Research projects
Algorithms and machine learning
Approximation and learning algorithms (ALEA)
Period: 1/2005-12/2008
Researchers: Krishnan Narayanan, Jyrki Kivinen
Funding: Academy of Finland
Approximation algorithms are methods with which to search for less-than-best solutions for computational problems. The idea is to save computational resources with problems for which it is difficult to find exact solutions. The basic method in machine learning is to create a simple hypothesis based on sample data to explain the samples. In such cases, it usually makes sense to use approximation algorithms, because it may be difficult computationally to find a simple hypothesis, and it is often unnecessary to even try to explain all the samples exactly. The samples may contain mistakes that the learning algorithm had better not copy. The project studies approximation algorithms with special attention to the needs of machine learning. One goal is to modify traditional approximation algorithm that are based on analysing worst-case scenarios, so that they can utilise the more commonly occurring easy cases better. Another approach is the development of models and methods for machine learning that bypass the notoriously difficult approximation problem, either by changing the representation of the hypothesis or by formalising the learning problem in a whole different way.
Content-Based Retrieval and Analysis of Harmony and other Music Structures (C-BRAHMS)
Period: 8/2005-7/2010
Researchers: Kjell Lemström, Väinö Ala-Härkönen, Johan Brunberg, Niko Mikkilä, Veli Mäkinen
Funding: Academy of Finland
This project designs and implements algorithms and data structures for the analysis of symbolically encoded music and for content-based retrieval. The results of the C-BRAHMS research are used in a prototype system that is distributed in accordance with the GNU GPL licence. Please see http://www.cs.helsinki.fi/group/cbrahms/ for the system and further information on the project.
Statistical Multilingual Analysis for Retrieval and Translation (SMART)
Period: 10/2006-9/2009
Researchers: Juho Rousu, Wray Buntine (NICTA), Matti Kääriäinen, Vladimir Poroshin, Kimmo Valtonen, Matti Vuorinen, Huizhen Yu
Funding: EU
The goal of this project is to develop new methods of statistics and machine learning for multilingual information retrieval and machine translation. The academic partners of the project are the universities of Southampton and Bristol , University College London, Università degli Studi di Milano, Josef Stefan Institute and National Research Council Canada. In the year 2007 Department of Computer Science developed new methods for decoding, methods for predicting the fluency and grammaticality of the translations, new learning methods for structured prediction and methods for multilingual lexicon extraction