Fitness importance for online evolution

To complement standard fitness functions, we propose "Fitness Importance" (FI) as a novel meta-heuristic for online learning systems. We define FI and show how it can be used to dynamically manipulate the population composition in order to vary the instantaneous system performance at a tradeoff to learning capability. The effect of FI is demonstrated on a simple light-sensing and light-actuating optimisation problem on a collection of physical wireless sensor network devices. We also describe the In situ Distributed Genetic Programming (IDGP) framework which has been developed for online evolution of logic on resource-constrained computing devices and demonstrate how FI can be used with the framework in order to achieve a dynamic balance of learning and performing. Categories and Subject Descriptors I.2.2 [Computing Methodologies]: Artificial Intelligence-Automatic Programming [Program synthesis] General Terms Algorithms, Design Keywords Late Breaking Abstract, Fitness, Obje...

• Type: Conference
• Year: 2010
• Where: GECCO
• Authors: Philip Valencia, Raja Jurdak, Peter Lindsay
• Keywords: GECCO 2010, Optimization, Physical Wireless Sensor, Standard Fitness Functions, Wireless Sensor Network