Content: Ninth Workshop

Ninth Workshop


The ninth Hybris Workshop will take place on June 12th-13th 2017 at the Leipzig University, Germany.

Venue

The workshop will be held at Leipzig University, Germany, Felix-Klein-Hörsaal, Paulinum, 5th floor, Augustusplatz 10.

Program

TimeTopicSpeaker
Monday, June 12th
12:45-13:00Registration and Welcome
13:00-13:30ASP with Linear Constraints and its ApplicationsMax Ostrowski
University of Potsdam
13:30-14:00Generalized Answer Set Planning with Incomplete InformationJavier Romero
University of Potsdam
14:00-14:30plasp 3: Towards Effective ASP PlanningMartin Gebser
University of Potsdam
14:30-15:00A Mobile App for ArgumentationJörg Pührer
University of Leipzig
15:00-15:30Coffee Break
15:30-16:30Invited Talk: Decomposition & Learning: Recent Advances in Classical Planning
Icon slides (3.0 MB)
Joerg Hoffmann
Saarland University
16:30-17:00Using Ontologies to Query Probabilistic Numerical DataPatrick Koopmann
TU Dresden
17:15-18:15Hybris PI Meeting
Tuesday, June 13th
9:00-10:00Epistemic General Problem Solving
Icon slides (2.2 MB)
Michael Thielscher
University of New South Wales
10:00-10:30Coffee Break
10:30-11:00Decidable Verification of Decision-Theoretic GologJens Claßen
RWTH Aachen University
11:00-11:30A First-Order Logic of Limited BeliefChristoph Schwering
University of New South Wales
11:30-12:00Learning Actions and Tasks from a few DemonstrationsTim Welschehold
University of Freiburg
12:00-13:30Lunch
13:30-15:30Invited Talk: A Short Course in Belief Revision
Icon slides (607.7 KB)
Pavlos Peppas
University of Patras
15:30Coffee and Farewell

Abstracts

Decomposition & Learning: Recent Advances in Classical Planning

Invited talk by Joerg Hoffmann, Saarland University

Abstract:Classical Planning is a combinatorial search problem, capturing sequential decision-making in its most basic form where the solution sought is a path in a large state space graph. Decomposition and learning methods are general algorithmic ideas that have been successful in many combinatorial search problems, including but not limited to classical planning. The talk covers three recent methods developed in the speaker's research group, each of which has potential for application beyond classical planning:

  • State-Space Conflict Based Learning transfers the wide-spread ideas of conflict analysis, backjumping, and nogood learning, quintessential in CSP and SAT, to state space reachability analysis. In difference to previous work in this direction, our algorithms do not assume a given solution length bound.
  • Online Relaxation Refinement improves the accuracy of a heuristic function during search, based on the difficulties encountered. We show that, for recent heuristic functions from classical planning with powerful convergence properties, the approach boosts hill-climbing searches in planning, rendering them complete in theory and competitive with the state of the art in practice.
  • Star-Topology Decoupled State Space Search partitions the state variables into components whose interactions form a star topology. This captures a form of "conditional independence" where leaf components are independent given the center; we show how to exploit this through search over center paths only, avoiding the multiplication of states across leaves. Optimality and completeness of standard search algorithms are preserved, and the empirical benefits are dramatic on benchmarks with a pronounced star topology.
We look at each of these techniques in turn, highlighting the key ideas, the main empirical results, as well as the major lines of future research arising from the current state of affairs.
 

Epistemic General Problem Solving

Invited talk by Michael Thielscher, University of New South Wales

Abstract: General problem-solving systems can understand descriptions of new problems at runtime and solve them without human intervention. I will introduce GDL-III, a formal description language for epistemic planning and general game playing. GDL-III provides a simpler representation language for actions and knowledge than existing formalisms: All that’s required are objective rules about what agents observe and can do. I will define its formal semantics and demonstrate that the language is expressive enough to model common epistemic problems. A proof will be presented that termination of GDL-III games is undecidable even when the game state space is guaranteed to be finite. I will give an overview of two reasoners for general epistemic problems that we are currently developing, and I will discuss how we plan to embed these into a hybrid control architecture for general problem-solving robots to cooperate with humans.

A Short Course in Belief Revision

Invited talk by Pavlos Peppas, University of Patras

Abstract: Thirty years ago, a new research area, now known as Belief Revision, emerged at the cross-section of Formal Philosophy, Cognitive Science, and Artificial Intelligence. In Belief Revision the problem at focus is the process by which a rational agent changes her beliefs in the light of new information. This process plays a central role in many different areas including cognitive robotics, argumentation theory, ontologies, multi-agent systems and software engineering. What makes the problem of revising beliefs non-trivial is that, in principle, it is not enough to simply add the new information to one's stock of beliefs; some of the old beliefs need to be withdrawn on pain of inconsistency. Furthermore, there is typically more than one choice on which beliefs to give up. In this tutorial we will present the most important ideas and results in the area of Belief Revision, and discuss some of its major open problems.