RaVeN

RaVeN : Relational Verification of Neural Networks

Relational erifier runs through the unittest framework. A new unit test can be added to run the verifier with a specific configuration. Current unit tests are located in src/tests/test_raven.py. Conda Environment at environment.yml, install with conda env create -f environment.yml.

Reproducing Experiments

Move to the RaVeN directory

cd RaVeN

Then run the following any of the following commands

Untargeted UAP Verification

Run any experiment by replacing test_name with any test from class TestDifferenceAblation from the file RaVeN/src/tests/test_raven.py

Small networks

python -m unittest -v src.tests.test_raven.TestUntargetedUapSmall.test_name

Large networks

python -m unittest -v src.tests.test_raven.TestUntargetedUapLarge.test_name

Targeted UAP Verification

Run any experiment by replacing test_name with any test from class TestTargetedUap from the file RaVeN/src/tests/test_raven.py

python -m unittest -v src.tests.test_raven.TestTargetedUap.test_name

Monotonicity Verification

Run any experiment by replacing test_name with any test from class TestMonotonicity from the file RaVeN/src/tests/test_raven.py

python -m unittest -v src.tests.test_raven.TestMonotonicity.test_name

Worst Worst-case hamming distance Verification

Run any experiment by replacing test_name with any test from class TestWorstCaseHamming from the file ``RaVeN/src/tests/test_raven.py`

python -m unittest -v src.tests.test_raven.TestWorstCaseHamming.test_name

Ablation w.r.t Difference Constraints

Run any experiment by replacing test_name with any test from class TestDifferenceAblation from the file RaVeN/src/tests/test_raven.py

python -m unittest -v src.tests.test_raven.TestDifferenceAblation.test_name

Designing Custom Tests

1. Update the parameters in the test_custom from TestRavenCustom in the file RaVeN/src/tests/test_raven.py. The class RaVeNArgs from RaVeN/src/raven_args.py defines the parameters

2. After updating parameters Move to the RaVeN directory

cd RaVeN

3. Run the following command from

python -m unittest -v src.tests.test_raven.TestRavenCustom.test_custom