Agnostic Learner

Assigned Computer Science Games Engineering eXtended Artifical Intelligence Bachelor Master Project Thesis Games Research Lab

Background

There are different ways to simulate real-world phenomena. Consider, for instance, the ghosts in the video game pac-man. They see pac-man when moving along the same path segment and then start chasing him. These non-player characters are modelled as so-called agents equipped with sensors, actuators and controllers that connect the received information to the resulting actions. Agent-based modelling is a very wide-spread approach to modelling and simulating various real-world phenomena—from the interaction of biological cells over the simulation of crowd movement, or population-wide happiness. All the empirically unearthed aspects of real-world agents can be reflected by their digital counterparts. However, since the interactions of agents are so open-ended, their degrees of interaction so great, the ensuing simulations are quickly computationally rather costly.

Task

Based on the background explained above, in order to arrive at viable computational loads, the degrees of freedom have to be systematically reduced - but only in such a way that relevant simulated results would not be impeded. To this end, we need to identify and learn the interaction patterns that unfold in the simulation. This is the task of this project/thesis work. You need to setup a simulation in which patterns at various spatio-temporal scales occur and design, implement and evaluate an algorithmic framework that captures these patterns and detects when they unfold.

In ordert to ensure a smooth and easy foundation for your experiments, I suggest you:

• rely on a cellular automaton (find an existing implementation or do it yourself),
• configure it to yield patterns of various levels of complexity (recipies can be the rules of Conways’ Game of Life and other patterns can be looked up in Wolfram’s book, see below).

For the algorithm’s design, I would recommend:

• setting up a time-series representation that can reflect the interactions and state changes in the observed model,
• observing patches of the cellular automaton of different sizes,
• observing these patches for a given time interval (“sliding window”),
• comparing observations with recordings in order to determine whether a pattern has been detected and also how strong/significant it is.

Literature

• Wolfram, S. (2002). A new kind of science (Vol. 5, p. 130). Champaign, IL: Wolfram media.
• http://www.scholarpedia.org/article/Game_of_Life (this is a legit scientific, peer-reviewed source, it’s not Wikipedia!)
• von Mammen, S., Steghöfer, J. P., Denzinger, J., & Jacob, C. (2011). Self-organized middle-out abstraction. In Self-Organizing Systems: 5th International Workshop, IWSOS 2011, Karlsruhe, Germany, February 23-24. 2011. Proceedings 5 (pp. 26-31). Springer Berlin Heidelberg.

Contact Persons at the University Würzburg