Automatically Balancing Model Accuracy and Complexity using Solution and Fitness Evolution (SAFE)

  title={Automatically Balancing Model Accuracy and Complexity using Solution and Fitness Evolution (SAFE)},
  author={Moshe Sipper and Jason H. Moore and Ryan J. Urbanowicz},
. When seeking a predictive model in biomedical data, one of-ten has more than a single objective in mind, e.g., attaining both high accuracy and low complexity (to promote interpretability). We investigate herein whether multiple objectives can be dynamically tuned by our recently proposed coevolutionary algorithm, SAFE (Solution And Fitness Evolution). We find that SAFE is able to automatically tune accuracy and complexity with no performance loss, as compared with a standard evolutionary… 

Figures and Tables from this paper



Investigating the parameter space of evolutionary algorithms

It is shown that parameter space tends to be rife with viable parameters, at least for the problems studied herein, and the implications for researcher employing EC are discussed.

Predicting the difficulty of pure, strict, epistatic models: metrics for simulated model selection

This study formally identifies and evaluates metrics which quantify model detection difficulty and utilizes these metrics to intelligently select models from a population of potential architectures for improved simulation study design which accounts for differences in detection difficulty attributed to model architecture.

GAMETES: a fast, direct algorithm for generating pure, strict, epistatic models with random architectures

GAMETES is a fast, flexible, and precise tool for generating complex n-locus models with random architectures, and is proficient at generating the lower heritability models typically used in simulation studies evaluating new algorithms.

Solution and Fitness Evolution (SAFE): Coevolving Solutions and Their Objective Functions

This work presents Solution And Fitness Evolution (SAFE), a commensalistic coevolutionary algorithm that maintains two coevolving populations: a population of candidate solutions and apopulation of candidate objective functions, and shows that SAFE successfully evolves not only solutions within a robotic maze domain, but also the objective functions needed to measure solution quality during evolution.

Solution and Fitness Evolution (SAFE): A Study of Multiobjective Problems

An investigation of SAFE’s adaptation and application to multiobjective problems, wherein candidate objective functions explore different weightings of each objective, suggests that SAFE, and the concept of coevolving solutions and objective functions, can identify a similar set of optimal multiObjective solutions without explicitly employing a Pareto front for fitness calculation and parent selection.

ExSTraCS 2.0: description and evaluation of a scalable learning classifier system

Performance over a complex spectrum of simulated genetic datasets demonstrated that these new mechanisms dramatically improve nearly every performance metric on datasets with 20 attributes and made it possible for ExSTraCS to reliably scale up to perform on related 200 and 2000-attribute datasets.

A Large-Scale Experimental Evaluation of High-Performing Multi- and Many-Objective Evolutionary Algorithms

A systematic, comprehensive evaluation of a large number of MOEAs that covers a wide range of experimental scenarios and confirms some of the assumed knowledge in the field, while at the same time providing new insights on the relative performance ofMOEAs for many-objective problems.

A fast and elitist multiobjective genetic algorithm: NSGA-II

This paper suggests a non-dominated sorting-based MOEA, called NSGA-II (Non-dominated Sorting Genetic Algorithm II), which alleviates all of the above three difficulties, and modify the definition of dominance in order to solve constrained multi-objective problems efficiently.

Fuzzy CoCo: a cooperative-coevolutionary approach to fuzzy modeling

This paper introduces a novel algorithm, called Fuzzy CoCo (fuzzy cooperative coevolution), and demonstrates the efficacy by applying it to a hard, real-world problem - breast cancer diagnosis, obtaining the best results to date while expending less computational effort than previous processes.

The Use of an Analytic Quotient Operator in Genetic Programming

It is demonstrated that this AQ operator systematically yields lower mean squared errors over a range of regression tasks, due principally to removing the discontinuities or singularities that can often result from using either protected or unprotected division.