Skip to content

JothiRamaswamy/Unlearn-Metrics

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

EVALUATING VARIOUS METRICS FOR MEASURING THE EFFICACY OF MACHINE UNLEARNING ALGORITHMS

Files changed/added in forked repository

Added Folders

results: contains resulting metrics of each of the models and in each unique seed (used for selecting forget set that model leverages for unlearning). More details in README.md file inside folder.

final_metrics: contains final metrics of each of the models, averaged across all seeds (used for selecting forget set that model leverages for unlearning). More details in README.md file inside folder

Added Files

load_model.py: contains code for saving (after untarring) + loading the model after unlearning (in the FASRC cluster), obtaining model predictions + prediction logits, obtaining the forget set used for unlearning

metrics.py: code for generating metric results given model weights + predictions (and logits). Metrics created in this file: wasserstein distance on forget/test set predictions in output space, KL divergence of model weights in weight space, L_2 distance of model weights in weight space.

metric_analysis.py: contains primary code for generating plots of metrics obtained from models after unlearning.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%