• LuxLogAI 2018

    Luxembourg Logic for AI Summit

    17-26 September 2018


Invited Speakers

LuxLogAI is pleased to feature various top-class invited keynotes of its participating events, including speakers of the 2nd International Joint Conference on Rules and Reasoning (RuleML+RR 2018), 4th Global Conference on Artificial Intelligence (GCAI 2018) and the Deduktionstreffen 2018. A summary of the talks can be found below. See also the list of invited tutorials at LuxLogAI.

You can click on the following links to jump to the talk(s) of the respective day.

Monday, 17 September [top]

Marco Lippi
University of Modena and Reggio Emilia
Artificial Intelligence for Consumer Law

Monday, 17 September, 09:00, room: MSA 3.120, MIREL

Abstract: TBA

Daniele Nardi
Sapienza Università di Roma
Knowledgeable Robots

Monday, 17 September, 09:30, room: MSA 4.530, GCAI

Abstract: In the talk we will first introduce the approach followed by our recent research in Artificial Intelligence and Robotics, which we regard as an attempt towards general Artificial Intelligence. Our aim is to build systems that achieve high levels of competence in specialized domains, by learning it incrementally, as opposed to the main trend of creating systems, that can work from scratch in any domain. Our long term plan is to address three types of knowledge: about environment, tasks and user.
As of today, we can report results in the first two realms, in particular, we shall present our recent work on semantic mapping and task learning. In addition, we shall focus on the domain of service robotics and address performance evaluation of the systems we are developing, specifically focusing on robot competitions in this domain.

Daphna Weinshall
Hebrew University of Jerusalem
New old frontiers in deep learning: curriculum learning, generative models

Monday, 17 September, 14:00, room: MSA 4.530, GCAI

Abstract: In the first part of the lecture I will talk about curriculum learning, where a learner is exposed to examples whose difficulty level is gradually increased. This heuristic has been empirically shown to improve the outcome of learning in various models. Our main contribution is a theoretical result, showing that learning with a curriculum speeds up the rate of learning in the context of the regression loss. Interestingly, we also show how curriculum learning and hard-sample mining, although conflicting at first sight, can coexist harmoniously within the same theoretical model. Specifically, we show that it benefits to start training with easier examples with respect to the global optimum of the model, while at the same time preferring the more difficult examples with respect to the current estimate of the model’s parameters. Finally, we show an empirical study using deep CNN models for image classification, where curriculum learning is shown to speed up the rate of learning, AND improve the final generalization performance.
In the second part of the lecture I will talk about a new GAN variant, which we call Multi-Modal-GAN. I will show how this model can be used for novelty detection, and also augment data in a semi-supervised setting when the labeled sample is small. Finally, I will show interesting unsupervised clustering results, with comparable results to state-of-the-art supervised classification using the MNIST dataset.

Guillaume Aucher
University of Rennes 1/INRIA
Principles for a judgement editor based on BDD

Monday, 17 September, 14:00, room: MSA 4.120, MIREL

Abstract: We describe the theoretical principles that underlie the design of a software tool which could be used by judges for writing judgements and for making decisions about litigations. The tool is based on Binary Decision Diagrams (BDD), which are graphical representations of truth–valued functions associated to propositional formulas. Given a specific litigation, the tool asks questions to the judge; each question is represented by a propositional atom. Their answers, true or false, allow to evaluate the truth value of the formula which encodes the overall recommendation of the software about the litigation. Our approach combines some sort of ‘theoretical’ or ‘legal’ reasoning dealing with the core of the litigation itself together with some sort of ‘procedural’ reasoning dealing with the protocol that has to be followed by the judge during the trial: some questions or group of questions must necessarily be examined and sometimes in a specific order. That is why we consider extensions of BDD called Multi-BDD. They are BDD with multiple entries corresponding to the different specific issues that must necessarily be addressed by the judge during the trial. We illustrate our ideas on a case study dealing with French union trade elections, an example that has been used throughout a project with the French Cour de cassation.

Tuesday, 18 September [top]

Valeria de Paiva
Nuance Communications
Bridging Trouble

Tuesday, 18 September, 09:30, room: MSA 3.520, Joint invited talk of RuleML+RR and GCAI

Abstract: Some ten years ago, when I left Xerox PARC to work for a search startup, I hadn’t realized how much the work I had done till then was not mine and could not be continued, for licensing reasons. For almost nine years at PARC I worked on a project to create logic from language, the Bridge project, using a collection of technologies developed by a strong collection of researchers, through at least two decades, under the leadership of Bobrow and Kaplan. I decided that I needed to redo my part of this work, using only open source tools, as I was not ready to give up on the idea of logic from language. I gave a talk at SRI, explaining my reasons and plans, published in ENTCS as ”Bridges from Language to Logic: Concepts, Contexts and Ontologies”, LSFA2010. This talk recalls and unifies some of the research that came up from this project and that is scattered in applications. We focus on a methodology for producing specific domain knowledge from text that we hope to improve, but that is already producing promising initial results, based on Universal Dependencies.

Wednesday, 19 September [top]

Philipp Slusallek
Saarland University/DFKI

Wednesday, 19 September, 09:30, room: MSA 3.510, Joint invited talk of RuleML+RR and GCAI

Abstract: tba.

Georg Gottlob
University of Oxford
Vadalog: A Language and System for Knowledge Graphs

Wednesday, 19 September, 14:00, room: MSA 3.520, Joint invited talk of RuleML+RR and GCAI

Abstract: With the introduction of its Knowledge Graph, Google has coined the name for a new generation of knowledge-based systems that go beyond what was previously expected of areas that include graph databases, knowledge bases, machinelearning systems and rule-based logical reasoners. Beyond Google, companies are recognizing the need for making use of their vast amounts of data and knowledge in the form of enterprise knowledge graphs. In the same way that databases created the need for Database Management Systems (DBMS) and knowledge bases fostered the creation of Knowledge Base Management Systems (KBMS), the interest in Knowledge Graphs creates a need for academia and industry to understand and develop knowledge graph management systems (KGMS).

Thursday, 20 September [top]

Hannah Bast
University of Freiburg
Efficient and Convenient Search on Very Large Knowledge Bases

Thursday, 20 September, 14:00, room: MSA 3.520, RuleML+RR

Abstract: Knowledge bases like Freebase or Wikidata have hundreds of millions of entities and billions of triples. Searching such knowledge bases is challenging in many ways. First, already importing the data dumps from such knowledge bases into standard triples stores is hard: it can take forever or does not work at all without preprocessing. Second, even relatively simple queries can take a very long time to process, in particular queries with large result sets. Third, formulating queries in SPARQL is hard even for experts, since it requires knowledge of the exact names or ids of the involved predicates and entities. Fourth, it is often desirable to combine knowledge base search with keyword search, but basic SPARQL provides little support for this. We will present ideas and solutions for all four of these challenges, as well as various demos based on these ideas and solutions.

Friday, 21 September [top]

Guido Governatori
Modal Rules: Extending Defeasible Logic with Modal Operators

Friday, 21 September, 09:00, room: MSA 3.520, RuleML+RR

Abstract: In this talk we present a general methodology to extend Defeasible Logic with modal operators. We motivate the reasons for this type of extension and we argue that the extension will allow for a robust knowledge framework in different application areas.

Pascal Fontaine
Université de Lorraine
Quantifier handling in SMT

Friday, 21 September, 11:00, room: MSA 3.120, Deduktionstreffen

Abstract: Satisfiability Modulo Theories (SMT) solvers are increasingly used in verification as back-ends to check the satisfiability of formulas in presence of interpreted symbols. After a brief overview of their architecture, we will review the current methods used in those solvers to handle quantifiers. An early technique, E-matching or trigger-based instantiation, generates instances when some patterns of terms are present in the formula. Although it is experimentally quite successful, it might generate many useless instances. Model-based quantifier instantiation is a more recent semantic instantiation technique: instances are built to fix a tentative model that does not agree with the quantified formulas. An even more recent method to which we have contributed is conflict-based instantiation; it can be seen as a weaker and more efficient form of model-based instantiation, that generates only the instances that contradict the partial model in the ground solver. Finally, we will also report on our results on revisiting enumerative instantiation.

Cynthia Kop
Radboud University Nijmegen
Wanda: A Higher-Order Termination Tool

Friday, 21 September, 16:00, room: MSA 3.120, Deduktionstreffen

Abstract: Wanda is a fully automatic tool to analyse termination of higher-order term rewriting systems in various styles. In this talk I will present Wanda and discuss the underlying methodology, which includes both a discussion of the theory involved and the strategies for automation.