Various methods based on the Shapley value have enjoyed notable success in recent years within the field of Explainable AI (XAI), in particular as feature selection mechanisms and for providing feature attributions for explaining machine learning models. Nevertheless, recent studies have raised concerns regarding the use of the Shapley value in this framework. In this paper, we delve deeper into these limitations through the lens of the axiomatic analysis of the Shapley value and its implications in the realm of machine learning. Leveraging on specific examples of classification models, we compare the effects of axioms for the Shapley value with other axioms for ranking methods based on a coalitional framework, where features are the “players” and the worth of a coalition of features corresponds to their predictive capacity. As an alternative feature selection method we pay particular attention to the lex-cel, a social ranking solution introduced in the recent literature at the intersection between coalitional games and social choice theory. Our analysis suggests that axioms characterizing the lex-cel, under certain circumstances, are more suitable for ranking features in machine learning models, compared to axioms satisfied by the Shapley value. Furthermore, through experiments conducted on public datasets, we show that the lex-cel outperforms some commonly employed feature selection algorithms based on the Shapley value, in particular with respect to the capacity of selecting less redundant features. An approximated version of the lex-cel, showing a satisfactory compromise between scalability of the approach and selection performance, is also presented and discussed.

Coalition formation over graphs is a well studied class of games whose players are vertices and feasible coalitions must be connected subgraphs. In this setting, the existence and computation of equilibria, under various notions of stability, has attracted a lot of attention. However, the natural process by which players, starting from any feasible state, strive to reach an equilibrium after a series of unilateral improving deviations, has been less studied. We investigate the convergence of dynamics towards individually stable outcomes under the following perspective: what are the most general classes of preferences and graph topologies guaranteeing convergence? To this aim, on the one hand, we cover a hierarchy of preferences, ranging from the most general to a subcase of additively separable preferences, including individually rational and monotone cases. On the other hand, given that convergence may fail in graphs admitting a cycle even in our most restrictive preference class, we analyze acyclic graph topologies such as trees, paths, and stars.

We consider committee election of k >= 3 (out of m >= k + 1) candidates, where the voters and the candidates are associated with locations on the real line. Each voter’s cardinal preferences over candidates correspond to her distance to the candidate locations, and each voter’s cardinal preferences over committees is defined as her distance to the nearest candidate elected in the committee. We consider a setting where the true distances and the locations are unknown. We can nevertheless have access to degraded information which consists of an order of candidates for each voter. We investigate the best possible distortion (a worst-case performance criterion) w.r.t. the social cost achieved by deterministic committee election rules based on ordinal preferences submitted by n voters and few additional distance queries. We show that for any k >= 3, the best possible distortion of any deterministic rule that uses at most k−3 distance queries cannot be bounded by any function of n, m and k. We present deterministic rules for k-committee election with distortion of O(n) with O(k) distance queries and O(1) with O(k log(n)) distance queries.

We study a hedonic game for which feasible coalitions are prescribed by a graph representing the agents’ social relations. A group of agents can form a feasible coalition if and only if their corresponding vertices can be spanned with a star. This requirement guarantees that agents are connected, close to each other, and one central agent can coordinate the actions of the group. In our game, everyone strives to join the largest feasible coalition. We study the existence and computational complexity of both Nash stable and core stable partitions. Then, we provide tight or asymptotically tight bounds on their efficiency, measured in terms of the price of anarchy and the price of stability, under two natural social functions, namely, the number of agents who are not in a singleton coalition, and the number of coalitions. We also derive refined bounds for games in which the social graph is claw-free. Finally, we investigate the complexity of computing socially optimal partitions, as well as extreme Nash stable ones.

This article deals with hedonic skill games, a non-transferable utility counterpart of coalitional skill games which model collaboration among entities through the abstract notions of tasks and the skills required to complete them. In the weighted tasks setting, we show that deciding whether an instance of the game admits a Nash stable outcome is NP-complete. We then characterize the instances admitting a Nash stable outcome. This characterization relies on the fact that every agent holds (resp., every task requires) either a single skill or more than one skill. For these instances, the complexity of computing a Nash stable outcome is determined, together with the possibility that natural dynamics converge to a Nash stable outcome from any initial configuration. Our study is completed with a thorough analysis of the price of anarchy of instances always admitting a Nash stable outcome.

Given two strings A and B such that B is a permutation of A, the max duo-preservation string mapping (MPSM) problem asks to find a mapping π between them so as to preserve a maximum number of duos. A duo is any pair of consecutive characters in a string and it is preserved by π if its two consecutive characters in A are mapped to same two consecutive characters in B. This problem has received a growing attention in recent years, partly as an alternative way to produce approximation algorithms for its minimization counterpart, min common string partition, a widely studied problem due its applications in comparative genomics. Considering this favored field of application with short alphabet, it is surprising that MPSM^𝓁, the variant of MPSM with bounded alphabet, has received so little attention, with a single yet impressive work that provides a 2.67-approximation achieved in O(n) [Brubach, 2018], where n = |A| = |B|. Our work focuses on MPSM^𝓁, and our main contribution is the demonstration that this problem admits a Polynomial Time Approximation Scheme (PTAS) when 𝓁 = O(1). We also provide an alternate, somewhat simpler, proof of NP-hardness for this problem compared with the NP-hardness proof presented in [Haitao Jiang et al., 2012].

International audience

In this paper, we study the problem of assigning PhD grants. Master students apply for PhD grants on different topics and the number of available grants is limited. In this problem, students have preferences over topics they applied to and the university has preferences over possible matchings of student/topic that satisfy the limited number of grants. The particularity of this framework is the uncertainty on a student's decision to accept or reject a topic offered to him. Without using probability to model uncertainty, we study the possibility of designing protocols of exchanges between the students and the university in order to construct a matching which is as close as possible to the optimal one i.e., the best achievable matching without uncertainty.

In this paper, we consider a variant of the facility location problem. Imagine the scenario where facilities are categorized into multiple types such as schools, hospitals, post offices, etc. and the cost of connecting a client to a facility is realized by the distance between them. Each client has a total budget on the distance she/he is willing to travel. The goal is to open the minimum number of facilities such that the aggregate distance of each client to multiple types is within her/his budget. This problem closely resembles to the set cover and r-domination problems. Here, we study this problem in different settings. Specifically, we present some positive and negative results in the general setting, where no assumption is made on the distance values. Then we show that better results can be achieved when clients and facilities lie in a metric space.

This article deals with object allocation where each agent receives a single item. Starting from an initial endowment, the agents can be better off by exchanging their objects. However, not all trades are likely because some participants are unable to communicate. By considering that the agents are embedded in a social network, we propose to study the allocations emerging from a sequence of simple swaps between pairs of neighbors in the network. This model raises natural questions regarding (i) the reachability of a given assignment, (ii) the ability of an agent to obtain a given object, and (iii) the search of Pareto-efficient allocations. We investigate the complexity of these problems by providing, according to the structure of the social network, polynomial and NP-complete cases.

In a voting system, voters may adopt a strategic behaviour in order to manipulate the outcome of the election. This naturally entails a game theoretic conception of voting. The specificity of our work is that we embed the voting game into a social context where agents and their relations are given by a graph, i.e. a social network. We aim at integrating the information provided by the graph in a refinement of the game-theotical analysis of an election. We consider coalitional equilibria immune to deviations performed by realistic coalitions based on the social network, namely the cliques of the graph. Agents are not fully selfish as they have consideration for their relatives. The corresponding notion of equilibrium was introduced by Hoefer et al. [12] and called considerate equilibrium. We propose to study its existence and the ability of the agents to converge to such an equilibrium in strategic voting games using well-known voting rules: Plurality, Antiplurality, Plurality with runoff, Borda, k-approval, STV, Maximin and Copeland.

This article deals with the fair allocation of indivisible goods and its generalization to matroids. The notions of fairness under consideration are equitability, proportionality and envy-freeness. It is long known that some instances fail to admit a fair allocation. However, an almost fair solution may exist if an appropriate relaxation of the fairness condition is adopted. This article deals with a matroid problem which comprises the allocation of indivisible goods as a special case. It is to find a base of a matroid and to allocate it to a pool of agents. We first adapt the aforementioned fairness concepts to matroids. Next we propose a relaxed notion of fairness said to be near to fairness. Near fairness respects the fairness up to one element. We show that a nearly fair solution always exists and it can be constructed in polynomial time in the general context of matroids.