Research map / Probabilistic and scientific modeling / Causal discovery
Exploiting Experts’ Knowledge for Structure Learning of Bayesian Networks
Hossein Amirkhani, Mohammad Rahmati, Peter Lucas, Arjen Hommersom
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 experts' opinions via marginalization-based score instead of data only | hill-climbing BDeu · Hailfinder simulated (weak experts, |D|=1000, DAG) | SHD: 67.0 -> 37.9 | Improved |
| add experts' opinions via marginalization-based score instead of data only | hill-climbing BDeu · Breast cancer network (real experts, average over |D|) | SHD: 8.8 -> 7.3 | Improved |
| use raw opinions as perfect experts for weak population (ablation) | explicit-accuracy score · Hailfinder simulated (weak experts, |D|=1000, DAG) | SHD: 67.0 -> 85.3 | Hurt |
| use experts' opinions only without data (ablation) | majority vote · Hailfinder simulated (weak experts, |D|=1000, DAG) | SHD: 67.0 -> 388.1 | Hurt |
| estimate accuracies with EM instead of fixed mean accuracy, weak experts (ablation) | explicit-accuracy score · Hailfinder simulated (weak experts, |D|=1000, DAG) | SHD: 43.5 -> 46.0 | Hurt |
| estimate accuracies with EM instead of fixed mean accuracy, good experts (ablation) | explicit-accuracy score · Alarm simulated (good experts, |D|=1000, DAG) | SHD: 18.6 -> 12.7 | Improved |
| estimate accuracies with EM on real radiologists (ablation) | explicit-accuracy score · Breast cancer network (real experts, average over |D|) | SHD: 7.1 -> 20.3 | Hurt |
| use marginalization instead of EM-estimated accuracies, weak experts (ablation) | marginalization-based score · Alarm simulated (weak experts, |D|=1000, DAG) | SHD: 35.1 -> 27.8 | Improved |
| increase training data from 1000 to 5000 (ablation) | Data only (BDeu) · Hailfinder simulated (weak experts, DAG) | SHD: 67.0 -> 69.4 | Hurt |
| vary marginalization coefficient c in {0.1, 1, 10, 100} (ablation) | marginalization-based score · Asia and Alarm simulated (Table 9) | SHD | No clear effect |
Closest papers by meaning
- Extendable and Iterative Structure Learning Strategy for Bayesian Networks (ICLR 2025 · similarity 0.90)
- Improving Bayesian Network Structure Learning in the Presence of Measurement Error (JMLR 2022 · similarity 0.90)
- Structure Learning in Bayesian Networks of a Moderate Size by Efficient Sampling (JMLR 2016 · similarity 0.89)
- Mitigating Prior Errors in Causal Structure Learning: A Resilient Approach via Bayesian Networks (TPAMI 2025 · similarity 0.89)
- DiBS: Differentiable Bayesian Structure Learning (NeurIPS 2021 · similarity 0.89)
- Finding All Bayesian Network Structures within a Factor of Optimal (AAAI 2019 · 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.
