The PhD Research Fellow Per-Arne Andersen from UiA’s CAIR centre won the award for best presentation at a Cambridge conference. 260 researchers of artificial intelligence from all over the world participated.
Professors and researchers from all over the world presented new discoveries and perspectives regarding artificial intelligence at the SGAI International Conference on Artificial Intelligence in Cambridge.
There were over 260 presentations at the conference. The best one was the presentation of the dreaming algorithm by Per-Arne Andersen from the University of Agder. Andersen held his award-winning presentation at the International Conference on Artificial Intelligence at Cambridge University on 13 December.
Under the supervision of Associate Professor Morten Goodwin and Professor Ole-Christoffer Granmo from UiA’s Centre for Artificial Intelligence Research (CAIR), Andersen has created a dreaming algorithm that creates new landscapes and new possibilities in a computer game through dreaming.
Requires small amounts of data
Artificial intelligence algorithms, also called recipes, are self-learning. They usually require large amounts of data to make intelligent decisions, but the recipe Andersen has created does not. He has created an algorithm that only requires small amounts of data to learn the game.
"This algorithm can create new game situations and develop its own data while it is running," Morten Goodwin said.
Goodwin definitely believes getting an award for best study already as a first-year PhD research fellow is a feather in the cap for Andersen.
Goodwin also points out that Cambridge University is among the universities with a long and prestigious history within research and development of artificial intelligence. Alan Turing was one the field’s pioneers and researched and developed his theories at Cambridge.
"I have named the algorithm The Dreaming Algorithm, and, in short, it is made in such a way that on its own it can create the data it needs to play well," Per-Arne Andersen said.
The algorithm carries out random actions to investigate how the game works. After the algorithm has created a certain amount of data, it has, so to speak, learned the grammar or anatomy of the game. From that, it can create new future visions. The dreaming algorithm dreams up possible development traits, and finally imagines all the future development traits that exist in the game.
He believes the algorithm can be of commercial interest."The work on the algorithm started with my master’s thesis and the use of artificial intelligence in games. In so-called research learning where the algorithm teaches itself to understand a game, the algorithm usually requires very large amounts of data. This makes the learning process slow and hard to carry out with limited computing power. The idea behind this algorithm was that it should learn from parts of the game and from that imagine new game stories," the researcher said.
"An example world be environments where the consequences for making mistakes are fatal. You could then use an algorithm like this to teach the other algorithms in a dream environment that test different solutions without there being no damage in case of mistakes," he said.
Another area where the algorithm could be used is medicine.
"There is a shortage of data on medical diagnoses. This algorithm could potentially dream up new examples of x-ray images and show a development that the doctor would otherwise not have discovered," Morten Goodwin said.
Expanding the research front
The study he presented in Cambridge is the third one he has presented at a conference. In total, he now has six public publications on his CV, including his master’s thesis. The young researcher’s goal is that his doctoral thesis will expand the research front for using artificial intelligence in games.
"The dreaming algorithm is a small step in the right direction towards more effective and less computer-demanding algorithms, but this is not the end of the research," he said.
Per-Arne Andersen walked up the podium at Cambridge University and accepted the award for the best study at the artificial intelligence conference on 13 December. This was not the end. It was only the beginning.