Relational DNN Verification With Cross Executional Bound Refinement
This work addresses the challenge of precise relational verification for DNNs, which is crucial for ensuring safety and reliability in applications like adversarial robustness, but it is incremental as it builds on prior methods by extending dependency analysis.
The paper tackles the problem of verifying relational properties in deep neural networks, such as robustness against universal adversarial perturbations, by developing a scalable verifier called RACoon that leverages cross-execution dependencies across all layers, achieving substantial precision improvements over state-of-the-art baselines on various datasets, networks, and properties.
We focus on verifying relational properties defined over deep neural networks (DNNs) such as robustness against universal adversarial perturbations (UAP), certified worst-case hamming distance for binary string classifications, etc. Precise verification of these properties requires reasoning about multiple executions of the same DNN. However, most of the existing works in DNN verification only handle properties defined over single executions and as a result, are imprecise for relational properties. Though few recent works for relational DNN verification, capture linear dependencies between the inputs of multiple executions, they do not leverage dependencies between the outputs of hidden layers producing imprecise results. We develop a scalable relational verifier RACoon that utilizes cross-execution dependencies at all layers of the DNN gaining substantial precision over SOTA baselines on a wide range of datasets, networks, and relational properties.