Seminar Computational Intelligence B (708.112)

WS 2012

Institut für Grundlagen der Informationsverarbeitung (708)

o.Univ.-Prof.Dr.Wolfgang Maass

E-mail: maass@igi.tugraz.at

Assoc. Prof. Dr. Robert Legenstein

E-mail: robert.legenstein@igi.tugraz.at

Objectives of the seminar:

This seminar will introduce into a few research areas related to Computational Intelligence.
It is designed for master students in their second year (possibly also in the first year, if they take simultaneously NN A or ML B).
The papers and talks will be on an introductory level, and will be accessible to an audience with minimal technical background.
Each of the talks is related to possible projects and thesis topics for a master, and could serve as preparation for that.

One of the courses NN A, NN B, ML A, ML B suffices.

Location:    IGI-seminar room, Inffeldgasse 16b/I, 8010 Graz

Date:            Preparatory Meeting on Monday Oct 8th 2012, 16.15 p.m. - 18.00 p.m.

                    03.12.2012    Grabuschnig Stefan, slides
                    10.12.2012    Haas Gabriel, slides
                    17.12.2012    entfällt
                    07.01.2013    Hell Timotheus, slides
                    14.01.2013    Lamprecht Daniel, slides
                                          Hubner Florian, slides
                    21.01.2013    Ofner Patrick, slides
                                    Klatzer Teresa, slides
                    28.01.2013    Rath Michael, slides

Content of the seminar
Several topics will be discussed:

Topics related to the AMARSI Project (Robotics)

AMARSI is an EU project that aims at a qualitative jump in robotic motor skills towards biological richness.
It examines to what extent movement control of robots can be learnt. Another innovation of the project is that it
investigates a new generation of compliant robot. A first prototype of a compliant robot developed in this project
will be available for experiments at IGI. The IGI is a partner in this project.

Topics and papers:

Hopfield and Tank: "Neural" computation of decisions in optimization problems.
Waegeman, Wyffles, Schrauwen: Towards a neural hierarchy of time scales for motor control, 2012.
McDannald et al: Model-based learning and the contribution of the orbitofrontal cortex to the model-free world, 2012.
Gläscher et al: States versus Rewards: Dissociable Neural Prediction Error Signals Underlying Model-Based and Model-Free Reinforcement Learning, 2010.
Hoerzer, Legenstein, Maass: Emergence of complex computational structures from chaotic neural networks through reward-modulated Hebbian learning, 2012.

Topics related to the Blue-Brain Project

The Blue-Brain Project (BBP) is a large research project situated at EPFL lead by Prof. Henry Markram.
The goal of the project is to use supercomputers to simulate one or several cortical columns of mice.
The IGI will collaborate with Prof. Markram on research related to the BBP.

A new type of EU research project, a socalled flagship project, is currently submitted as an extension of the BPP:
www.humanbrainproject.eu.

Topics and papers:

The paper The Blue Brain Project provides an introduction into the goals and strategies of the BBP.
The report The Human Brain Project discusses the planned flagship project (a subset of the content will be chosen for presentation).
Hill et al. 2012 discusses results about brain connectivity discovered within the BBP.
The paper Probst et al. Liquid Computing in a Simplified Model of Cortical Layer IV: Learning to Balance a Ball, 2012
shows how a simplified model of a cortical microcircuit can be used for a control task.
The computational capabilities of simulated cortical columns can be analyzed through methods developed at our institute:
Häusler and Maass 2007 or Legenstein and Maass 2007.

Topics related to the PNEUMA Project (Memristors)

PNEUMA is an international project that aims to utilize memristors - a novel electronic circuit element that is basically a resistor
with memory - for brain-inspired computation. The IGI is a partner in this project.
New hardware technologies related to memristive elements on the nanoscale have motivated major companies like IBM and HP
to investigate new computing architectures related to neural circuits (see e.g. the recent workshop at the IBM Research Center in Zürich).
In the seminar, we will discuss the basics of memristors.

Topics and papers:

GK Johnsen - Journal of Electrical Bioimpedance, 2012 provides an introduction to memristors.
L Chua - Applied Physics A: Materials Science & Processing, 2011 provides a tutorial from the theoretical side.
B Linares-Barranco, T Serrano-Gotarredona - Nature precedings, 2009 shows how synaptic plasticity rules
can be implemented through memristors.

2-photon calcium imaging data

2-photon Calcium imaging (2p-imaging) is a novel experimental technique that makes it possible to image the activity of hundreds
to thousands of neurons in the living brain. We collaborate with several researchers to study suitable analysis methods
for this novel high-dimensional data.

Topics and papers:

Huber et al, Multiple dynamic representations in the motor cortex during sensorimotor learning. Nature, 2012 presents results
about changes in neural activity during learning, recorded through 2p Calcium imaging.
Harvey et al., Choice-specific sequences in parietal cortex during a virtual-navigation decision task. Nature, 2012 presents nice data
about neural trajectories during a decision task, recorded through 2p Calcium imaging.

Stochastic synaptic transmission

Synapses transmit information between neurons in a stochastic manner. An interesting research question is how neuronal networks
can utilize this stochasticity for computation and learning.

Topics and papers:

Faisal et al., Nature Rev. Neurosci. 2008 reviews the sources of noise in the nervous system.
More papers...

Further topics may include

Retrograde plasticity signals
Clever exploration and creativity (intrinsic motivation)