Preference aggregation, preference elicitation and preference learning

Abstract

Preferences are a widely studied concept in a number of domains including economics, decision theory, game theory.
In computer science, researchers are interested in efficient computational tools for efficiently reasoning about preferences in order to support decision-making for individual or group of users.
Preferences are often interpreted in a quantitative way in terms of “utility” values that are not directly observed but inferred.
Note that policy design is essentially based on values which need to be learned.

Some of the challenging questions that the researchers are the following:

• How to efficiently elicit user preferences (i.e. utilities) in an interactive way (by asking informative questions) in order to limit the cognitive burden posed to the user ?
• How do we elicit and aggregate the preferences of a group of users ?
• How do we learn preferences from a dataset of ranking data ?

In this talk I will describe some methods for interactive preference elicitation, including “setwise max-margin” optimization, a paradigm for elicitation that determine informative queries and can deal with large configuration spaces, it is robust to user inconsistencies in preference feedback, and it can be coupled with regularization terms if sparsity is required. I will then consider the problem of preference elicitation in multi user settings, and present some ideas on “interactive social choice”, where voting protocols interactively ask specific agents to report their vote. Finally I will describe preference learning methods to efficiently aggregate preference rankings into clusters.

Slides

Video

About the lecturer

Paolo Viappiani is a researcher of the French Centre National de la Recherche Scientifique (CNRS) affiliated with the LIP6 laboratory of the University Pierre et Marie Curie (UPMC) in Paris. He holds an engineering degree from Politecnico di Milano (Italy) and a PhD in Computer Science from EPFL (Switzerland). His research interests are within artificial intelligence, algorithmic decision theory, recommender systems, interactive optimization and machine learning. He is currently a member of the editorial board of the EURO Journal of Decision Processes (EJDP) and has been a member of the editorial board of the Journal of Artificial Intelligence Research (JAIR); he often serves as a PC member of highly reputable conferences such as AAAI, IJCAI, ECAI. He has been co-organizer of several workshops related to preferences in artificial intelligence.