Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Additive Causal Bandits with Unknown Graph
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 global interventions and MODL rather than identifying the parents before optimizing | Causal bandits with unknown graph · Randomly generated additive causal models | sample complexity | Helped |
| Use global interventions and MODL rather than identifying the parents before optimizing | Causal bandits with unknown graph · Randomly generated additive causal models | final-action gap | Helped |
| Use full-product-space Successive Elimination instead of exploiting additive structure | Best-arm identification for causal interventions · Randomly generated additive causal models with few variables | sample complexity: roughly 200 times | Hurt |
| Provide the number of parents so MODL can terminate once that many variables are identified | MODL · Randomly generated additive causal models | sample complexity | Helped |
| Add marginal action-set restriction to action elimination (ablation) | MODL · Additive combinatorial linear bandit with full-bandit feedback | sample complexity | Mixed |
| Add nonlinear interaction terms to the additive outcome model (ablation) | MODL · Randomly generated causal models with nonlinear interaction terms | final-action gap | Hurt |
| Add nonlinear interaction terms to the additive outcome model (ablation) | Parents-first (P1) · Randomly generated causal models with nonlinear interaction terms | final-action gap | Hurt |
| Use MODL without prior parent identification rather than P1 | Causal bandits with unknown graph · Randomly generated causal models with nonlinear interaction terms | final-action gap | Helped |
| Increase the number of parents | MODL · Randomly generated additive causal models | sample complexity | Helped |
| Increase the number of parents | MODL · Randomly generated additive causal models | sample complexity | Mixed |
| Change the degree of the sampled causal graph while keeping the parent count fixed (ablation) | MODL · Randomly generated causal models with K=30 and 10 parents | sample complexity | No clear effect |
| Set the parent count to K so MODL and the oracle use the same intervention set (ablation) | MODL · Randomly generated additive causal models with parent count K | sample complexity | No clear effect |
Closest papers by meaning
- Partial Structure Discovery is Sufficient for No-regret Learning in Causal Bandits (NeurIPS 2024 · similarity 0.93)
- Linear Causal Bandits: Unknown Graph and Soft Interventions (NeurIPS 2024 · similarity 0.92)
- Causal Bandits: The Pareto Optimal Frontier of Adaptivity, a Reduction to Linear Bandits, and Limitations around Unknown Marginals (ICML 2024 · similarity 0.91)
- Adaptive Online Experimental Design for Causal Discovery (ICML 2024 · similarity 0.89)
- Robustly Improving Bandit Algorithms with Confounded and Selection Biased Offline Data: A Causal Approach (AAAI 2024 · similarity 0.89)
- Stochastic contextual bandits with graph feedback: from independence number to MAS number (NeurIPS 2024 · similarity 0.89)
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.