Research map / Adversarial robustness and security / Certified and empirical robustness
Learning to Repair Software Vulnerabilities with Generative Adversarial Networks
Jacob Harer, Onur Ozdemir, Tomo Lazovich, Christopher Reale, Rebecca Russell, Louis Kim, peter chin
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 |
|---|---|---|---|
| add Auto or Freq self-regularization to the unregularized GAN (ablation) | Proposed GAN · Sorting | Seq. Acc.: 82.8 -> 98.9; 82.8 -> 99.3 | Improved |
| add curriculum learning without self-regularization (ablation) | Proposed GAN · Sorting | Seq. Acc.: 82.8 -> 81.5 | Hurt |
| add Auto regularization without curriculum (ablation) | Proposed GAN · CFG | CFG Acc: 98.0 -> 96.5 | Hurt |
| add Freq regularization without curriculum (ablation) | Proposed GAN · CFG | CFG Acc: 98.0 -> 97.5 | Hurt |
| add curriculum learning without self-regularization (ablation) | Proposed GAN · SATE IV paired | BLEU-4: 84.2 -> 88.3 | Improved |
| add Auto regularization without curriculum (ablation) | Proposed GAN · SATE IV paired | BLEU-4: 84.2 -> 85.7 | Improved |
| add Freq regularization without curriculum (ablation) | Proposed GAN · SATE IV paired | BLEU-4: 84.2 -> 86.2 | Improved |
| add curriculum learning without self-regularization (ablation) | Proposed GAN · SATE IV unpaired | BLEU-4: 79.3 -> 81.1 | Improved |
| add Auto regularization without curriculum (ablation) | Proposed GAN · SATE IV unpaired | BLEU-4: 79.3 -> 79.2 | No clear effect |
| use unpaired rather than paired training data | Proposed GAN Base · SATE IV | BLEU-4: 84.2 -> 79.3 | Hurt |
Closest papers by meaning
- Adversarial Robustness for Code (ICML 2020 · similarity 0.85)
- Conserve-Update-Revise to Cure Generalization and Robustness Trade-off in Adversarial Training (ICLR 2024 · similarity 0.82)
- Improved Generation of Adversarial Examples Against Safety-aligned LLMs (NeurIPS 2024 · similarity 0.82)
- Understanding and Enhancing the Transferability of Jailbreaking Attacks (ICLR 2025 · similarity 0.82)
- Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models (NeurIPS 2024 · similarity 0.81)
- Adversarial Mixup Unlearning (ICLR 2025 · similarity 0.81)
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.
