Research map / Adversarial robustness and security / Backdoor and poisoning
Architectural Backdoors in Neural Networks
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 |
|---|---|---|---|
| Add an architectural backdoor with a checkerboard trigger detector | VGG-11 · CIFAR-10, direct use | triggered accuracy: 77.8% -> 10.0% | Helped |
| Add an architectural backdoor with a checkerboard trigger detector | VGG-11 · CIFAR-10, direct use | task accuracy: 81.4% -> 80.2% | No clear effect |
| Use an architectural backdoor rather than a data-poisoning backdoor | VGG-11 · CIFAR-10, direct use | triggered accuracy: 10.1% -> 10.0% | No clear effect |
| Use an architectural backdoor rather than a handcrafted weight backdoor | VGG-11 · CIFAR-10, direct use | triggered accuracy: 19.6% -> 10.0% | Helped |
| Retain the architectural trigger detector through fine-tuning instead of embedding the backdoor in weights | VGG-11 · GTSB to BTSC fine-tuning | triggered accuracy | Helped |
| Retain the architectural backdoor when re-training from scratch rather than relying on a weight-based backdoor | VGG-11 · IMDB-Wiki re-training | triggered accuracy ratio: 8x | Helped |
| Replace a white-patch trigger detector with a detector requiring both white and black pixels in a 3x3 region (ablation) | NaiveMAB · CIFAR-10 | trigger detection robustness | Helped |
| Search for backdoors using the modified DARTS objective (ablation) | DARTS · MNIST | triggered accuracy | No clear effect |
| Use the DARTS-searched architecture after fine-tuning (ablation) | DARTS-searched backdoor · MNIST | backdoor persistence | Hurt |
Closest papers by meaning
- Statistically Undetectable Backdoors in Deep Neural Networks (ICML 2026 · similarity 0.92)
- 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 0.90)
- BAN: Detecting Backdoors Activated by Adversarial Neuron Noise (NeurIPS 2024 · similarity 0.90)
- BackdoorDM: A Comprehensive Benchmark for Backdoor Learning on Diffusion Model (NeurIPS 2025 · similarity 0.89)
- BaDExpert: Extracting Backdoor Functionality for Accurate Backdoor Input Detection (ICLR 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.