Several fully funded Ph.D. positions in machine learning, speech recognition, handwriting recognition and robotics are available at the Reservoir Lab (http://reslab.elis.ugent.be) and the Speech Lab (http://speech.elis.ugent.be), both part of the Electronics and Information Systems Department, faculty of Engineering of the Ghent University, Belgium (http://ugent.be).
Project
Current state-of-the-art speech & handwriting recognition systems still perform much worse than human beings who can effortlessly decode the speech or handwriting of most people, even in fairly adverse conditions (e.g. the presence of noise in case of speech recognition). The fact that the human brain works so efficiently is owed to its self-organizing capacity, its deeply hierarchical approach, its adoption of unsupervised and supervised learning strategies, its capacity to adapt almost instantly to new circumstances, etc. Why not try to build an automatic speech recognizer and handwriting recognition engine that incorporates the same principles? This is exactly what we will do in two recently approved projects:
The research concerns the investigation of architectures and algorithms for the efficient learning of large recurrent neural networks based on the Reservoir Computing concept (where only a linear readout layer is learned in a supervised way whereas the recurrent connections are fixed or trained in an unsupervised way). Important research topics are the unsupervised learning of a large hierarchy of recurrent sub-layers, and the integration of various adaptation techniques. The application domains are off-line handwriting recognition, speech recognition and various aspects of robotics (such as robot localization, motion control, ...). So far we were able to demonstrate that reservoirs can give rise to the robust recognition of digits spoken or written in isolation, but now we want to demonstrate that they can also yield robust recognition of continuous speech and handwriting (large vocabulary).
Requirements
Candidates should have a Masters degree in Electrical, Computer or Physics Engineering; or in Physics, Mathematics or Computer Science. A good knowledge of English is essential. No professional background is required, but the ideal candidates have some acquaintance with Machine Learning, programming (Python, Matlab, ...), statistics, signal processing, speech recognition, control engineering, or robotics.
What we offer
We offer an opportunity to perform at least three years of research in a new promising domain, and to get a doctoral degree in this domain. There will be ample opportunities for establishing international contacts (stays at partner universities, participation to international conferences). As an employee of the university you will receive a competitive salary (starting with a net monthly salary of approximately 1.600€) as well as excellent secondary benefits (holiday allowance, etc.). Belgium was ranked first on the “Best Countries for Academic Research” worldwide list (The Scientist, 2007), and Ghent University was appointed second place on the “Best Places to Work in Academia” non-US list (The Scientist, 2006).
Application and timing
If you are interested in one of the Ph.D. vacancies, please send in electronic format to Benjamin Schrauwen (Benjamin “dot” Schrauwen “at” UGent “dot” be): a detailed curriculum vitae, a motivation letter, your course program, your grades, two letters of recommendation and, if applicable, a publication list and selected publications. Do also mention your topics of preferences within the projects (e.g. robotics, speech, no preference, etc.). Some positions start on April 1, 2009, others in September 2009, meaning that persons who expect to graduate in July 2009 are welcome to apply. Applications which are received before February 1, 2009 get priority.
Project
Current state-of-the-art speech & handwriting recognition systems still perform much worse than human beings who can effortlessly decode the speech or handwriting of most people, even in fairly adverse conditions (e.g. the presence of noise in case of speech recognition). The fact that the human brain works so efficiently is owed to its self-organizing capacity, its deeply hierarchical approach, its adoption of unsupervised and supervised learning strategies, its capacity to adapt almost instantly to new circumstances, etc. Why not try to build an automatic speech recognizer and handwriting recognition engine that incorporates the same principles? This is exactly what we will do in two recently approved projects:
- “Self-organized Recurrent Neural Learning for Language Processing” (ORGANIC), funded by the European Commission within the 7th Framework Program. Details about the project can be found at the preliminary reservoir computing website (http://reservoir-computing.org).
- “Reservoir Computing for auditory pattern recognition” (RECAP), funded by the Research Program of the Research Foundation - Flanders (FWO).
The research concerns the investigation of architectures and algorithms for the efficient learning of large recurrent neural networks based on the Reservoir Computing concept (where only a linear readout layer is learned in a supervised way whereas the recurrent connections are fixed or trained in an unsupervised way). Important research topics are the unsupervised learning of a large hierarchy of recurrent sub-layers, and the integration of various adaptation techniques. The application domains are off-line handwriting recognition, speech recognition and various aspects of robotics (such as robot localization, motion control, ...). So far we were able to demonstrate that reservoirs can give rise to the robust recognition of digits spoken or written in isolation, but now we want to demonstrate that they can also yield robust recognition of continuous speech and handwriting (large vocabulary).
Requirements
Candidates should have a Masters degree in Electrical, Computer or Physics Engineering; or in Physics, Mathematics or Computer Science. A good knowledge of English is essential. No professional background is required, but the ideal candidates have some acquaintance with Machine Learning, programming (Python, Matlab, ...), statistics, signal processing, speech recognition, control engineering, or robotics.
What we offer
We offer an opportunity to perform at least three years of research in a new promising domain, and to get a doctoral degree in this domain. There will be ample opportunities for establishing international contacts (stays at partner universities, participation to international conferences). As an employee of the university you will receive a competitive salary (starting with a net monthly salary of approximately 1.600€) as well as excellent secondary benefits (holiday allowance, etc.). Belgium was ranked first on the “Best Countries for Academic Research” worldwide list (The Scientist, 2007), and Ghent University was appointed second place on the “Best Places to Work in Academia” non-US list (The Scientist, 2006).
Application and timing
If you are interested in one of the Ph.D. vacancies, please send in electronic format to Benjamin Schrauwen (Benjamin “dot” Schrauwen “at” UGent “dot” be): a detailed curriculum vitae, a motivation letter, your course program, your grades, two letters of recommendation and, if applicable, a publication list and selected publications. Do also mention your topics of preferences within the projects (e.g. robotics, speech, no preference, etc.). Some positions start on April 1, 2009, others in September 2009, meaning that persons who expect to graduate in July 2009 are welcome to apply. Applications which are received before February 1, 2009 get priority.
