Evolving simple and accurate symbolic regression models via asynchronous parallel computing

In machine learning, reducing the complexity of a model can help to improve its computational efficiency and avoid overfitting. genetic programming (GP), reduction is often achieved by size evolved expressions. However, previous studies have demonstrated that expression does not necessarily prevent Therefore, this paper uses evaluation time – required evaluate GP on data as estimate complexity. The depends only expressions but also their composition, thus acting more nuanced measure than alone. To discourage complexity, study employs novel method called asynchronous parallel (APGP) introduces race condition in evolutionary process GP; offers an advantage simple solutions when accuracy competitive. proposed method, it compared standard (GP) with bloat control (GP+BC) methods six challenging symbolic regression problems. APGP produced models are significantly accurate (on 6/6 problems) those both GP+BC. terms control, prevailed over GP+BC; however, GP+BC simpler at cost test-set accuracy. Moreover, took lower number evaluations meet target training fitness all tests. Our analysis involved: (1) ablation separated from (2) initialisation scheme encourages functional diversity initial population improved results for methods. These question overall benefits endorse employment controlling it.