Research map / Probabilistic and scientific modeling / Causal discovery
A Causal View on Robustness of 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 |
|---|---|---|---|
| use deep CAMA instead of a discriminative DNN on MNIST FGSM/PGD attacks | deep CAMA · MNIST adversarial examples | accuracy | Helped |
| add test-time fine-tuning on unlabeled attacked data (ablation) | deep CAMA · MNIST FGSM/PGD | accuracy: 20%-40% accuracy increase | Helped |
| add test-time fine-tuning (ablation) | deep CAMA · CIFAR-binary FGSM/PGD | accuracy: 5% to 8% | Mixed |
| add test-time fine-tuning on shifted test data (ablation) | deep CAMA trained on clean MNIST · MNIST vertical shift | accuracy | Helped |
| augment training data with one shift type | discriminative DNN · MNIST unseen shifts | accuracy | Hurt |
| augment training data with one shift type | deep CAMA · MNIST unseen shifts | accuracy | No clear effect |
| fine-tune with only 1% of manipulated test data (ablation) | deep CAMA · MNIST vertical shift | accuracy | Helped |
| use causally consistent graph (children/co-parents correct) vs mis-specified (ablation) | deep CAMA · synthetic measurement data | accuracy | Helped |
| enlarge discriminative network width (ablation) | discriminative DNN · MNIST unseen shifts | accuracy | No clear effect |
Closest papers by meaning
- Robustly Disentangled Causal Mechanisms: Validating Deep Representations for Interventional Robustness (ICML 2019 · similarity 0.88)
- A Statistical Approach to Assessing Neural Network Robustness (ICLR 2019 · similarity 0.87)
- Interpreting Robustness Proofs of Deep Neural Networks (ICLR 2024 · similarity 0.85)
- Causal Adversarial Perturbations for Individual Fairness and Robustness in Heterogeneous Data Spaces (AAAI 2024 · similarity 0.84)
- Robustness Reprogramming for Representation Learning (ICLR 2025 · similarity 0.84)
- Since Faithfulness Fails: The Performance Limits of Neural Causal Discovery (ICML 2025 · similarity 0.84)
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.