Research map / Learning theory and algorithms / Combinatorial and search optimization
On the Role of Sparsity and DAG Constraints for Learning Linear DAGs
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 |
|---|---|---|---|
| replace least-squares objective with hard DAG constraint (NOTEARS-L1) by likelihood objective with soft constraints (GOLEM) | linear DAG learning · large ER2 graphs, Gaussian-EV | normalized SHD / TPR | Helped |
| use GOLEM-NV instead of NOTEARS-L1 | Gaussian-NV ER2/ER4 graphs · nonidentifiable Gaussian-NV | SHD | Helped |
| GOLEM-EV instead of NOTEARS-L1 on sparse vs dense graphs | Gaussian-NV · ER1 vs ER4 | SHD | Mixed |
| GOLEM-NV vs GOLEM-EV when noise variances are unequal | GOLEM · Gaussian-NV | SHD | Helped |
| GOLEM-NV vs NOTEARS-L1 on real data | GOLEM-NV · Sachs | SHD: 15 -> 14 | Helped |
| add DAG penalty to l1 penalty (ablation) | GOLEM-NV · Gaussian-NV, 100-node ER1/ER4 | TPR | Helped |
| add DAG penalty to l1 penalty (ablation) | GOLEM-EV · Gaussian-EV, small samples/dense graphs | SHD/TPR | Helped |
| add l1 and DAG penalties vs none (ablation) | GOLEM-EV · Gaussian-EV | TPR | Helped |
| initialize GOLEM-NV with GOLEM-EV solution | GOLEM-NV · synthetic | avoiding local solutions | Helped |
| GOLEM-EV vs NOTEARS-L1 under weight scaling | GOLEM-EV · 50-node ER2, weight scale alpha 0.3-1.0 | SHD/SID/TPR | Helped |
Closest papers by meaning
- Analytic DAG Constraints for Differentiable DAG Learning (ICLR 2025 · similarity 0.91)
- CoLiDE: Concomitant Linear DAG Estimation (ICLR 2024 · similarity 0.90)
- Markov Equivalence and Consistency in Differentiable Structure Learning (NeurIPS 2024 · similarity 0.89)
- ProDAG: Projected Variational Inference for Directed Acyclic Graphs (NeurIPS 2025 · similarity 0.88)
- Boosting Causal Structure Learning: An Asymmetric Exponential Modulation Gaussian-Based Adaptive Sample Reweighting Framework (AAAI 2025 · similarity 0.88)
- Learning Gaussian DAG Models without Condition Number Bounds (ICML 2025 · similarity 0.88)
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.