Policy Learning Experiments

In our AAAI 2020 paper, we ran FastLAS on two access control policy learning datasets. The first dataset was the Amazon dataset, from:

Amazon. 2013. Amazon.com employee access challenge. http://www.kaggle.com/c/amazon-employee-access-challenge.

The second was the Project Management dataset, from:

Xu, Z., and Stoller, S. D. 2014. Mining attribute-based access control policies from logs. In IFIP Annual Conference on Data and Applications Security and Privacy, 276–291. Springer.

We transformed both datasets into sets of FastLAS tasks for cross vaildation. These FastLAS tasks are available here.

You should create the directory results in the policy_learning directory before running the experiments below. Note that if the experiments are interrupted, they will resume from the last completed experiment. To restart the experiments completely, you should empty the results folder.

Running the Full Amazon/Project Management Experiments

In the paper, we reported the results of various experiments, using both datasets to learn either accept or reject rules with three different scoring functions. You can reproduce these experiments by running:

./full_cross_validation.rb *dataset* *experiment_type* *scoring_function*

Where *dataset* is replaced with either amazon or PM, *experiment_type* is replaced with either accept or reject and *scoring_function* is replaced with either len, cov or uni.

The script will print a summary of the results (iteratively updating the averages).

Running the Amazon Experiments for a Single Resource

In the paper, we did a further set of experiments using only a single resource from the Amazon dataset and evaluated with the universal F1 metric from:

Cotrini, C.; Weghorn, T.; and Basin, D. 2018. Mining abac rules from sparse logs. In 2018 IEEE European Symposium on Security and Privacy (EuroS&P), 31–46. IEEE.

To reproduce these experiments, run:

./resource_cross_validation.rb *resource_id* *scoring_function*

Where *resource_id* is replaced with 25993, 4675 75078 or 79092 and *scoring_function* is replaced with len, cov uni or uf1.

The script will print a summary of the results (iteratively updating the averages).