A multitasks learning approach to autonomous agent based on Genetic Network Programming

The standard methodology in machine learning is to learn one problem at a time. But, many real-world problems are complex and have multitasks, and it is a bit hard to learn them well by one machine learning approach. So, the simultaneous learning of several tasks has been considered, that is, so-called multitask learning. This paper describes a new approach to the autonomous agent problem using the multitask learning scheme based on Genetic Network Programming (GNP), called ML-GNP, where each GNP is used to learn one corresponding task. ML-GNP has some charateristics, such as distribution, interaction and autonomy, which are helpful for learning multitask problems. The experimental results illustrate that ML-GNP can give much better performance than learning all the tasks of the problem by one GNP algorithm.