Research map / Adversarial robustness and security / Certified and empirical robustness
Sparse DNNs with Improved Adversarial Robustness
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Increase activation sparsity from 45.25% to 75.92% with L1 activation regularization | LeNet-300-100 · MNIST | r_infinity: 0.2862±0.0113 -> 0.3993±0.0079 | Helped |
| Increase activation sparsity from 45.25% to 75.92% with L1 activation regularization | LeNet-300-100 · MNIST | r_2: 1.3213±0.0207 -> 1.5903±0.0240 | Helped |
| Raise activation sparsity further to 95.22% (ablation) | LeNet-300-100 · MNIST | r_infinity: 0.3993±0.0079 -> 0.2098±0.0133 | Hurt |
| Raise activation sparsity further to 95.22% (ablation) | LeNet-300-100 · MNIST | r_2: 1.5903±0.0240 -> 1.1440±0.0402 | Hurt |
| Increase activation sparsity from 51.26% to 97.54% with L1 activation regularization | LeNet-5 · MNIST | r_infinity: 0.7388±0.0188 -> 0.7729±0.0081 | Helped |
| Increase activation sparsity from 51.26% to 97.54% with L1 activation regularization | LeNet-5 · MNIST | r_2: 2.7831±0.1490 -> 3.1688±0.1203 | Helped |
| Raise activation sparsity further to 99.64% (ablation) | LeNet-5 · MNIST | r_infinity: 0.7729±0.0081 -> 0.6741±0.0162 | Hurt |
| Raise activation sparsity further to 99.64% (ablation) | LeNet-5 · MNIST | r_2: 3.1688±0.1203 -> 2.0799±0.0522 | Hurt |
| Prune LeNet-300-100 to 98.21% weight sparsity | LeNet-300-100 · MNIST | r_infinity: 0.2663 -> 0.3823 | Helped |
| Prune LeNet-300-100 to 98.21% weight sparsity | LeNet-300-100 · MNIST | r_2: 1.3899 -> 1.1058 | Hurt |
| Prune LeNet-5 to 99.07% weight sparsity | LeNet-5 · MNIST | r_infinity: 0.7887 -> 0.6791 | Hurt |
| Prune LeNet-5 to 99.07% weight sparsity | LeNet-5 · MNIST | r_2: 2.7226 -> 1.7383 | Hurt |
Closest papers by meaning
- Feature compression is the root cause of adversarial fragility in neural networks (ICLR 2026 · similarity 0.91)
- Towards Deep Learning Models Resistant to Adversarial Attacks (ICLR 2018 · similarity 0.91)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.90)
- Sparsity Winning Twice: Better Robust Generalization from More Efficient Training (ICLR 2022 · similarity 0.90)
- Enhancing Adversarial Robustness in SNNs with Sparse Gradients (ICML 2024 · similarity 0.88)
- Interpreting Robustness Proofs of Deep Neural Networks (ICLR 2024 · similarity 0.88)
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.