Research map / Adversarial robustness and security / Certified and empirical robustness
FAME: $\underline{F}$ormal $\underline{A}$bstract $\underline{M}$inimal $\underline{E}$xplanation for neural networks
Ryma Boumazouza, Raya Elsaleh, Melanie Ducoffe, Shahaf Bassan, Guy Katz
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use greedy instead of MILP for abstract batch freeing (single round) | FAME · MNIST/GTSRB models | runtime: 9x to 12x speedup | Improved |
| use greedy instead of MILP for abstract batch freeing (iterative refinement) | FAME · GTSRB-CNN | runtime: up to 2.4x speedup | Improved |
| add iterative refinement of the perturbation domain to single-round FAME (ablation) | FAME · MNIST-CNN | explanation size: 190.29 -> 122.09 | Improved |
| add iterative refinement of the perturbation domain to single-round FAME (ablation) | FAME · GTSRB-CNN | explanation size: 379.34 -> 322.42 | Improved |
| use FAME-accelerated VERIX+ instead of VERIX+ alone | VERIX+ · GTSRB-CNN | runtime (s): 185.03 -> 138.12 | Improved |
| use FAME-accelerated VERIX+ instead of VERIX+ alone | VERIX+ · MNIST-FC | explanation size: 280.16 -> 224.41 | Improved |
| use FAME iterative greedy non-minimal set instead of VERIX+ minimal set | VERIX+ · GTSRB-CNN | runtime: 338.28 features in 185.03s -> 322.42 features in 7.4s | Improved |
| add cardinality-constrained perturbation domain (ablation) | FAME · MNIST/GTSRB models | explanation size | Improved |
| use FAME single-round on GTSRB-FC instead of VERIX+ | VERIX+ · GTSRB-FC | explanation size: 313.42 -> 243.18 | Improved |
Closest papers by meaning
- Explaining, Fast and Slow: Abstraction and Refinement of Provable Explanations (ICML 2025 · similarity 0.86)
- On Completeness-aware Concept-Based Explanations in Deep Neural Networks (NeurIPS 2020 · similarity 0.84)
- Beyond Topological Self-Explainable GNNs: A Formal Explainability Perspective (ICML 2025 · similarity 0.84)
- Provably Explaining Neural Additive Models (ICLR 2026 · similarity 0.83)
- CoSy: Evaluating Textual Explanations of Neurons (NeurIPS 2024 · similarity 0.83)
- GNNX-BENCH: Unravelling the Utility of Perturbation-based GNN Explainers through In-depth Benchmarking (ICLR 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
