Research map / Adversarial robustness and security / Backdoor and poisoning
Excess Capacity and Backdoor Poisoning
Naren Sarayu Manoj, Avrim Blum
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 PGD adversarial training instead of vanilla training | CNN · MNIST target 0, alpha=0.05 | training robust loss: 1.00 -> 0.13 | Improved |
| use PGD adversarial training instead of vanilla training | CNN · MNIST target 0, alpha=0.05 | backdoor success rate: 1.00 -> 0.00 | Improved |
| use PGD adversarial training instead of vanilla training | CNN · MNIST target 0, alpha=0.00 | testing robust loss: 1.00 -> 0.09 | Improved |
| use PGD adversarial training instead of vanilla training | CNN · MNIST target 0, alpha=0.05 | testing 0-1 loss: 0.01 -> 0.03 | Hurt |
| increase poisoned fraction alpha from 0.00 to 0.30 under adversarial training | CNN with PGD adversarial training · MNIST target 0 | training robust loss: 0.09 -> 0.41 | Hurt |
| increase poisoned fraction alpha from 0.00 to 0.30 under adversarial training | CNN with PGD adversarial training · MNIST target 0 | testing robust loss: 0.09 -> 0.19 | Hurt |
| increase poisoned fraction alpha from 0.00 to 0.30 under adversarial training | CNN with PGD adversarial training · MNIST target 0 | backdoor success rate: 0.00 -> 0.05 | No clear effect |
| increase alpha with vanilla training | CNN with vanilla training · MNIST target 0 | training 0-1 loss: 0.01 -> 0.01 | No clear effect |
| vary target label t from 0 to 9 | CNN with PGD adversarial training · MNIST | robust loss and backdoor success | No clear effect |
Closest papers by meaning
- Adversarial-Inspired Backdoor Defense via Bridging Backdoor and Adversarial Attacks (AAAI 2025 · similarity 0.91)
- Demystifying Poisoning Backdoor Attacks from a Statistical Perspective (ICLR 2024 · similarity 0.91)
- Generalization Bound and New Algorithm for Clean-Label Backdoor Attack (ICML 2024 · similarity 0.90)
- Universal Backdoor Attacks (ICLR 2024 · similarity 0.90)
- Wicked Oddities: Selectively Poisoning for Effective Clean-Label Backdoor Attacks (ICLR 2025 · similarity 0.90)
- A Theoretical Analysis of Backdoor Poisoning Attacks in Convolutional Neural Networks (ICML 2024 · similarity 0.89)
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.
