Research map / Adversarial robustness and security / Certified and empirical robustness
On the Decision Boundaries of Neural Networks: A Tropical Geometry Perspective
Motasem Alfarra, Adel Bibi, Hasan Abed Al Kader Hammoud, Mohamed Gaafar, Bernard Ghanem
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 |
|---|---|---|---|
| prune with tropical decision-boundary objective instead of Class Blind/Uniform/Distribution | AlexNet classifier · SVHN | pruning-accuracy AUC: better in AUC by 3% | Improved |
| prune with tropical objective | AlexNet classifier · CIFAR10 | pruning-accuracy AUC: better in AUC by 3% | Improved |
| prune with tropical objective | AlexNet classifier · CIFAR100 | pruning-accuracy AUC: better in AUC by 2% | Improved |
| prune with tropical objective | VGG16 classifier · CIFAR100 | pruning-accuracy AUC | Hurt |
| prune with tropical objective | VGG16 classifier · SVHN and CIFAR10 | pruning-accuracy AUC | No clear effect |
| tropical pruning instead of SM heuristic | LeNet · MNIST, 90% neurons kept | test accuracy: 95.71 -> 98.82 | Improved |
| tropical pruning instead of SM heuristic | LeNet · MNIST, 25% neurons kept | test accuracy: 91.04 -> 98.36 | Improved |
| tropical pruning instead of SM heuristic | LeNet · MNIST, 5% neurons kept | test accuracy: 92.93 -> 94.91 | Improved |
| add decision-boundary polytope perturbation (eps2=1.0) to structured attack (ablation) | structured adversarial attack · MNIST, eps1=0.1 | robust accuracy: 44.8 -> 46.1 | Hurt |
| add decision-boundary polytope perturbation (eps2=2.0) to structured attack (ablation) | structured adversarial attack · MNIST, eps1=0.1 | robust accuracy: 44.8 -> 43.8 | Improved |
| add decision-boundary polytope perturbation (eps2=2.0) to structured attack (ablation) | structured adversarial attack · MNIST, eps1=0.2 | robust accuracy: 12.6 -> 11.0 | Improved |
| initialize pruned network with the original (lottery ticket) initialization instead of Xavier/Gaussian (ablation) | pruned 2-layer network · synthetic 2D binary data | orientation of decision boundaries polytope | Improved |
Closest papers by meaning
- Empirical Study of the Topology and Geometry of Deep Networks (CVPR 2018 · similarity 0.89)
- What do Graph Neural Networks learn? Insights from Tropical Geometry (NeurIPS 2024 · similarity 0.88)
- Tropical Attention: Neural Algorithmic Reasoning for Combinatorial Algorithms (NeurIPS 2025 · similarity 0.86)
- No Wrong Turns: The Simple Geometry Of Neural Networks Optimization Paths (ICML 2024 · similarity 0.86)
- Mind the Boundary: Coreset Selection via Reconstructing the Decision Boundary (ICML 2024 · similarity 0.85)
- Geometric Inductive Biases of Deep Networks: The Role of Data and Architecture (ICLR 2025 · similarity 0.85)
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.
