Efficient generator of mathematical expressions for symbolic regression

Sebastian Mežnar, Sašo Džeroski, Ljupčo Todorovski

Machine Learning 2023 article cited by 20

Source Open-access PDF More in Mathematics

importance 4/10 partially solved Evolutionary Algorithms and Applications scientific computing stage: research

Problem

Symbolic regression — finding mathematical equations that fit data — is computationally hard.

Why it matters

Recovering interpretable equations from data supports scientific discovery across many fields.

What the AI did

A hierarchical variational autoencoder (HVAE) learns a smooth latent space of mathematical expressions from small corpora; the EDHiE system searches this latent space with an evolutionary algorithm to reconstruct equations.

Results

Random search in HVAE latent space outperforms random search over grammar-generated expressions; EDHiE reconstructs equations from a standard symbolic regression benchmark (details truncated).

Impact

Peer-reviewed demonstration of a working symbolic regression approach on a standard benchmark.

Record

Venue	Machine Learning
Published	2023
Work type	article
Authors	Sebastian Mežnar, Sašo Džeroski, Ljupčo Todorovski
Citations	20 (OpenAlex, at import time)
References	39
Model or method	HVAE; EDHiE
Organization	not stated in source
Primary source	https://doi.org/10.1007/s10994-023-06400-2
Open access	hybrid
Keywords	symbolic regression, variational autoencoder, latent space, evolutionary algorithm, equation discovery

Importance rating

4 / 10

real but narrower advance

Symbolic regression is a useful scientific tool; abstract demonstrates benchmark improvements only, with no stated application scale or beneficiary count.

Basis: inferred. Judged from the source text, which gives no explicit scale. Stated scale: not stated in source. Method and limits.

More in Mathematics

#	Paper	Year	Citations	Importance
1	Discovering faster matrix multiplication algorithms with reinforcement learning Alhussein Fawzi, Matej Balog, Aja Huang, Thomas Hubert, Bernardino Rom...	2022	476	7
2	Solving olympiad geometry without human demonstrations Trieu H. Trinh, Yuhuai Wu, Quoc Viet Le, He He, Thang Luong	2024	307	7
3	Mathematical discoveries from program search with large language models Bernardino Romera‐Paredes, Mohammadamin Barekatain, Alexander Novikov,...	2023	509	6
4	Superhuman AI for multiplayer poker Noam Brown, Tüomas Sandholm	2019	583	6
5	Symbolic regression as a feature engineering method for machine and deep learning regression tasks Assaf Shmuel, Oren Glickman, Teddy Lazebnik	2024	19	4