Finding and inspecting equations in data
Symbolic regression searches for mathematical expressions that describe patterns in data. An equation makes the candidate relationships explicit so researchers can inspect them alongside the data and scientific context.
Dr. Costilla Reyes’s work studies how the search can reuse useful structure rather than rebuilding every expression from scratch. InceptionSR carries partial equations into later rounds as features; a learned concept-library approach explores reusable abstractions for equation discovery.
Comparing search strategies and candidate models
The publication record covers recursive equation synthesis, learned concepts, and evolutionary search. A body-fat estimation study compares evolutionary-computation methods and produces mathematical expressions for a defined dataset. A later clinician-evaluated preprint examines how language models can support review of those expressions and their explanations.
These studies address complementary parts of the problem: finding candidate equations, reusing structure, and reviewing results against the study data and evaluation criteria.
Evaluating candidate equations
The research evaluates candidate equations against the data and criteria used in each study. Domain expertise helps researchers interpret the resulting expressions and assess their relevance to the scientific question.