PhD Job Vacancy in The Netherlands
The Netherlands: PhD Position in Adaptive Distance Measures in Relevance Learning Vector Quantization, University of Groningen
Project Description:
Learning Vector Quantization (LVQ) is a family of appealing algorithms used in a variety of classification problems. An iterative training process determines from a given set of example data a set of prototypes which represent typical features and parameterize a distance based classification scheme. A key difficulty is the choice of an appropriate distance measure. Frequently, Euclidean metrics are employed without further justification, or different measures are compared in a
trial-and-error approach. An appealing solution to this problem is the use of adaptive metrics. Relevance learning schemes have been proposed which assign a relevance factor to each dimension in feature space. These factors and the prototypes are updated at the same time during training. One aim of the project is a better theoretical understanding of relevance learning. To this end, we will extend our previous mathematical description of the learning dynamics and performance of LVQ.
The focus of the project will be on an important extension of relevance learning that we have recently suggested: the adaptation of relevance matrices during training. Matrix Relevance Learning does not only weight single features differently, but takes into account correlations between them, as well. As a testbed for the novel class of algorithms, it will be implemented and benchmarked in practical problems of, e.g., medical data analysis, classification of bioinformatics data and image processing problems.
Qualification:
You hold a university degree (diploma or master of science) in Computer Science, Physics or a related discipline with an excellent academic record. You are highly interested in both, theoretical and applicational aspects of machine learning. Proficiency in English (both oral and written) and excellent communication skills are indispensible for this project. This concerns, in particular, the ability to write scientific articles and reprorts.
Organization:
The project will be embedded in the research group Intelligent Systems with Dr. Michael Biehl as thesis supervisor.
Conditions of employment:
PhD students are employed for a maximum period of four years. Salaries are according to the standard salary scale for PhD students with an estimated gross salary of 1956,- Euros per month in the first year, increasing to a monthly salary of 2502,- Euros in the fourth year.
Contact information:
Dr. Michael Biehl (m.biehl@rug.nl, Tel: +31 50 363 3997)
Information about on-going research, re- and preprints etc. are available from: www.cs.rug.nl/~biehl
Application:
Initially, send a short CV including information about your academic degrees and grades, as well as a letter of motivation to Michael Biehl (preferred format: pdf or ps). The position will be open until a suitable candidate is found.
Project Description:
Learning Vector Quantization (LVQ) is a family of appealing algorithms used in a variety of classification problems. An iterative training process determines from a given set of example data a set of prototypes which represent typical features and parameterize a distance based classification scheme. A key difficulty is the choice of an appropriate distance measure. Frequently, Euclidean metrics are employed without further justification, or different measures are compared in a
trial-and-error approach. An appealing solution to this problem is the use of adaptive metrics. Relevance learning schemes have been proposed which assign a relevance factor to each dimension in feature space. These factors and the prototypes are updated at the same time during training. One aim of the project is a better theoretical understanding of relevance learning. To this end, we will extend our previous mathematical description of the learning dynamics and performance of LVQ.
The focus of the project will be on an important extension of relevance learning that we have recently suggested: the adaptation of relevance matrices during training. Matrix Relevance Learning does not only weight single features differently, but takes into account correlations between them, as well. As a testbed for the novel class of algorithms, it will be implemented and benchmarked in practical problems of, e.g., medical data analysis, classification of bioinformatics data and image processing problems.
Qualification:
You hold a university degree (diploma or master of science) in Computer Science, Physics or a related discipline with an excellent academic record. You are highly interested in both, theoretical and applicational aspects of machine learning. Proficiency in English (both oral and written) and excellent communication skills are indispensible for this project. This concerns, in particular, the ability to write scientific articles and reprorts.
Organization:
The project will be embedded in the research group Intelligent Systems with Dr. Michael Biehl as thesis supervisor.
Conditions of employment:
PhD students are employed for a maximum period of four years. Salaries are according to the standard salary scale for PhD students with an estimated gross salary of 1956,- Euros per month in the first year, increasing to a monthly salary of 2502,- Euros in the fourth year.
Contact information:
Dr. Michael Biehl (m.biehl@rug.nl, Tel: +31 50 363 3997)
Information about on-going research, re- and preprints etc. are available from: www.cs.rug.nl/~biehl
Application:
Initially, send a short CV including information about your academic degrees and grades, as well as a letter of motivation to Michael Biehl (preferred format: pdf or ps). The position will be open until a suitable candidate is found.
Labels: Netherlands
<< Home