ConstraintFlow

DNN Certifier Specifications

The DNN certifier specifications can be found in:

• verification/data/test_cases_correct

Set up environment

sudo apt update
sudo apt install python
pip install -r requirements.txt

To re-generate the lexer and parser

sudo apt install antlr4
cd ast_cflow
antlr4 -Dlanguage=Python3 -visitor dsl.g4

ConstraintFlow Verification

ConstraintFlow verification procedure can be used to automatically check the soundness of the DNN certifiers.

Run ConstraintFlow Verification procedure for checking the soundness of a DNN certifier

python -m verification.experiments.experiments_correct certifier_file [n_prev] [n_sym]

Example Usage:

python -m verification.experiments.experiments_correct verification/data/test_cases_correct/deeppoly_relu 1 1

To randomly introduce bugs to a DNN certifier and check for soundness

python -m verification.experiments.experiments_incorrect certifier_file [n_prev] [n_sym]

Example:

python -m verification.experiments.experiments_incorrect verification/data/test_cases_correct/deeppoly_relu 1 1

To run all experiments presented in the paper

python -m verification.experiments.experiments_full 

ConstraintFlow Compiler

ConstraintFlow Compiler generates a PyTorch based executable for a DNN certifier.

Compile a DNN certifier in ConstraintFlow

python -m compile certifier_file 

Example Usage:

python -m compile certifier/testcases/deeppoly

Run the compiled certifier

python -m run_certifier onnx_network_file batch_size 

Example Usage:

python -m run_certifier certifier/nets/mnist_relu_3_50.onnx 1

📜 Citation

@InProceedings{constraintflow,
    author={Avaljot Singh and Yasmin Sarita and Charith Mendis and Gagandeep Singh},
    title={ConstraintFlow: A DSL for Specification and Verification of Neural Network Analyses},
    booktitle="Static Analysis",
    year="2024",
    publisher="Springer Nature Switzerland",
}