VeriX: Towards Verified Explainability of Deep Neural Networks
About
We present VeriX (Verified eXplainability), a system for producing optimal robust explanations and generating counterfactuals along decision boundaries of machine learning models. We build such explanations and counterfactuals iteratively using constraint solving techniques and a heuristic based on feature-level sensitivity ranking. We evaluate our method on image recognition benchmarks and a real-world scenario of autonomous aircraft taxiing.
Min Wu, Haoze Wu, Clark Barrett• 2022
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Standard robust explanation computation | MNIST first 100 images (test) | Explanation Size (|Ex|)82.74 | 6 | |
| Standard robust explanation computation | GTSRB first 100 images (test) | Explanation Complexity (|Ex|)287 | 6 | |
| OVAL-optimal robust explanation generation | MNIST (first 10 images) (test) | Magnitude Cx129.1 | 3 | |
| OVAL-optimal robust explanation generation | CIFAR-10 first 10 images (test) | Complexity |Cx|209.3 | 3 | |
| Robust Explanation Computation | CIFAR-10 first 100 images (test) | Explanation Complexity (|Ex|)204.2 | 3 | |
| Sufficient Explanation Generation | Breast cancer | Explanation Size16.58 | 3 | |
| Sufficient Explanation Generation | Credit | Model Size12.42 | 3 | |
| Sufficient Explanation Generation | FICO HELOC | Explanation Size15.6 | 3 |
Showing 8 of 8 rows