Research map / Adversarial robustness and security / Certified and empirical robustness
Neural Networks with Recurrent Generative Feedback
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 recurrent generative feedback (CNN-F avg) to adversarially trained CNN | CNN · Fashion-MNIST | PGD (first) accuracy: 77.09 -> 79.55 | Helped |
| add recurrent generative feedback (CNN-F avg) to adversarially trained CNN | CNN · CIFAR-10 | PGD (first) accuracy: 42.31 -> 48.72 | Helped |
| add generative feedback | CNN · Fashion-MNIST | clean accuracy: 89.97 -> 89.77 | No clear effect |
| add generative feedback (avg) | CNN · CIFAR-10 | clean accuracy: 79.09 -> 80.27 | No clear effect |
| average logits over iterations instead of using last iteration (ablation) | CNN-F · CIFAR-10 | PGD (e2e) accuracy: 49.35 -> 55.02 | Helped |
| average logits over iterations instead of last (ablation) | CNN-F · Fashion-MNIST | PGD (e2e) accuracy: 78.34 -> 79.89 | Helped |
| reconstruct to earlier layer (ind=3 vs ind=5), avg logits (ablation) | CNN-F · CIFAR-10 | PGD (e2e) accuracy: 55.02 -> 67.44 | Helped |
| reconstruct to earlier layer (ind=3 vs ind=5), avg logits (ablation) | CNN-F · CIFAR-10 | PGD (first) accuracy: 48.72 -> 45.61 | Hurt |
| train with more feedback iterations (standard training) (ablation) | CNN-F · Fashion-MNIST | PGD-40 accuracy | Helped |
| larger cross-entropy coefficient on clean images (ablation) | CNN-F · CIFAR-10 | PGD (e2e) vs first accuracy | Mixed |
Closest papers by meaning
- Feedback Adversarial Learning: Spatial Feedback for Improving Generative Adversarial Networks (CVPR 2019 · similarity 0.88)
- Task-Driven Convolutional Recurrent Models of the Visual System (NeurIPS 2018 · similarity 0.88)
- Recurrent neural network dynamical systems for biological vision (NeurIPS 2024 · similarity 0.87)
- Forward Learning with Top-Down Feedback: Empirical and Analytical Characterization (ICLR 2024 · similarity 0.85)
- Recall-Oriented Continual Learning with Generative Adversarial Meta-Model (AAAI 2024 · similarity 0.85)
- RTify: Aligning Deep Neural Networks with Human Behavioral Decisions (NeurIPS 2024 · 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.