Seminar Computational Intelligence B (708.112)

Content of the seminar:

In addition talks can be presented on the following material, that is closely related to research projects at our Institute. Some of these topics could also serve as first step towards a Master project or -thesis.

The remaining slots for talks (two 40-minute talks at each seminar meeting) will be assigned at the Organization meeting on march 2.

If you are particularly interested in a particular one of these topics (or want to propose an additional one) you can also send anytime email to Wolfgang Maass.

------------------------------------------------------------------

Probabilistic Inference in Graphical Models:

Selected sections from the book:

Koller-Friedman, Probabilistic Graphical Models MIT-Press, 2009,

chosen with regard to their possible relevance for our research:
(I would like to suggest, that most talks in this seminar should cover material from this book, and/or alternative sources for the same material)


--parts of ch. 13 on MAP inference

--ch. 15: Inference in Temporal Models

--pp 741-754: Bayesian parameter estimation in Bayesian Networks

--parts of ch. 18: Structure learning in Bayesian networks

--ch. 19: Partially Observed Data

--ch. 20: Learning Undirected Models

--ch. 23: Structured Decision Problems


--------------------------------------------------------------------------------------------------

Date of presentations:

16.03.11:    Gerhard Neumann, Slides,

06.04.11:    Dejan Pecevski

13.04.11:    Stefan Klampfl, Slides, Bernhard Nessler,Slides

11.05.11:    Johannes Bill, Slides, David Kappel, Slides, 

20.05.11:    12.00 - 14.00h - Brown-Bag-Seminar
                   Stefan H�usler, Slides, Stefan Habenschuss, Slides,

25.05.11:    Jing Fang, Slides,

01.06.11:    Sabine Schneider, Slides, Tim Genewein, Slides,

15.06.11:    Patrick Ofner, Gernot Griesbacher,Slides, Zeno Jonke, Slides,

--------------------------------------------------------------------------------------------------                       

Applications of Probabilistic Inference in Motor Control for Robots etc

--Tobias Lang, Marc Toussaint (2010): Probabilistic backward and forward reasoning in stochastic relational worlds. 26th International Conference on Machine Learning (ICML 2010).


--Marc Toussaint: Approximate inference control  (preprint),
[outlines a promising approach for solving motor control tasks via Bayesian networks, shows that stochastic optimal control is a special case of this approach, an MA-thesis in this direction could complement already ongoing work in this direction at IGI]

--Tobias Lang, Marc Toussaint (2010): Planning with Noisy Probabilistic Relational Rules. Journal of Artificial Intelligence Research, 39, 1-49.
http://www.jair.org/media/3093/live-3093-5172-jair.pdf
[very interesting new approach involving both probabilistic inference and relational rules; an MA-thesis in this direction could complement already ongoing work in this direction at IGI ]

--SJ Gershman, Y Niv: Learning latent structure: carving nature at its joints, Current Opinion in Neurobiology, 2010 http://www.sciencedirect.com/science/article/pii/S0959438810000309
[This paper provides very important new ideas for robot learning etc, exploiting learning simultaneously on several layers of abstraction, on which we will work at our Institute; the paper itself review primarily related evidence from biological motor control. In conjunction with some cited papers this review paper would also provide material for more than 1 talk. This paper would provide a good basis for a MA-thesis on the development of methods for fast robot learning.]

---------------------------------------------------------------------------------------------------

Computational Neuroscience

--Litvak, Ullman, Cortical circuitry implementing graphical models von.  Neural Computation, 2009.  https://www.mitpressjournals.org/doi/abs/10.1162/neco.2009.05-08-783


--S. Shinomoto et al, Relating neuronal firing patterns to functional differentiation of cerebral cortex, PLOS Computational Biology 2009
http://www.jneurosci.org/cgi/reprint/29/10/3233

--W. Li, V. Piech, and C. D Gilbert. Perceptual learning and top-down influences in primary visual cortex. Nature Neuroscience, 7(6):651-657, 2004.http://www1.nin.knaw.nl/viscog/temp/learning%20v1%20Gilbert.pdf http://www1.nin.knaw.nl/viscog/temp/learning v1 Gilbert.pdf
[This paper provides the main background for a potential MA-thesis (supervised by Dr. Legenstein). MA thesis short description: "Self-organization of task-dependent computation in sensory circuits". Results from Li, Piech, and Gilbert (2004) show that neurons in primary visual cortex can drastically and rapidly change their feature preferences in dependence of the currently performed task. This ability suggests that sensory cortical areas act as adaptive processors, performing different calculations according to the immediate perceptual demands (Gilbert et al. 2009). Such complex neuronal behavior is probably a result of local neural circuits that exploit nonlinear dendritic properties of single neurons as well as feedback from other cortical areas. This MA thesis will explore how feedback indicating the current task can influence the computation of local circuits of spiking neurons and how the computation can be self-organized through synaptic plasticity mechanisms.
Further readings:  C. D. Gilbert, W. Li, and V. Piech. Perceptual learning and adult cortical plasticity . The Journal of Physiology, pages
2743-2751, 2009. (review)
M. Sigman and C. D. Gilbert. Learning to find a shape. Nature Neurosci., 2000.]


--R. Pascanu and H. Jaeger, A Neurodynamical Model for Working Memory
Preprint, 2010

--A Mazzoni, S Panzeri, N Logothetis, N Brunel (2008) Encoding of Naturalistic Stimuli by Local Field Potential Spectra in Networks of Excitatory and Inhibitory Neurons, PLOS Comp. Biol., http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1000239

--A. Mazzoni, K. Whittingstall, N. Brunel, N. K. Logothetis, S. Panzeri
Understanding the relationships between spike rate and delta/gamma frequency bands of LFPs and EEGs using a local cortical network model
NeuroImage 52, 956-972, 2010

-------------------------------------------------------------------------------

Applications of Probabilistic Inference in Cognitive Science (with the prospect of porting similar capabilities into machines)

 Theory acquisition as stochastic search. T. D. Ullman, N. D. Goodman and J. B. Tenenbaum (2010). Proceedings of the Thirty-Second Annual Conference of the Cognitive Science Society http://web.mit.edu/tomeru/www/papers/tlss2010.pdf

Kemp, C., Goodman, N. & Tenenbaum, J. (2010). Learning to learn causal models. Cognitive Science, 34(7),1185-1243 http://www.psy.cmu.edu/~ckemp/papers/kempgt10_learningtolearncausalmodels.pdf

Theory acquisition and the language of thought. C. Kemp, N. D. Goodman, and J. B. Tenenbaum (2008). Proceedings of the Thirtieth Annual Conference of the Cognitive Science Society. http://repository.cmu.edu/psychology/967/

Learning Structured Generative Concepts. A. Stuhlmueller, J. B. Tenenbaum, and N. D. Goodman (2010). Proceedings of the Thirty-Second Annual Conference of the Cognitive Science Society.http://www.mit.edu/~ast/papers/structured-generative-concepts-cogsci2010.pdf
Shi, L., Feldman, N. H., & Griffiths, T. L. (2008).

Performing Bayesian inference with exemplar models. Proceedings of the 30th Annual Conference of the Cognitive Science Society http://cocosci.berkeley.edu/tom/papers/exemplar.pdf

(more details are in  Shi, L., Griffiths, T. L., Feldman, N. H, & Sanborn, A. N. (in press). Exemplar models as a mechanism for performing Bayesian inference. Psychonomic Bulletin & Review. http://cocosci.berkeley.edu/tom/papers/mechanism.pdf )