Research map / Label-efficient and robust learning / Recognition and long-tail
Granger-Causal Attentive Mixtures of Experts: Learning Important Features with 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 the Granger-causal objective to attentive expert gating (ablation) | AME · Cancer gene associations | Recall@10: 2 -> 10 | Helped |
| Add the Granger-causal objective to attentive expert gating (ablation) | AME · Cancer classification test set | Error rate: 4.86±0.093% -> 2.16±0.009% | Helped |
| Add the Granger-causal objective to attentive expert gating (ablation) | AME · Prescription demand test set | SMAPE: 33.08 -> 33.85 | Hurt |
| Use attentive gating without the Granger-causal objective (ablation) | AME · Cancer gene associations | Recall@10: 10 -> 2 | Hurt |
| Increase α from zero to 0.04 (ablation) | AME on prescription demand · Prescription demand test set | SMAPE: 33.08 -> 33.85 | Hurt |
| Increase α from 0 to approximately 0.03 (ablation) | AME on prescription demand · Prescription demand test set | MGE | Helped |
| Increase α from 0 to approximately 0.03 (ablation) | AME on prescription demand · Prescription demand test set | MSE | Hurt |
| Train AME with the Granger-causal objective rather than attention alone (ablation) | AME on prescription demand · Prescription demand feature attribution | Pearson correlation r² with desired attributions: 0.19 -> 0.84 | Helped |
| Use AME feature attribution rather than DeepLIFT | Prescription demand feature attribution · Prescription demand | Attribution computation time: 24 -> 2 CPU seconds | Helped |
| Use AME feature attribution rather than LIME | Prescription demand feature attribution · Prescription demand | Attribution computation time: 10464 -> 2 CPU seconds | Helped |
| Use AME feature attribution rather than SHAP | Prescription demand feature attribution · Prescription demand | Attribution computation time: 729068 -> 2 CPU seconds | Helped |
| Use AME(α=0.05) rather than RF importance | Cancer gene association analysis · Cancer gene associations | Recall@10: 10 vs 10 | No clear effect |
Closest papers by meaning
- Variable Importance in High-Dimensional Settings Requires Grouping (AAAI 2024 · similarity 0.85)
- Fine-Grained Neural Network Explanation by Identifying Input Features with Predictive Information (NeurIPS 2021 · similarity 0.85)
- Testing Semantic Importance via Betting (NeurIPS 2024 · similarity 0.83)
- MoDE: A Mixture-of-Experts Model with Mutual Distillation among the Experts (AAAI 2024 · similarity 0.82)
- Regional Explanations: Bridging Local and Global Variable Importance (NeurIPS 2025 · similarity 0.82)
- Optimizing importance weighting in the presence of sub-population shifts (ICLR 2025 · similarity 0.82)
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.