On Relating Explanations and Adversarial Examples
Authors: Alexey Ignatiev, Nina Narodytska, Joao Marques-Silva
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental evidence demonstrating the relationship between explanations and adversarial examples is analyzed in Section 4. The section practically illustrates the described duality between the concepts of absolute explanation and counterexample for a given model prediction. To do this, the following experiment was performed on a Macbook Pro with an Intel Core i5 2.3GHz CPU and 16GB of memory. The experiment targets the well-known and widely used MNIST digits database. |
| Researcher Affiliation | Collaboration | Alexey Ignatiev Monash University, Australia alexey.ignatiev@monash.edu Nina Narodytska VMWare Research, CA, USA nnarodytska@vmware.com Joao Marques-Silva ANITI, Toulouse, France joao.marques-silva@univ-toulouse.fr |
| Pseudocode | Yes | Algorithm 1 shows a Pythonic-style algorithm to compute the complete set E. |
| Open Source Code | Yes | The developed Python-based prototype5 follows the prime compilation approach of Algorithm 1 and uses CPLEX 12.8.0 [20] as an MILP oracle, which is invoked at each iteration of the algorithm. The implementation of minimum hitting set enumeration of Algorithm 1 is based on an award-winning maximum satisfiability solver RC26 [22] written on top of the Py SAT toolkit [21]. 5https://github.com/alexeyignatiev/xpce-duality/ |
| Open Datasets | Yes | The experiment targets the well-known and widely used MNIST digits database4 as it enables a visual demonstration of the discovered duality relationship. 4http://yann.lecun.com/exdb/mnist/ |
| Dataset Splits | No | The paper mentions using the MNIST dataset and training networks to distinguish digits, but it does not specify explicit training, validation, or test splits by percentage or count. It states, "For the sake of simplicity, the networks used are trained to distinguish two digits, e.g. 5 and 6". |
| Hardware Specification | Yes | The following experiment was performed on a Macbook Pro with an Intel Core i5 2.3GHz CPU and 16GB of memory. |
| Software Dependencies | Yes | The developed Python-based prototype... uses CPLEX 12.8.0 [20] as an MILP oracle... The implementation... is based on an award-winning maximum satisfiability solver RC26 [22] written on top of the Py SAT toolkit [21]. |
| Experiment Setup | Yes | For the sake of simplicity, the networks used are trained to distinguish two digits, e.g. 5 and 6... (1) only pixels from a predefined patch area can participate in an explanation/counterexample with the other pixels being fixed; (2) the images were binarized, i.e. every pixel can be either black or white. |