Séminaires 2023-2024

• Mardi 3 octobre 2023, salle S/2: Margot Herin (LIP6, Sorbonne Université), "Learning Compact Preference Representations based on Choquet Integrals for Multicriteria Decision Making"

Multicriteria decision-making requires defining the result of conflicting and possibly interacting criteria. The Choquet or bi-Choquet integral, that represents preference by a non-additive multiattribute utility function, is a major example of preference model allowing interactions of criteria. This preference model is parameterized by one-dimensional utility functions measuring the attractiveness of consequences w.r.t various point of views and one or two set functions (capacities) used to weight the coalitions and control the intensity of interactions among criteria, on the positive and possibly the negative sides of the utility scale. The interplay of both types of parameters and the combinatorial nature of this weighting system makes the preference learning task difficult.

This talk presents an approach to learn successively marginal utilities from properly chosen preference examples and then learn where the interactions really matter in the overall model. In particular, we will see how a sparse representation of the interactions can be learned efficiently using a method based on iterative reweighted least square and dualization. 

• Mardi 17 octobre 2023, salle S/2: Fuad Aleskerov (HSE University, ICS RAS), "Several Urgent Problems in the Arctic"

We consider territorial interests of countries in the Arctic, and the problem of fair division of these territories. The oil reserves found in the Arctic seas increase interest in oil transportation through Arctic seas. We consider related ecological problems. 

• Mardi 7 novembre 2023, salle S/2: Aurélie Beynier (LIP6, Sorbonne Université),  "Resource allocation under diversity constraints and fairness between groups of agents"

We consider the problem of allocating indivisible items to agents where both agents and items are partitioned into disjoint groups. 

The first part of the talk will deal with diversity constraints. Following previous works on public housing allocation, each item (or house) belongs to a block (or building) and each agent is assigned a type (e.g. ethnicity group). The allocation problem consists in assigning at most one item to each agent in a good way while respecting diversity constraints. In this context, we investigate the issue of stability, understood here as the absence of mutually improving swaps, and we define the cost of requiring it. Then we study the behaviour of two existing allocation mechanisms: an adaptation of the sequential mechanism used in Singapore and a distributed procedure based on mutually improving swaps of items.

The second part of the talk will deal with fairness among groups of agents. More specifically we consider the notion of envy-freeness and investigate how it can be adapted to measure the envy between groups of different sizes in house allocation settings.

• Jeudi 23 novembre 2023: OSGAD seminar

• Mardi 12 décembre 2023, salle S/2: Dinko Dimitrov (Saarland University), "Incomplete information and the matching of likes"

We consider a model where market participants can observe the identity of the agents they can be matched with but not their types. A mechanism generates a matching and a public announcement at each reported type profile. For the case of one-sided incomplete information, we fully characterize the set of matching states which are both minimally informative and stable, and show that not every assortative matching mechanism is ex-post incentive compatible. Our main result states that every seemingly informative assortative matching mechanism is ex-post incentive compatible. Such mechanisms become manipulable when incompleteness of information applies to both sides of the market.

• Mardi 19 décembre 2023, salle S/2: Khaled Belahcène (MICS, CentraleSupélec, Université Paris-Saclay), "Explaining and reasoning about redistributive and incomplete preference"

In different decision-making situations, such as group- or multiple criteria-decision-making, it is envisioned to support recommendations with explanations based on normative properties. We shall focus on the case where preference (either those of a single decision-maker or those of society as a whole) are redistributive, in the sense that they tend to favour those who are satisfied the least, and incomplete, in the sense that we eschew the usual assumption of a weak ordering of alternatives. Under these assumptions, we leverage ancient results concerning Lorenz dominance and Pigou-Dalton transfers so as to axiomatically characterize and provide a sound and complete explanation engine based on robust ordered weighted averaging operators (OWAs, a.k.a. generalized Gini indices). These results are based on a foundational formal system (i.e. syntax and semantics) that underlies the reasoning about redistributive and incomplete preference.