Decoupling Representation and Learning in Genetic Programming: The LaSER Approach

Genetic Programming (GP) has traditionally entangled the evolution of symbolic representations with their performance-based evaluation, often relying solely on raw fitness scores. This tight coupling makes GP solutions more fragile and prone to overfitting, reducing their ability to generalize. In this work, we propose LaSER (Latent Semantic Representation Regression)–a general framework that decouples representation evolution from lifetime learning. At each generation, candidate programs produce features which are passed to an external learner to model the target task. This approach enables any function approximator, from linear models to neural networks, to serve as a lifetime learner, allowing expressive modeling beyond conventional symbolic forms. We show that LaSER can outperform standard GPstandard GP, particularly when equipped with nonlinear lifetime learners. While LaSER with linear models (e.g., ridge regression) already offers strong performance on smooth, low-complexity problems, its true advantage emerges as target functions grow in nonlinearity and complexity–where nonlinear learners provide a stronger inductive bias. We also present a first attempt at demonstrating an instance of the Baldwin EffectBaldwin Effectin symbolic regressionsymbolic regression: under LaSER, evolved representations become increasingly innate, reducing reliance on lifetime learning across generations. By explicitly separating the roles of representation and adaptation, LaSER offers a principled and extensible framework for symbolic modeling tasks.