JSAI2019

Presentation information

Interactive Session

[4Rin1] Interactive Session 2

Fri. Jun 7, 2019 9:00 AM - 10:40 AM Room R (Center area of 1F Exhibition hall)

9:00 AM - 10:40 AM

[4Rin1-40] Introducing Modular Neural Networks in Learning Action Games

〇Toshinori Takahashi1, Masaharu Munetomo2 (1. Hokkaido University, 2. Information Initiative Center, Hokkaido University)

Keywords:AI, Neural network, NEAT, Game learning, Modular network

In this paper, we introduce a modular network approach in neuroevolution for action game learning.
We employ NEAT (NeuroEvolution of Augmenting Topologies) to generate modular networks learning game
stages under different conditions, which are combined to obtain networks that can adapt to difficult situations
appeared in actual game playing. We show the effectiveness of our approach in a Pygame instance
compared to the original NEAT.