Current Research:

I currently follow my master degree in Game AI at the Laval University, Department of Computer Science and Software Engineering  under the direction of Mr. Luc Lamontagne. My research is focused primarily on the video games and the artificial intelligence. We currently work on the resolution of the problem of Tetris game by using the techniques of the case based reasoning and reinforcement learning. The infinite space of observations of the game obliges the reduction of complexity by using the similarity. The intelligence is enforced with time by the reinforcement learning process. The use of the forgetting (elimination on unused cases) considerably reduces the size of the case base and improved game performance.

Research interest:

I am personally interested by the artificial intelligence and the intelligent processes. The improvement of the behavioral aspect in the industrial field, robots and games belongs to my first priorities. My work also involves the interactive and intelligent interfaces in the multimedia applications. I am also interested by integration of intelligent decision systems in the real-time applications (military and industrial applications, intelligent systems constrained by time, real time systems) and the operating systems of the future (intelligent interfaces and OS/kernels).

Publications:

• Romdhane, H.; Lamontagne, L. (2008), “Forgetting Reinforced Cases”, Proceedings of the European Conference on CBR, Springer-Verlag. pdf

• Romdhane, H., Lamontagne, L., (2008) Reinforcement of Local Pattern Cases for Tetris, Proceedings of Twenthy First International FLAIRS Conference, AAAI Press, pp. 263-269. Florida, USA. pdf

Tetris Project: We created a CBR system that proceed by similarity to learn tetris how to play the game, using only past experience. We have introduced reinforcement learning to re-evaluate cases, and forgetting policy to reduce case base size after training. This image shows Tetris game after reinforcing cases for more than 500 plays, and forgetting cases using usage and value of cases. This is the performance of the resulting case base with about 5000 cases.