Research map / Probabilistic and scientific modeling / Molecules and proteins
Breaking the Simplification Bottleneck in Amortized Neural Symbolic Regression
Paul Saegert, Ullrich Koethe
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 |
|---|---|---|---|
| replace SymPy with SimpliPy for simplification | Flash-ANSR training and inference · simplification of 64k expressions | simplification time: ~100ms -> millisecond regime | Improved |
| use SimpliPy instead of SymPy (Lmax >= 5) | simplification engine · simplification ratio | expression length ratio | No clear effect |
| train on simplified expressions instead of unsimplified (ablation) | Flash-ANSR 120M · FastSRB | vNRR | Mixed |
| train on simplified expressions instead of unsimplified (ablation) | Flash-ANSR 120M · FastSRB | bloat and rewrites: 21x more rewrites without simplification | Improved |
| use SymPy instead of SimpliPy in training (ablation) | Flash-ANSR 120M · FastSRB | vNRR: 20x slower | No clear effect |
| use softmax sampling instead of beam search (ablation) | Flash-ANSR 120M · FastSRB | vNRR: 9.4 pp fewer problems recovered by beam search | Improved |
| use softmax sampling instead of beam search (ablation) | Flash-ANSR 120M · FastSRB | distinct canonical hypotheses: 33% -> 48% | Improved |
| use 32-bit instead of 16-bit input encoding (ablation) | Flash-ANSR 120M · FastSRB | vNRR | Improved |
| use Pre-Norm instead of Post-Norm (ablation) | Flash-ANSR 120M · FastSRB | training stability: Post-Norm training failed | Improved |
| use N(0,5) constant prior instead of U(1,5) (ablation) | Flash-ANSR 120M · FastSRB | vNRR | No clear effect |
| use RMSSetNorm instead of LayerNorm (ablation) | Flash-ANSR 120M · FastSRB | vNRR | No clear effect |
| use Levenberg-Marquardt instead of BFGS (ablation) | Flash-ANSR · FastSRB | vNRR: ~10 pp higher vNRR | Improved |
Closest papers by meaning
- SMILE: Bridging Continuous Optimization and Discrete Symbolic Recovery (NeurIPS 2026 · similarity 0.88)
- Pareto-Optimal Fronts for Benchmarking Symbolic Regression Algorithms (ICML 2025 · similarity 0.87)
- Ab Initio Nonparametric Variable Selection for Scalable Symbolic Regression with Large $p$ (ICML 2025 · similarity 0.87)
- Symbolic regression via MDLformer-guided search: from minimizing prediction error to minimizing description length (ICLR 2025 · similarity 0.86)
- ParFam -- (Neural Guided) Symbolic Regression via Continuous Global Optimization (ICLR 2025 · similarity 0.86)
- SymMaP: Improving Computational Efficiency in Linear Solvers through Symbolic Preconditioning (NeurIPS 2025 · similarity 0.86)
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.
