|
Size: 4581
Comment:
|
Size: 4591
Comment:
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 14: | Line 14: |
---- == Planned Talks == This is a preliminary schedule, and might be changed on short notice. === Wednesday, 04.11.09 14.15 FR6046 === ==== Talk invited speaker ==== '''Aapo Hyvarinen:''' ''Unsupervised learning by discriminating data from artificial noise'' Abstract: We propose to learn the structure of data based on the simple idea of learning to discriminate data from artificially generated noise. Such learning is basically unsupervised, although we formulate it as logistic regression. The method can be shown to estimate a parametric probabilistic model for the data. Furthermore, the probabilistic model does not need to be normalized (i.e. it can be energy-based) because the normalization parameter can actually be estimated just like the other parameters. We apply the method to learning two- and three-layer networks from natural images, where the noise is white and gaussian. (Chair:Klaus-Robert Müller) === Wednesday, 11.11.09 14.15 FR6046 === ==== Research Ideas ==== * '''Daniel Bartz''', * '''Felix Bießmann''', * '''Tobias Lang''' (Chair:Nils Plath) |
|
| Line 53: | Line 22: |
---- == Planned Talks == This is a preliminary schedule, and might be changed on short notice. |
|
| Line 129: | Line 107: |
| === Wednesday, 11.11.09 14.15 FR6046 === ==== Research Ideas ==== * '''Daniel Bartz''', * '''Felix Bießmann''', * '''Tobias Lang''' (Chair:Nils Plath) === Wednesday, 04.11.09 14.15 FR6046 === ==== Talk invited speaker ==== '''Aapo Hyvarinen:''' ''Unsupervised learning by discriminating data from artificial noise'' Abstract: We propose to learn the structure of data based on the simple idea of learning to discriminate data from artificially generated noise. Such learning is basically unsupervised, although we formulate it as logistic regression. The method can be shown to estimate a parametric probabilistic model for the data. Furthermore, the probabilistic model does not need to be normalized (i.e. it can be energy-based) because the normalization parameter can actually be estimated just like the other parameters. We apply the method to learning two- and three-layer networks from natural images, where the noise is white and gaussian. (Chair:Klaus-Robert Müller) |
Doktoranden-/Diplomandenseminar WS 09/10
In this seminar, master/phd students of the IDA group present their current work and research ideas.
When? |
usually on Wednesdays, 2:15 pm (s.t.) to 3:15, or as announced (see below) |
Where? |
at TU Berlin , FR 6046, or as announced (see below) |
How? |
More information for speakers and participants here. |
Next Talk
Wednesday, 18.11.09 14.15 FR6046
Research Ideas
Paul von Bünau,
Katja Hansen,
Tammo Krüger
(Chair:Pascal Lehwark)
Planned Talks
This is a preliminary schedule, and might be changed on short notice.
Wednesday, 25.11.09 14.15 FR6046
Research Ideas
Alexander Binder,
Frank Meinecke,
Claudia Sannelli
(Chair:Felix Bießmann)
Wednesday, 02.12.09 14.15 FR6046
Talk invited speaker
Adam Stanski,
(Chair: Klaus-Robert Müller)
07.-12.12.09 NIPS
Wednesday, 16.12.09 14.15 FR6046
NIPS Conference Report
Wednesday, 21.12.09 - 02.01.2010 no lectures
Wednesday, 6.1.10 14.15 FR6046
Research Ideas
Matthias Jugel,
Christian Gehl,
Nils Plath
(Chair:Claudia Sanelli)
Wednesday, 13.1.10 14.15 FR6046
Research Ideas
Siamac Fazli,
Cecilia Maeder,
Irene Sturm
(Chair:Alexander Binder)
Wednesday, 20.1.10 14.15 FR6046
Research Ideas
Bastian Venthur,
Stanimir Dragiev,
Grégoire Montavon
(Chair:Matthias Jugel)
Wednesday, 27.1.10 14.15 FR6046
Research Ideas
Anne Porbadnigk,
Marko Ristin,
n.n.
(Chair:Marton Danoczy)
Wednesday, 03.2.10 14.15 FR6046
Research Ideas
Marton Danoczy,
Patrick Düssel,
Pascal Lehwark
(Chair: Christian Gehl)
Previous Talks
Wednesday, 11.11.09 14.15 FR6046
Research Ideas
Daniel Bartz,
Felix Bießmann,
Tobias Lang
(Chair:Nils Plath)
Wednesday, 04.11.09 14.15 FR6046
Talk invited speaker
Aapo Hyvarinen:
Unsupervised learning by discriminating data from artificial noise
Abstract: We propose to learn the structure of data based on the simple idea of learning to discriminate data from artificially generated noise. Such learning is basically unsupervised, although we formulate it as logistic regression. The method can be shown to estimate a parametric probabilistic model for the data. Furthermore, the probabilistic model does not need to be normalized (i.e. it can be energy-based) because the normalization parameter can actually be estimated just like the other parameters. We apply the method to learning two- and three-layer networks from natural images, where the noise is white and gaussian.
(Chair:Klaus-Robert Müller)
Wednesday, 28.10.09 14.15 FR6046
Talk
Timon Schroeter
(Chair: Katja Hansen)
Wednesday, 21.10.09 14.15 FR6046
Research Ideas
Nikolay Jetchev: Trajectory Prediction and Parallel Exploration of Movement Trajectories for Online Planning,
Martijn Schreuder: Performance feedback during (auditory) P300 operation,
nn
(Chair:Daniel Bartz)
Wednesday, 14.10.09 14.15 FR6046
Tutorial
Matthias Treder
Experimental Design in Human Research
Abstract:This tutorial is devoted the first steps in human experimental research, that is, defining a research question and then translating this research question into an experimental design, and finally setting up the experiment. According concepts and terminology from psychological methodology are introduced, such as types of research, experimental designs, the definition of measurements, and the assessment of their quality. Possible remedies are worked out in a hopefully interactive fashion.
Archive
-- Nicole Kraemer / Katja Hansen -- Aug 2009