The 14th Finnish Artificial Intelligence Conference (STeP 2010), co-organized by FAIS and A?, was held 17-18 August in Otaniemi. The conference had invited speakers around Finland: Jarmo Alander from University of Vaasa, Timo Honkela from Aalto University, Ville Kyrki from Lappeenranta University of Technology, Tapio Salakoski from University of Turku and Harri Valpola from Aalto University and ZenRobotics Ltd. There were also presentations for ten contributed papers and a couple of demos.
Jarmo Alander talked about computational intelligence in Vaasa, where especially evolutionary algorithms and their applications are investigated.
Timo Honkela discussed subjectivity, which is mostly ignored in artificial intelligence systems. The first level of subjectivity considers the fact that each individual agent has its own model of the world. The second level of subjectivity goes deeper and takes into account that symbols in the models are not shared between agents, but have to be grounded individually for each agent.
Ville Kyrki made the claim that robots haven't really improved from Shakey, that was built in the late 60s and early 70s. The most recent significant developments have been made with robots operating in unstructured environments, such as the BigDog robot or the DARPA Urban Challenge.
Tapio Salakoski showed how methods for natural language processing (NLP) can be applied in the bio-medical domain, for instance, by extracting events from millions of scientific publications and using machine learning for finding connections between the events.
Harri Valpola discussed how artificial intelligence should be constructed gradually. They tackle the goal of constructing an artificial brain in such manner, taking into account the evolutionary development of the vertebrate brain. They also use their models for motor control, which they say is the original purpose of the brain.
The presented papers had a range of topics, including solving the Rubik's cube with a genetic algorithm, new genetic operators for image registration, building a platform for software program comprehension, domain adaption for statistical machine translation, and handling missing values with principal component analysis.
1 comment:
An interesting recent news is that every position of the Rubik's cube can be solved at most in 20 moves. Humans typically solve the cube using algorithms that require more moves.
Post a Comment