Research map / Probabilistic and scientific modeling / Molecules and proteins
Deep Generative Symbolic Regression with Monte-Carlo-Tree-Search
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 larger mutation size (pre-trained on mutation size ~10) instead of size 1 (ablation) | DGSR-MCTS mutation model · Synthetic in-domain and out-of-domain | % solved datasets: 74.8 vs 52.2 (in-domain); 44.0 vs 26.8 (out-of-domain) | Helped |
| Predict full expression in one step (no search-based incremental mutation) (ablation) | DGSR-MCTS mutation model · Synthetic out-of-domain | % solved datasets: 16.8 vs 44.0 for mutation size 10 | Hurt |
| Increase number of sampled expressions per expansion from greedy K=1 to K=8-16 (ablation) | DGSR-MCTS · Synthetic in-domain and out-of-domain | % solved datasets: K=16: 54.0 vs K=1: 9.6 (in-domain); K=8: 19.6 vs K=1: 0.8 (out-of-domain) | Helped |
| Perform constant optimization (BFGS) after every mutation instead of never (ablation) | DGSR-MCTS · Synthetic in-domain and out-of-domain | % solved datasets: 79.6 vs 74.8 (in-domain); 66.2 vs 44.0 (out-of-domain) | Helped |
| Add synthetic datasets for training critic and fine-tuning mutation model (ablation) | DGSR-MCTS · SRBench black-box and Feynman | median R^2 (black-box) / proportion R^2>0.99 (Feynman): black-box 0.801->0.812 (no simultaneous); 0.823->0.846 (simultaneous); Feynman 0.655->0.748, 0.689->0.796 | Helped |
| Train simultaneously on multiple SRBench datasets instead of one dataset at a time (ablation) | DGSR-MCTS · SRBench black-box and Feynman | median R^2 (black-box) / proportion R^2>0.99 (Feynman): black-box 0.801->0.823 (no synthetic); 0.812->0.846 (with synthetic); Feynman 0.655->0.689 and 0.748->0.796 | Helped |
Closest papers by meaning
- Improving Monte Carlo Tree Search for Symbolic Regression (NeurIPS 2025 · similarity 0.91)
- A Unified Framework for Deep Symbolic Regression (NeurIPS 2022 · similarity 0.91)
- Symbolic Regression with a Learned Concept Library (NeurIPS 2024 · similarity 0.89)
- Symbolic regression via MDLformer-guided search: from minimizing prediction error to minimizing description length (ICLR 2025 · similarity 0.87)
- FunctionEvolve: Structure-Guided Symbolic Regression with LLMs (NeurIPS 2026 · similarity 0.87)
- A Neural-Guided Dynamic Symbolic Network for Exploring Mathematical Expressions from Data (ICML 2024 · similarity 0.87)
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.