhenyu Zhang, Jun Wu, Randy Huang, Sijia Liu, Shiyu Chang, Zhangyang Wang

• Host: GitHub
• URL: https://github.com/vita-group/lth-pass
• Owner: VITA-Group
• License: mit
• Created: 2021-12-06T06:36:41.000Z (about 3 years ago)
• Default Branch: main
• Last Pushed: 2022-09-08T00:37:07.000Z (over 2 years ago)
• Last Synced: 2024-04-16T07:18:14.572Z (8 months ago)
• Topics: adversarial-robustness, explanability, flatness, generalization, interpretability, loss-landscape, lottery-ticket-hypothesis, out-of-distribution-detection, pac-bayes, robustness, uncertainty, winning-tickets
• Language: Python
• Homepage:
• Size: 8.54 MB
• Stars: 10
• Watchers: 7
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Can You Win Everything with Lottery Ticket?

[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](https://opensource.org/licenses/MIT)

Code for this **preprint** paper [Can You Win Everything with Lottery Ticket?]()

Tianlong Chen, Zhenyu Zhang, Jun Wu, Randy Huang, Sijia Liu, Shiyu Chang, Zhangyang Wang

## Overview

## Experiment Results

![](./Figs/Res.png)

## Prerequisites

- pytorch

- torchvision

- advertorch

## Usages

Evaluation on CIFAR-10/100 

```

python -u main_eval.py \

	--data [data-direction] \

	--dataset cifar10 \ # choose from [cifar10, cifar100]

	--arch resnet20s \ # choose from [resnet20s,resnet18,wideresnet]

	--pretrained [pretrained-weight] \ 

	--eval_mode accuracy,robustness,corruption,ood,calibration,interpretation,pac_bayes_weight,pac_bayes_input \

	--output_file result.pt \

	--test_randinit_off \

	--image_number 1000 

```

Evaluation on ImageNet 

```

python -u main_eval_imagenet.py \

    --data [data-direction] \

    --arch resnet50 \

    --pretrained [pretrained-weight] \

    --eval_mode accuracy,robustness,corruption,ood,calibration,interpretation,pac_bayes_weight,pac_bayes_input \

    --output_file result.pt 

```

Evaluation for Hessian

```

python -u vis_pyhessian_analysis.py \

    --data [data-direction] \

    --dataset [dataset] \ choose from [cifar10, cifar100, imagenet]

    --arch [network-architecture] \ choose from [resnet20s,resnet18,wideresnet,resnet50]

    --pretrained [pretrained-weight] \

    --output_file result.pt \

    --mode weight,input

```

Composition neurons

`cd composition-neuron`, which is modified from https://github.com/jayelm/compexp

## Citation

```

TBD

```