Research map / Probabilistic and scientific modeling / Causal discovery
DISCO: Mitigating Bias in Deep Learning with Conditional Distance Correlation
Emre Kavak, Tom Nuno Wolf, Christian Wachinger
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 |
|---|---|---|---|
| add DISCOm conditional independence penalty | ResNet · dSprites | R2: 0.417 -> 0.688 | Improved |
| add sDISCO penalty | ResNet · Blob | R2: 0.281 -> 0.751 | Improved |
| add DISCOm penalty | ResNet-18 · YaleB | balanced accuracy: 0.607 -> 0.804 | Improved |
| add sDISCO penalty | ResNet-18 · Waterbirds | balanced accuracy: 0.730 -> 0.874 | Improved |
| add sDISCO penalty | TinyBERT · MNLI | worst-group accuracy: 0.616 -> 0.760 | Improved |
| use CIRCE instead of sDISCO | ResNet-18 · Waterbirds | balanced accuracy: 0.874 -> 0.894 (CIRCE better) | Mixed |
| use corrected exact sDISCO factorization instead of earlier buggy implementation (ablation) | sDISCO · dSprites | R2: 0.619 -> 0.687 | Improved |
| use corrected sDISCO instead of earlier implementation (ablation) | sDISCO · YaleB | balanced accuracy: 0.694 -> 0.770 | Improved |
| use corrected sDISCO instead of earlier implementation (ablation) | sDISCO · Blob | R2: 0.788 -> 0.751 | Hurt |
| regularize both bias variables instead of only one (ablation) | sDISCO · dSprites counterfactual | counterfactual accuracy: 0.83 -> 0.98 | Improved |
| add sDISCO on both biases | ResNet · dSprites counterfactual | sensitivity S_X: 0.302 -> 0.066 | Improved |
| add adversarial debiasing instead of DISCO | ResNet · dSprites | R2: 0.688 -> 0.467 (DISCOm vs Adversarial) | Hurt |
Closest papers by meaning
- Causal Fairness under Unobserved Confounding: A Neural Sensitivity Framework (ICLR 2024 · similarity 0.87)
- From Hope to Safety: Unlearning Biases of Deep Models via Gradient Penalization in Latent Space (AAAI 2024 · similarity 0.86)
- FADES: Fair Disentanglement with Sensitive Relevance (CVPR 2024 · similarity 0.86)
- Causal Representation Learning via Counterfactual Intervention (AAAI 2024 · similarity 0.85)
- Since Faithfulness Fails: The Performance Limits of Neural Causal Discovery (ICML 2025 · similarity 0.85)
- Modular Learning of Deep Causal Generative Models for High-dimensional Causal Inference (ICML 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.
