NSVKI Student Conference
November, 16th 2012
This year was the 4th year in row that we, the NSVKI, organized the student conference. Students from Amsterdam, Groningen, Nijmegen and Utrecht gathered in Utrecht to listen to each other’s AI-projects.
After a shift in the program due to the sudden absence of one of the speakers, the conference started with Peter Smit from Rijksuniversiteit Groningen.
His talk was about a machine learning project on the game quantum tic-tac-toe. He built a bot that could play this game. A problem of this quantum version of the game compared to the normal tic-tac-toe game is that the number of possible states of the game is too large to build a separate strategy for every state. Peter used a neural network to approximate a
good strategy without having the problem of the large number of states of the game. He showed us that this network was able to learn to play the game quite well.
The second speaker was Sjoerd Dost from Universiteit Utrecht. He built a theorem prover for Lambek-Grishin calculus. He showed us that formal languages can have different expressivity and complexity. He pointed out that natural language is somewhere between context-free and fully contextsensitive languages in the Chomsky hierarchy. He showed us how Lambek-Grishin calculus might be able to describe natural languages and how his theorem prover could determine the validity of sentences using such categorical grammars as the Lambek-Grishin calculus. After Sjoerd’s talk there was a break with coffee and tea.
Rijk’s talk was about his bachelorproject on the philosopher John Ralph. His main question was: ‘How can we distribute welfare in a fair way using principles of justice and game-theory?’ He explained what Ralph’s idea of a just welfare distribution was. Where welfare distributors have no knowledge about their own position in society and the maxmin principle is used as a goal to optimize welfare of the total society.
Roland Meertens, from the Radboud University in Nijmegen, talked about gesture based flight control. He used a Microsoft kinect to collect the raw input data. He compared several classification methods and experimented with different amount of distinct gestures. He also tried training the classifier on one person and testing it on another, but doing this caused the classification rate to drop quite a bit (from 80-90% to 30-40%).
To fill in the gap from the missing speaker, Robert Jansen and Kim Feenstra Kuiper, also from the Radboud University, were invited to present their Robotics project. For one of their courses they designed a Pacman game played with robots. The robots were built using lego NXT bricks. The field was implemented on a large wooden plate, using paper to stick the paths on it. They explained how the two ghost robots tried to enclose the
pacman robot.
