eng
Schloss Dagstuhl – Leibniz-Zentrum für Informatik
Dagstuhl Seminar Proceedings
1862-4405
2010-11-05
10302
1
15
10.4230/DagSemProc.10302.1
article
10302 Abstracts Collection – Learning paradigms in dynamic environments
Hammer, Barbara
Hitzler, Pascal
Maass, Wolfgang
Toussaint, Marc
From 25.07. to 30.07.2010, the Dagstuhl Seminar 10302 ``Learning paradigms in dynamic environments '' was held in Schloss Dagstuhl~--~Leibniz Center for Informatics.
During the seminar, several participants presented their current
research, and ongoing work and open problems were discussed. Abstracts of
the presentations given during the seminar as well as abstracts of
seminar results and ideas are put together in this paper. The first section
describes the seminar topics and goals in general.
Links to extended abstracts or full papers are provided, if available.
https://drops.dagstuhl.de/storage/16dagstuhl-seminar-proceedings/dsp-vol10302/DagSemProc.10302.1/DagSemProc.10302.1.pdf
Recurrent neural networks
Dynamic systems
Speech processing
Neurobiology
Neural-symbolic integration
Autonomous learning
eng
Schloss Dagstuhl – Leibniz-Zentrum für Informatik
Dagstuhl Seminar Proceedings
1862-4405
2010-11-05
10302
1
4
10.4230/DagSemProc.10302.2
article
10302 Summary – Learning paradigms in dynamic environments
Hammer, Barbara
Hitzler, Pascal
Maass, Wolfgang
Toussaint, Marc
The seminar centered around problems which arise in the context of machine
learning in dynamic environments. Particular emphasis was put on a
couple of specific questions in this context: how to represent and abstract
knowledge appropriately to shape the problem of learning in a partially unknown
and complex environment and how to combine statistical inference
and abstract symbolic representations; how to infer from few data and how
to deal with non i.i.d. data, model revision and life-long learning; how to
come up with efficient strategies to control realistic environments for which
exploration is costly, the dimensionality is high and data are sparse; how to
deal with very large settings; and how to apply these models in challenging
application areas such as robotics, computer vision, or the web.
https://drops.dagstuhl.de/storage/16dagstuhl-seminar-proceedings/dsp-vol10302/DagSemProc.10302.2/DagSemProc.10302.2.pdf
Summary
eng
Schloss Dagstuhl – Leibniz-Zentrum für Informatik
Dagstuhl Seminar Proceedings
1862-4405
2010-11-05
10302
1
16
10.4230/DagSemProc.10302.3
article
Neurons and Symbols: A Manifesto
d'Avila Garcez, Artur S.
We discuss the purpose of neural-symbolic integration including its
principles, mechanisms and applications. We outline a cognitive computational model for neural-symbolic integration, position the model
in the broader context of multi-agent systems, machine learning and
automated reasoning, and list some of the challenges for the area of
neural-symbolic computation to achieve the promise of effective integration of robust learning and expressive reasoning under uncertainty.
https://drops.dagstuhl.de/storage/16dagstuhl-seminar-proceedings/dsp-vol10302/DagSemProc.10302.3/DagSemProc.10302.3.pdf
Neuro-symbolic systems
cognitive models
machine learning
eng
Schloss Dagstuhl – Leibniz-Zentrum für Informatik
Dagstuhl Seminar Proceedings
1862-4405
2010-11-05
10302
1
9
10.4230/DagSemProc.10302.4
article
One-shot Learning of Poisson Distributions in fast changing environments
Tino, Peter
In Bioinformatics, Audic and Claverie were among the first to systematically study the influence of random fluctuations and sampling size on the reliability of digital expression profile data.
For a transcript representing a small fraction of the library and a large number N of clones, the probability of observing x tags of the same gene will be well-approximated by the Poisson distribution parametrised by its mean (and variance) m>0,
where the unknown parameter m signifies the number of transcripts of the given type (tag) per N clones in the cDNA library.
On an abstract level, to determine whether a gene is differentially expressed or not, one has two numbers generated from two distinct Poisson distributions and based on this (extremely sparse) sample one has to decide whether the two Poisson distributions are identical or not. This can be used e.g. to determine equivalence of Poisson photon sources (up to time shift) in gravitational lensing.
Each Poisson distribution is represented by a single measurement only, which is, of course, from a purely statistical standpoint very problematic.
The key instrument of the Audic-Claverie approach is a distribution P over tag counts y in one library informed by the tag count x in the other library, under the null hypothesis that the tag counts are generated from the same but unknown Poisson distribution. P is obtained by Bayesian averaging (infinite mixture) of all possible Poisson distributions with mixing proportions equal to the posteriors (given x) under the flat prior over m.
We ask: Given that the tag count samples from SAGE libraries are *extremely* limited, how useful actually is the Audic-Claverie methodology? We rigorously analyse the A-C statistic P that forms a backbone of the methodology and represents our knowledge of the underlying tag generating process based on one observation.
We show will that the A-C statistic P and the underlying Poisson distribution of the tag counts share the same mode structure. Moreover, the K-L divergence from the true unknown Poisson distribution to the A-C statistic is minimised when the A-C statistic is conditioned on the mode of the Poisson distribution. Most importantly (and perhaps rather surprisingly), the expectation of this K-L divergence never exceeds 1/2 bit! This constitutes a rigorous quantitative argument, extending the previous empirical Monte Carlo studies, that supports the wide spread use of Audic-Claverie method, even though by their very nature, the SAGE libraries represent very sparse samples.
https://drops.dagstuhl.de/storage/16dagstuhl-seminar-proceedings/dsp-vol10302/DagSemProc.10302.4/DagSemProc.10302.4.pdf
Audic-Claverie statistic
Bayesian averaging
information theory
one-shot learning
Poisson distribution
eng
Schloss Dagstuhl – Leibniz-Zentrum für Informatik
Dagstuhl Seminar Proceedings
1862-4405
2010-11-05
10302
1
15
10.4230/DagSemProc.10302.5
article
Some steps towards a general principle for dimensionality reduction mappings
Hammer, Barbara
Bunte, Kerstin
Biehl, Michael
In the past years, many dimensionality reduction methods have been
established which allow to visualize high dimensional data sets. Recently,
also formal evaluation schemes have been proposed for data visualization,
which allow a quantitative evaluation along general principles. Most techniques
provide a mapping of a priorly given finite set of points only, requiring
additional steps for out-of-sample extensions. We propose a general
view on dimensionality reduction based on the concept of cost functions,
and, based on this general principle, extend dimensionality reduction to
explicit mappings of the data manifold. This offers the possibility of simple
out-of-sample extensions. Further, it opens a way towards a theory
of data visualization taking the perspective of its generalization ability
to new data points. We demonstrate the approach based in a simple
example.
https://drops.dagstuhl.de/storage/16dagstuhl-seminar-proceedings/dsp-vol10302/DagSemProc.10302.5/DagSemProc.10302.5.pdf
Visualization
dimensionality reduction
eng
Schloss Dagstuhl – Leibniz-Zentrum für Informatik
Dagstuhl Seminar Proceedings
1862-4405
2010-11-05
10302
1
2
10.4230/DagSemProc.10302.6
article
Why deterministic logic is hard to learn but Statistical Relational Learning works
Toussaint, Marc
A brief note on why we think that the statistical relational learning framework is a great advancement over deterministic logic – in particular in the context of model-based Reinforcement Learning.
https://drops.dagstuhl.de/storage/16dagstuhl-seminar-proceedings/dsp-vol10302/DagSemProc.10302.6/DagSemProc.10302.6.pdf
Statistical relational learning
relational model-based Reinforcement Learning