Research map / Learning theory and algorithms / Prediction and decision losses
Symbolic Metamodels for Interpreting Black-Boxes Using Primitive Functions
Mahed Abroshan, Saumitra Mishra, Mohammad Mahdi Khalili
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 |
|---|---|---|---|
| Use SMPF instead of SP | Symbolic metamodeling of e^{-3x_0+x_1} · e^{-3x_0+x_1} | MSE: 0.009 ± 0.004 -> 0.001±0.0002 | Improved |
| Use SMPF instead of SP | Symbolic metamodeling of e^{-3x_0+x_1} · e^{-3x_0+x_1} | R²: 0.958± 0.014 -> 0.996 ± 0.002 | Improved |
| Use SMPF instead of SR | Symbolic metamodeling of sin(x_0x_1) · sin(x_0x_1) | MSE: 0.0004± 0.0002 -> 0.012± 0.002 | Hurt |
| Use SMPF instead of SR | Symbolic metamodeling of sin(x_0x_1) · sin(x_0x_1) | R²: 0.988± 0.003 -> 0.962 ± 0.004 | Hurt |
| Use SMPF instead of SP | Symbolic metamodeling of x_0x_1/(x_0²+x_1) · x_0x_1/(x_0²+x_1) | R²: 0.878± 0.021 -> 0.895 ± 0.013 | Improved |
| Use SMPF instead of SP | Symbolic metamodeling of sinc(x_0²+x_1) · sinc(x_0²+x_1) | MSE: 0.009 ± 0.002 -> 0.004± 0.0004 | Improved |
| Use SMPF instead of SP | Symbolic metamodeling of sinc(x_0²+x_1) · sinc(x_0²+x_1) | R²: 0.937± 0.015 -> 0.952± 0.003 | Improved |
| Use SMPF instead of SP | Approximating an MLP black box on Yacht · Yacht test, MLP black box | R² to black box: 0.978 ± 0.016 -> 0.993± 0.003 | Improved |
| Use SMPF instead of SP | Approximating an SVM black box on Yacht · Yacht test, SVM black box | MSE to black box: 0.014 ± 0.015 -> 0.029 ± 0.013 | Hurt |
| Use SMPF instead of SP | Approximating an SVM black box on Yacht · Yacht test, SVM black box | R² to black box: 0.974 ± 0.078 -> 0.967 ± 0.120 | Hurt |
| Use simpler primitive functions and KST tree search rather than Meijer G-based SM | Symbolic metamodeling · e^{-3x_0+x_1} | R²: 0.273± 0.019 -> 0.996 ± 0.002 | Improved |
Closest papers by meaning
- Demystifying Black-box Models with Symbolic Metamodels (NeurIPS 2019 · similarity 0.94)
- Learning outside the Black-Box: The pursuit of interpretable models (NeurIPS 2020 · similarity 0.87)
- End-to-end Symbolic Regression with Transformers (NeurIPS 2022 · similarity 0.85)
- SYMBOL: Generating Flexible Black-Box Optimizers through Symbolic Equation Learning (ICLR 2024 · similarity 0.85)
- MetaSymNet: A Tree-like Symbol Network with Adaptive Architecture and Activation Functions (AAAI 2025 · similarity 0.85)
- Symbolic regression via MDLformer-guided search: from minimizing prediction error to minimizing description length (ICLR 2025 · similarity 0.84)
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.
