Research map / Probabilistic and scientific modeling / Causal discovery
Generative Interventions for Causal Learning
Check 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 |
|---|---|---|---|
| Steer the generator to intervene on nuisance factors rather than only sampling GAN images | ResNet50 trained on BigGAN data · ImageNet validation | top1 accuracy: 42.65% -> 45.06% | Helped |
| Add generated interventional data X_int to ImageNet training (ablation) | ResNet18 · ObjectNet test, Std. Augmentation | top1 accuracy: 20.48% -> 22.07% | Helped |
| Use transferred interventions X_itr with ImageNet training instead of only X_int (ablation) | ResNet18 · ObjectNet test, Std. Augmentation | top1 accuracy: 22.07% -> 22.29% | Helped |
| Add both generated X_int and transferred X_itr interventions (ablation) | ResNet18 · ObjectNet test, Std. Augmentation | top1 accuracy: 20.48% -> 22.34% | Helped |
| Add generated interventional data X_int to ImageNet training (ablation) | ResNet18 · ObjectNet test, Add. Augmentation | top1 accuracy: 24.42% -> 25.71% | Helped |
| Add transferred interventions X_itr to ImageNet training (ablation) | ResNet18 · ObjectNet test, Add. Augmentation | top1 accuracy: 24.42% -> 27.02% | Helped |
| Combine generated X_int and transferred X_itr interventions (ablation) | ResNet18 · ObjectNet test, Add. Augmentation | top1 accuracy: 24.42% -> 27.03% | Helped |
| Increase intervention strength while selecting for higher log P(x|z) (ablation) | GenInt · ImageNet and ObjectNet | accuracy | Helped |
| Use generative interventions instead of observational GAN sampling (ablation) | Classifier trained on GAN-generated data · ImageNet | top1 accuracy: 42.65% -> 45.06% | Helped |
| Intervene on latent principal directions to randomize background color (ablation) | CVAE classifier · Colored MNIST causal test | test accuracy: 11.255% -> 29.618% | Helped |
| Use Stylized ImageNet rather than ImageNet Only | ResNet152 · ImageNet validation | top1 accuracy: 78.57% -> 77.27% | Hurt |
Closest papers by meaning
- Toward Causal Generative Modeling: From Representation to Generation (AAAI 2025 · similarity 0.88)
- Object centric architectures enable efficient causal representation learning (ICLR 2024 · similarity 0.88)
- Visual Representation Learning through Causal Intervention for Controllable Image Editing (CVPR 2025 · similarity 0.87)
- CaRiNG: Learning Temporal Causal Representation under Non-Invertible Generation Process (ICML 2024 · similarity 0.86)
- Restyling Unsupervised Concept Based Interpretable Networks with Generative Models (ICLR 2025 · similarity 0.86)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.86)
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.