Email: ilker DOT yildirim AT boun DOT edu DOT tr
Here you can find a small survey on Bayesian learning and Bayesian models of cognition that I prepared.
Here you can find a Markov chain Monte Carlo (MCMC) example and implementation guidelines for it. It is helpful especially for dummies!
I use Repast for doing Agent Based Simulation, a common tool. I will try to share personal experience about running multiple simulations here soon.
In May 2008, I attended to the 10th European Agent Systems Summer School. The summer school was a remarkable experience for me, because I got introduced to most recent research directions and areas in multiagent systems. Apart from lectures, a student session is held during the summer school. My work, in which I described my proposal for my master thesis, is peer reviewed, and accepted as one of 6 papers to the student session. I presented this work during the student session in the EASSS08.
I am developing a multiagent system in which each agent models other agents in the system in terms of the extent and depth of their knowledge in a particular concept. But we want an agent to be able to effectively model another agent just after a few interactions with that other agent. More precisely, each agent has an inductive problem: after collecting a sparse data coming from few number of interactions with another agent, model that other agent in terms of the extent and depth of its knowledge. In particular, this multiagent system consists of agents that can recognize several signs in Turkish Sign Language, and continue to learn new signs and forget what they knew before. After a few interactions, which are sort of asking a particular agent to recognize a particular sign, the agent is to guess what is the extent and depth of other agents' knowledge of Turkish Sign Language. We are developing a Bayesian Model, by combining the data coming from few number of interactions and the possible structures of knowledge, to accomplish that task. I am preparing that work to a conference submission.
In this study we investigate the problem of selfish invasion in cooperative symbiotic groups. Bruce Edmonds proposed a tag based model in which symbiotic groups emerged in an environment such that agents are capable of harvesting different types of foods and they need to have all types of resources in order to survive. Although collaborative groups arises in the model of Bruce Edmonds, they are wiped out due to selfish individual domination, hence these groups are not maintained. To achieve symbiotic groups that do not suffer from selfish invasion and so maintained, we developed hybrid models of cooperation. In addition to tag mechanism from Bruce Edmonds' model, we proposed three different reciprocity mechanisms. In our hybrid model the symbiotic groups not only formed earlier, but also maintained far longer. Our results are published in the plenary session of fifth Conference of European Social Simulation Association, ESSA08.
This research initially was a term project for the course Complex Systems, that I attended during Fall 2008. Then I studied further on it. In that project, we investigated a crucial aspect of cooperation: is it better to cooperate with the similar, or with the complement. For that purpose we designed and implemented a simulation model. In a setting in which agents could cooperate with other similar agents as well as complement agents, we studied the success of agents who are more inclined to cooperate with similar agents and who are more inclined to cooperate with complement agents. Our results show that individuals that prefer to cooperate with complement agents are more successful. We are preparing that work for submission to a journal.