Veranstaltungen

Intelligent Systems and Applications

SGAICO

18.11.2014 10:00 - 18.11.2014 18:00
Universität Basel
Petersplatz 1, Kollegienhaus, Fakultätenzimmer 112 - 4001 Basel Schweiz

Annual assembly and workshop of the SGAICO group. Registration until Nov, 14 2014.

 

Universität Basel, Kollegienhaus, Fakultätenzimmer 112, Petersplatz 1, 4001 Basel

 

www.unibas.ch/unibas_lage/plan.cfm

The annual assembly and workshop of SGAICO will take place on November, 18 2014 SGAICO at the University of Basel and feature the topic Intelligent Systems and Applications.


To register for participation, please fill in the registration form at

http://www.networkinglabor.ch/sgaico-2014/ UNTIL FRIDAY NOV, 14 2014.

 

 

AGENDA

 

9:30Registration

10:00 - 10:30

General Assembly & Election of the new Steering Board and President 

 

Candidates for the new Steering Board:

Jean-Daniel Dessimoz (HESSO) (for deputy president)

Marc-Oliver Gewaltig (EPFL) (for deputy president)

Jana Koehler (HSLU) (for president)

Thilo Stadelmann (ZHAW) (for deputy president)

 

10:30 - 11:45

Dialogue on Education (Slides)

10:45

15'

Overview AI/CS Education in Switzerland - results from poll (Thilo Stadelmann)

11:00

10'

Course overview on "Automatisation avancée, intelligence artificielle et cognitique" (Jean-Daniel Dessimoz, HESSO; Slides)

11:15

10'

Course overview on "DAS Data Science" (Thilo Stadelmann, ZHAW)

 

11:30

15'

General discussion on next steps and the idea of a winter school (all)

 

11:45

        

End

 

11:45 - 12:30

Invited Keynote (Slides)

Deep Learning and the neural net RNNaisense (Jonathan Masci, IDSIA)

In the past few years AI experienced a rapid progress, far beyond expectations, given previous trends. This is mainly due to a neural network renaissance called Deep Learning.  Thanks to availability of large amounts of labeled training data, extremely powerful general purpose machines (i.e. GPU enabled), and a few small modifications to neural architectures devised more than two decades ago, we are now able to efficiently train networks with millions of weights, and up to 20 layers with plain back-propagation.
After introducing the foundations of deep learning I will present the convolutional neural net (convnet), a particularly powerful model for computer vision tasks, which is largely responsible for many of the recent, high impact successes of deep learning in object recognition and detection.

12:30 - 14:00Lunch
14:00 - 18:00

Technical Program & Workshop

 

14:00 - 14:25 LP-based Heuristics for Cost-optimal Planning (Florian Pommerening, Uni Basel; Slides)

Heuristics in cost-optimal planning estimate the cost to reach a goal. The calculation of many heuristics can be written declaratively as a linear program. We cover several interesting heuristics of this type by a common framework that fixes the objective function of the linear program. Within the framework, constraints from different heuristics can be combined in one heuristic estimate which dominates the maximum of the component heuristics. Different heuristics of the framework can be compared on the basis of their constraints. This enables us to better understand the relationships between different heuristics.


14:25 - 14:50 Directed model checking: Finding bugs through AI planning (Martin Wehrle, Uni Basel; Slides)

Model checking is an automated approach to check whether a model of a system satisfies a given property. An important practical aspect of model checking is bug finding. In particular, this is the case for (safety critical) concurrent systems, where subtle bugs can occur because of unexpected thread interleavings.Directed model checking is a version of model checking that applies AI planning techniques like heuristic search to find bugs in concurrent systems. This talk provides a general introduction to directed model checking, and presents a search technique that has been recently applied to find bugs in concurrent systems of timed automata.

 

14:50 - 15:15 Can Deep Learning solve the Sentiment Analysis Problem? (Mark Cieliebak, ZHAW; Slides)

Sentiment analysis appears to be one of the easier tasks in the realm of text analytics: given a text like a tweet or product review, decide whether it contains positive or negative opinion. This task is almost trivial for humans, but it turns out to be a true challenge for automated systems, in fact, state-of-the-art sentiment analysis tools are wrong on approx. 4 out of 10 documents. Current sentiment analysis tools are rule-based, feature-based, or combinations of both. However, recent research uses deep learning on very large sets of documents. In this talk, we will explain the intrinsic difficulties of automated sentiment analysis; present existing solution approaches and their performance; describe an architecture for a deep learning system; and explore whether deep learning can improve sentiment analysis accuracy.
Link to the word2vec demo mentioned in the talk: Word2vec Tutorial (towards the bottom of the page).

 

15:15  - 15:40 Cognition to perceive, explore and model the world ( Jean-Daniel Dessimoz, HESSO; Slides)

The presentation will develop the first of five theses recently published in the context of new research frontiers for intelligent autonomous systems: Cognition is necessary to perceive, explore and model the world. Therefore, first, a better understanding is required and, next, making the underlying capability automated will boost its deployment. Modeling is, more or less dynamically, at the core of cognitive processes, i. e. of cognition. Let us first see whether innate or acquired resources are at hand. Then modeling is addressed again. Finally, it is shown how models can be updated and refined, as a result of exploration, perception and experience.


15:40 - 16:00 Break

 

16:00 - 16:45 Invited Talk IBM Watson - Technical Deep Dive (Romeo Kienzler, IBM; Slides)

We are transitioning from the programmatic to the cognitive computing era.IBM Deep Blue won against the world champion in Chess 1996. IBM Watson won against the two world champions in the famous US quiz show "Jeopardy" 2011. Since then, the press heavily established the term "Cognitive Computing" to the public. I will explain how IBM Watson works internally and start with Algebraic Text Extraction.  DeepQA is the heart of IBM Watson and I will explain each component of this pipeline, the linguistic preprocessor, hypothesis generation, hypothesis and evidence scoring, final matching based on supervised learning and confidence estimation. Finally, I conclude with an overview of actual use cases and outline the roadmap of future work.

 

16:45 - 17:15 Detection and Quantification of Hand Eczema by Visible Spectrum Skin Pattern Analysis (Marc Pouly, HSLU; Slides)

Hand and whole-body eczema are frequent dermatoses with severe health and financial consequences to patients and society. They follow a chronic course and persist up to 15 years after onset. A new generation of highly effective drugs has recently been pushed onto the market, but due to the exorbitant costs of this new treatment, health insurances cover expenses only in severe cases. Dermatologists have therefore been developing different scoring systems to objectively measure and document eczema severeness. However, assessing the parameters for these scores in practice is difficult as it for example requires to estimate the percentage of body surface with eczema infestation. In this talk we are going to present a prototype-based feasibility study of automated detection and quantification of hand eczema using texton-based imaging and machine-learning techniques; a multi-disciplinary research project between the university hospitals of Zurich, Basel and the Lucerne university of applied sciences and arts.

 

17:15 - 17:40 Complex Spatial Models and Constraint Solving in Ultra-Large Search Spaces (Jana Koehler, HSLU)

Cable-tree wiring is a complex assembly problem commonly involving manual activity of skilled humans. When trying to automate this problem, one major challenge results from modeling the behavior of cables and cable wiring machines. The other major challenge results from the need to solve constraint problems in ultra-large search spaces of 10^120 states. The talk discusses results and summarizes lessons learned from a project that aims at developing automated control software for commercial cable-tree wiring machines.

 

17:40 - 18:00 Closing and Outlook on next events

 

To cover the costs of the event, a nominal participation fee of 20 CHF will be collected per participant. Lunch vouchers will be available for 15 CHF.
 

 

Zuständig:
Jana Koehler
044 349 3350 - jana.koehler(at)hslu.ch
Kosten: 20 CHF (mit Mittagessen 35 CHF)
Zurück zur Liste