DA2PL 2020
DA2PL 2020 vs COVID-19: In order to guarantee the safety, health and well-being of the participants we have decided that DA2PL 2020 will take place entirely virtually. In order to maximise its interactiveness, the event will be organised as a live remote conference.
DA2PL 2020 (From Multiple Criteria Decision Aid to Preference Learning) aims to bring together researchers from decision analysis and machine learning. It provides a forum for discussing recent advances and identifying new research challenges in the intersection of both fields, thereby supporting a cross-fertilisation of these disciplines.
Following the four previous editions of this workshop, which took place in Mons in 2012, Paris in 2014, Paderborn in 2016 and Poznan in 2018, DA2PL 2020 will be held at the University of Trento, Trento - Italy.
Following the four previous editions of this workshop (DA2PL 2012, DA2PL 2014 and DA2PL 2016, DA2PL 2018), the fifth edition has the goal to bring together researchers from decision analysis and machine learning. It aims at providing a forum for discussing recent advances and identifying new research challenges in the intersection of both fields, thereby supporting a cross-fertilisation of these disciplines.
The notion of “preferences” has a long tradition in economics and operational research, where it has been formalised in various ways and studied extensively from different points of view. Nowadays, it is a topic of key importance in fields such as game theory, social choice and the decision sciences, including decision analysis and multicriteria decision aiding. In these fields, much emphasis is put in properly modelling a decision maker’s preferences, and on deriving and (axiomatically) characterizing rational decision rules.
In machine learning, like in artificial intelligence and computer science in general, the interest in the topic of preferences arose much more recently. The emerging field of preference learning is concerned with methods for learning preference models from explicit or implicit preference information, which are typically used for predicting the preferences of an individual or a group of individuals in new decision contexts. While research on preference learning has been specifically triggered by applications such as “learning to rank” for information retrieval (e.g., Internet search engines) and recommender systems, the methods developed in this field are useful in many other domains as well.
Preference modelling and decision analysis on the one side and preference learning on the other side can ideally complement and mutually benefit from each other. In particular, the suitable specification of an underlying model class is a key prerequisite for successful machine learning, that is to say, successful learning presumes appropriate modelling. Likewise, data-driven approaches for preference modelling and preference elicitation are becoming more and more important in decision analysis nowadays, mainly due to large scale applications, the proliferation of semi-automated computerised interfaces and the increasing availability of preference data.
DA2PL 2020 solicits contributions to the usage of theoretically supported preference models and formalisms in preference learning as well as communications devoted to innovative preference learning methods in decision analysis and multicriteria decision aiding. Specific topics of interest include, but are not limited to:
quantitative and qualitative approaches to modelling preferences, user feedback and training data
preference representation in terms of graphical models, logical formalisms, and soft constraints
dealing with incomplete and uncertain preferences
preference aggregation and disaggregation
learning utility functions using regression-based approaches
preference elicitation and active learning
preference learning in combinatorial domains
learning relational preference models and related regression problems
classification problems, such as ordinal and hierarchical classification
inducing monotonic decision models for preference representation
comparison of different preference learning paradigms (e.g., monolithic vs. decomposition)
ranking problems, such as object ranking, instance ranking and label ranking
complementarity of preference models and hybrid methods
explanation of recommendations
applications of preference learning, such as web search, information retrieval, electronic commerce, games, personalization, recommender systems
We solicit full research papers as well as extended abstracts reporting on more preliminary results. Submissions must be written in English and formatted according to the electronic template. Please download the LaTeX template and an example tex file.
The page limit is 6 for full papers and 2 for extended abstracts. As usual for DA2PL, submissions are NOT anonymous. According to the template, names of authors should be stated in the manuscript. All submissions will be reviewed by at least two referees. Submissions must be submitted through EasyChair.
Following a successful experience from the latest editions, DA2PL will provide a special opportunity for doctoral students to explore and develop their research interests. A special session at the conference will be devoted to PhD students, for presenting and discussing their ongoing research work. Therefore, we specifically encourage young researchers to submit their work (as full paper or extended abstract, depending on the maturity of the PhD project); the submission site will provide the option to mark a submission as a student paper.
Submission site opens: April 1, 2020
Paper submission: September 26, 2020
Author notification: October 19, 2020
Registration deadline for authors: October 28, 2020
Camera-ready version: October 31, 2020
Registration deadline for all other participants: 5 pm (local time) November 4, 2020
Conference: November 5-6, 2020
Please find below some documents and videos of the speaches
Nataliya Sokolovska, NutriOmics, Sorbonne Université, nataliya.sokolovska@gmail.com
Abstract: In many prediction tasks such as medical diagnostics, sequential decisions are crucial to provide optimal individual treatment. Budget in real-life applications is always limited, and it can represent any limited resource such as time, money, or side effects of medications. The goal of cascade classifiers under budget constraints is to classify examples with low cost, and to minimise the number of expensive or time-consuming features (or measurements). I will present an approach to learn cost-sensitive heterogeneous cascading systems. Another important aspect of useful classifiers is interpretability. Learning compact but highly accurate models that help in human decision-making is challenging. Most such scoring systems were constructed by human experts using some heuristics. I will discuss principled methods with theoretical guarantees to learn interpretable simple rules purely from data.
Kristen Brent Venable, Florida Institute for Human and Machine Cognition (IHMC) and University of West Florida
bvenable@ihmc.us
https://www.ihmc.us/groups/
Abstract: Preference modeling and reasoning has attracted the interest of researchers from different disciplines, including, economics, operation research, psychology and, of course, artificial intelligence. There is common agreement that preference reasoning is a fundamental component of any realistic intelligent system and their importance has been recently magnified by the fast pace at which the field human-machine interaction has evolved.
In this talk I focus on cognitive models of choice and AI preference techniques and explore some opportunities for synergy between them.
Cognitive models of decision making are computational frameworks grounded in psychology designed to replicate the deliberation process of humans. Among the most successful are those based on the hypothesis that preferences accumulate over time until a certain threshold or deadline is reached and a chosen alternative emerges. They model settings with multiple choices, characterized by multiple attributes and, for the most part, are focussed on one-shot tasks. Their goal is to be good predictors of average human choice behavior and to capture typical context effects observed when humans deviate from rationality. We will give a brief overview of two examples of such models, namely, Multi-attribute Decision-Field (MDFT) Theory and the Multi-attribute Ballistic Accumulator (MLBA), highlighting their complementary properties in terms of expressiveness and computational tractability.
Rationality is instead the underlying assumption of most AI preference models, such as soft constraints which we will consider here as an example, and which are designed to represent compactly preferences over large combinatorial domains in order to allow efficient optimization.
These approaches face common challenges, such as preference elicitation, and bear a high potential of hybridization. We conclude by illustrating recent work in this direction, laying the ground on how cognitive models, AI preference reasoning, and ML can give rise to useful and trusted teammates in decision-making by pairing formal representation and reasoning about objectives with a realistic model of the users’ preferences and choice behavior.
Marc Pirlot, Université de Mons, Belgium
Abstract: Recent years have seen a blossoming of papers related to multiple criteria sorting methods, especially in the framework of the outranking approach. Stimulated by the boom of data mining and machine learning, this literature has developed different research themes within the field of multiple criteria decision analysis such as:
- designing algorithms for learning multicriteria sorting (or clustering) models from assignment examples;
- designing procedures for eliciting the parameters of such methods by questioning the decision maker;
- elaborating more complex or, on the contrary, simpler variants of previously proposed methods;
- characterizing sorting methods by identifying the underlying models.
These contributions showcase how a branch of decision analysis has drawn closer to machine learning, data analysis, AI and social choice, while maintaining specificities of decision theory as for instance, attachment to models and methods, small data rather than large data, lack of ground truth... The talk will survey the recent literature on multiple criteria sorting and clustering, trying to analyze how some themes develop within this community while they echo the methods followed in other domains.
Bio: Marc Pirlot holds a Master degree in Mathematics from Université Libre de Bruxelles and a PhD in Mathematics from Université de Mons-Hainaut (1981). He has been teaching operations research, statistics and decision analysis at the Faculty of Engineering, Université de Mons, since 1989. He served as president of ORBEL, the Belgian OR society (1996-1998) and as principal editor of its journal JORBEL (1992-2002). He is the author or co-author of over 80 scientific publications and 3books which mainly focus on multiple criteria decision analysis, preference modelling and preference learning. He has also been involved in applied research projects with the industry and the public sector.
Francesco Ricci, Free University of Bozen-Bolzano
Abstract: Recommender systems have been introduced already more than 20 years ago, as information search and filtering tools providing suggestions for items to be of use to a user. Group Recommender Systems (GRSs) are especially designed for supporting groups of various nature and size in finding items that may be satisfactorily consumed by the group's members together. Natural applications of GRSs are Travel&Tourism and Music. Standard approaches rely on preference aggregation techniques, and many alternative preference aggregation methods, implementing various types of heuristics, have been proposed. The major obstacle of these methods is that they are agnostic of the specific discussion phase that usually occurs in groups before a decision can be taken; they simply rely on the knowledge of individual preferences before the group meets. We will present and discuss some novel methods that try to either simulate the discussion phase, and can be used for inspecting the quality of a recommendation method, or to learn from observations of real group discussions the most likely decision that the group will make. Both methods try to generate more realistic predictions of the true group behaviour by leveraging personality features, network related measures and machine learning classifiers.
Bio: Francesco Ricci is full professor at the Faculty of Computer Science, Free University of Bozen-Bolzano. F.Ricci has established in Bolzano a reference point for the research on Recommender Systems. He has co-edited the Recommender Systems Handbook (Springer 2015), and has been actively working in this community as President of the Steering Committee of the ACM conference on Recommender Systems (2007-2010). He was previously (from 2000 to 2006) senior researcher and the technical director of the eCommerce and Tourism Research Lab (eCTRL) at ITC-irst (Trento, Italy). F.Ricci's research interests cover: machine learning, user modelling, recommender systems and ICT applications to travel and tourism. Francesco Ricci is author of approximately two hundred refereed publications and, according to Google Scholar, has H-index 54 and around 19,000 citations.
General chair
Andrea Passerini, University of Trento
Program chairs
Vincent Mousseau, CentraleSupélec
Andrea Passerini, University of Trento
Organizing committee
Andrea Passerini, University of Trento
Giovanni Pellegrini, University of Trento
Khaled Belahcene (Heudiasyc, UTC)
Denis Bouyssou (Université Paris Dauphine)
Róbert Busa-Fekete (Paderborn University)
Stéphan Clémençon (Telecom ParisTech)
Yves De Smet (Université Libre de Bruxelles)
Krzysztof Dembczynski (Poznan University of Technology)
Sébastien Destercke (CNRS, Heudiasyc)
Luis Dias (INESC Coimbra and Faculty of Economics, University of Coimbra)
Michael Doumpos (Technical University of Crete)
Paolo Dragone (Criteo)
Johanes Fürnkranz (TU Darmstadt)
Michel Grabisch (Univ. Paris I)
Salvatore Greco (University of Catania)
Eyke Hüllermeier (Paderborn University)
Ulrich Junker
Christophe Labreuche (Thales R&T)
Jérôme Lang (Lamsade - University Paris Dauphine)
Thibaut Lust (UPMC-LIP6)
Thierry Marchant (Universiteit Gent)
Brice Mayag (Lamsade - University Paris Dauphine)
Jérôme Mengin (IRIT - Université de Toulouse)
Patrick Meyer (IMT Atlantique)
Vincent Mousseau (MICS, CentraleSupélec)
Wassila Ouerdane (MICS, Ecole Centrale Paris)
Meltem Ozturk (Lamsade - University Paris Dauphine)
Andrea Passerini (Univ. Trento)
Patrice Perny (UPMC-LIP6)
Marc Pirlot (Université de Mons)
Antoine Rolland (Univ. Lyon 2)
Ahti Salo (Aalto Univ.)
Roman Slowinski (Poznan University of Technology)
Olivier Sobrie (Univ. Mons)
Bruno Teheux (University of Luxembourg)
Mikhail Timonin (Queen Mary University of London)
Alexis Tsoukias (Lamsade - University Paris Dauphine)
Aida Valls (University Rovira i Virgili)
Paolo Viappiani (CNRS and LIP6, Univ Pierre et Marie Curie)
Willem Waegeman (Ghent University)
Conference Office
Communication and Events Office – Polo Collina
University of Trento
comunicazione-collina@unitn.it