Eleventh Workshop

The eleventh Hybris Workshop takes place on May 7-8 2018 at the University of Potsdam, Germany.

Venue

The workshop sessions takes place at

Haus 4 (Institut fuer Informatik),
August-Bebel-Straße 89,
14482 Potsdam

in room 1.02 on the first floor.

Abstracts

The best of both worlds: Machine learning meets logical reasoning

Invited Talk by Holger H. Hoos Leiden Institute of Advanced Computer Science (LIACS), The Netherlands and University of British Columbia, Canada

Machine learning and logical inference are foundational pillars of artificial intelligence. Research in both areas has blossomed over the last twenty years and produced significant, rapidly increasing impact in many important applications. While machine learning and logical inference both play a key role in the AI revolution that is currently underway, on the surface, they seem rather different worlds. In this talk, I will explore an intriguing connection between those worlds: the use of machine learning for the automated design and analysis of logical inference engines. I will use prominent work from SAT solving to illustrate how this connection can be leveraged, from performance modelling and prediction to automated selection, configuration and parallelisation of state-of-the-art SAT solvers. I will then explain how this line of work brings about a fundamental change in the way in which we design and deploy not only cutting-edge solvers for SAT and its important generalisations, but algorithms and software for a broad range of problems from AI and beyond. I will also introduce the concept of automated AI (or AutoAI), which promises to make cutting edge AI techniques more accessible, more effective and more broadly applicable.

Slides

Go was solved with a hammer, then what are the other nails?

Invited Talk by Carmel Domshlak, Technion Israel Institute of Technology

From Go playing to focused web crawling to physical systems’ control automation, reinforcement learning algorithms, such as Q-learning and UCT, are perceived as key enablers of grand successes in large-scale multi-state decision problems. But what was the actual marginal contribution of these algorithms to the success stories? What *exactly* can we learn from these successes? Most importantly, what of these success stories can and should be translated nicely to many other decision problems of our interest?

A few years ago, these questions took us into a small journey that brought us to interesting, surprising, and sometimes even “shocking” revelations. Along the journey, we have also learned a few things about building competitive algorithms for action selection in multi-state decision processes, with (originally unintentional) focus on Monte-Carlo sampling algorithms. In this talk I will summarize for you some of our key findings, developments, and intermediate conclusions, while trying to make the presentation as self-contained as possible.

Joint work with Zohar Feldman and Alexander Shleyfman, and not to forget all my friends with whom I had pleasure to discuss the subject on a cup of coffee.

Program

Participants