---------- Forwarded message ----------
From: Nicolas Courty ncourty@gmail.com>
Date: Fri, Apr 12, 2013 at 12:38 PM
Subject: [ML-news] PhD position available in Vannes/Rennes (France)
To: ml-news@googlegroups.com
Dear all,
Please find enclosed the description of a PhD position available in Vannes/Rennes (France).
Feel free to circulate the call to your colleagues / students that may be interested !
Sincerely,
Nicolas Courty
--
The OBELIX Team is a team in creation from Irisa (http://www.irisa.fr/). The team is dedicated to environment observation problems, implying advanced
image processing techniques and machine learning. The team is co-located between Rennes and Vannes, two beautiful cities from Brittany, France.
We are seeking highly skilled PhD students with a strong interest in machine learning and image processing.
Deadline for application: 3 May 2013
*PhD Position at IRISA- Obelix Team*
*Topic:* Optimization of manifold learning techniques for large quantities of data
*Advisor:* Nicolas Courty / Thomas Corpetti (http://www-irisa.univ-ubs.fr/Nicolas.Courty/ , http://www.sites.univ-rennes2.fr/costel/corpetti/site/)
*Duration:* 3 years, starting end of 2013
*How to apply:* Please send a resume + motivation letter to ncourty@gmail.com + tcorpetti@gmail.com
Observation is one of the key issues in the understanding of environmental systems. A large amount of possibilities, ranging from local probes or networks to hyper-spectral remote sensing images, is at the moment available to sense and extract environmental parameters. The huge quantities of data collected by the new generation of captors that operate at fine temporal and spatial scales calls for new dedicated machine learning methods that can handle this complexity. One of the main research axis of the OBELIX team is to build new tools that allow to handle both theoretically and computationally this complexity. The underlying techniques refer to scale-space techniques and manifold learning for the theoretical part, and to massive computing using GPUs networks or cloud computing for the operational level. The applications of these works are concerned with classification or visualization.
More specifically, this PhD Thesis is concerned with the domain of manifold learning, where one seeks a sub-dimensional space which best characterizes the data, while offering a low-dimensional representation. Among a wide panel of methods, we have developed specifically a tool which is named PerTurbo. It is inspired by recent advances in computer graphics: Each class is characterized as a manifold in the input space (in a manner similar to the cloud of points giving birth to the 3D surface of a virtual object) thanks to an approximation of the Laplace-Beltrami operator. As this approximation happens to be the Gaussian kernel, its perturbation measure (when a test sample is added to the manifold) can be re-interpreted in the kernel machine learning setting. However, when the dimension of the learning sets is too large, computational problems arise, mostly because the algorithm complexity is $o(n^3)$.
Possible solutions to solve this problem imply that only a subset of the input learning set is kept: this problematic is then a coarse-graining problem. In order to build this coarse scale representation of the information, the subsampling may try to preserve to the maximum some features of the manifold. In the thesis, several aspects of this construction will be examined. Also, different types of methods will be tested mostly implying either sparse analysis but also random projections. The student will also try to adapt the corresponding solutions to the specificities of the considered data (spatiality and temporality), mostly related to environmental observations.
--
You received this message because you are subscribed to the Google Groups "Machine Learning News" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ml-news+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
From: Nicolas Courty ncourty@gmail.com>
Date: Fri, Apr 12, 2013 at 12:38 PM
Subject: [ML-news] PhD position available in Vannes/Rennes (France)
To: ml-news@googlegroups.com
Dear all,
Please find enclosed the description of a PhD position available in Vannes/Rennes (France).
Feel free to circulate the call to your colleagues / students that may be interested !
Sincerely,
Nicolas Courty
--
The OBELIX Team is a team in creation from Irisa (http://www.irisa.fr/). The team is dedicated to environment observation problems, implying advanced
image processing techniques and machine learning. The team is co-located between Rennes and Vannes, two beautiful cities from Brittany, France.
We are seeking highly skilled PhD students with a strong interest in machine learning and image processing.
Deadline for application: 3 May 2013
*PhD Position at IRISA- Obelix Team*
*Topic:* Optimization of manifold learning techniques for large quantities of data
*Advisor:* Nicolas Courty / Thomas Corpetti (http://www-irisa.univ-ubs.fr/Nicolas.Courty/ , http://www.sites.univ-rennes2.fr/costel/corpetti/site/)
*Duration:* 3 years, starting end of 2013
*How to apply:* Please send a resume + motivation letter to ncourty@gmail.com + tcorpetti@gmail.com
Observation is one of the key issues in the understanding of environmental systems. A large amount of possibilities, ranging from local probes or networks to hyper-spectral remote sensing images, is at the moment available to sense and extract environmental parameters. The huge quantities of data collected by the new generation of captors that operate at fine temporal and spatial scales calls for new dedicated machine learning methods that can handle this complexity. One of the main research axis of the OBELIX team is to build new tools that allow to handle both theoretically and computationally this complexity. The underlying techniques refer to scale-space techniques and manifold learning for the theoretical part, and to massive computing using GPUs networks or cloud computing for the operational level. The applications of these works are concerned with classification or visualization.
More specifically, this PhD Thesis is concerned with the domain of manifold learning, where one seeks a sub-dimensional space which best characterizes the data, while offering a low-dimensional representation. Among a wide panel of methods, we have developed specifically a tool which is named PerTurbo. It is inspired by recent advances in computer graphics: Each class is characterized as a manifold in the input space (in a manner similar to the cloud of points giving birth to the 3D surface of a virtual object) thanks to an approximation of the Laplace-Beltrami operator. As this approximation happens to be the Gaussian kernel, its perturbation measure (when a test sample is added to the manifold) can be re-interpreted in the kernel machine learning setting. However, when the dimension of the learning sets is too large, computational problems arise, mostly because the algorithm complexity is $o(n^3)$.
Possible solutions to solve this problem imply that only a subset of the input learning set is kept: this problematic is then a coarse-graining problem. In order to build this coarse scale representation of the information, the subsampling may try to preserve to the maximum some features of the manifold. In the thesis, several aspects of this construction will be examined. Also, different types of methods will be tested mostly implying either sparse analysis but also random projections. The student will also try to adapt the corresponding solutions to the specificities of the considered data (spatiality and temporality), mostly related to environmental observations.
--
You received this message because you are subscribed to the Google Groups "Machine Learning News" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ml-news+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
Ali Mustafa
__._,_.___
| Reply via web post | Reply to sender | Reply to group | Start a New Topic | Messages in this topic (1) |
MARKETPLACE
.
__,_._,___
No comments:
Post a Comment