Systèmes complexes en sciences sociales

Année 2017/2018

Henri Berestycki (directeur d’études à l’EHESS)
Jean-Pierre Nadal (directeur d’études à l’EHESS et directeur de recherche au CNRS)

Le séminaire « Systèmes complexes en sciences sociales » reprend à partir du vendredi 6 octobre, et se tiendra tous les 2ème et 4ème vendredis de cette année 2017-2018 (voir le site de l’Ehess), à 14h30, (sauf mention contraire) en salle A04-47, 54 bd Raspail, 75006 Paris.


  • Vendredi 23 février, 14h30, salle A4-47, EHESS, 54 boulevard Raspail, 75006 Paris.

Matthieu Latapy
Directeur de recherche au CNRS, LIP6, UPMC
Analyser des interactions temporelles avec les flots de liens
Des transferts d’argent ou de données, des contacts entre individus, ou des achats de produits, toutes ces activités sont assimilables à des interactions temporelles. Étudier la structure et la dynamique de ces interactions revêt un caractère crucial pour la compréhension de nombreux phénomènes et pour de nombreuses applications (détection d’événements dans du trafic, détection de fraudes, recommandation de produits, optimisation de réseaux, construction de relations pérennes à partir d’interactions ponctuelles, etc). La structure de telles interactions est étudiée en utilisant des graphes ou des réseaux (ensembles de nœuds et de liens) ; leur dynamique est étudiée en utilisant des signaux ou des séries temporelles (variations d’une propriété au cours du temps) ; pour étudier la dynamique de leur structure, on utilise des séquences de graphes. Toutefois, ces approches ne capturent que de façon très limitée la nature à la fois structurelle et temporelle des interactions, qui nécessite un cadre spécifique.
Nous présentons ici une généralisation des graphes, que nous appelons des flots de liens, permettant un traitement cohérent des deux aspects. Nous obtenons un langage pour l’étude directe des séquences d’interactions, similaire à celui des graphes pour l’étude des relations.

  • Lundi 15 janvier, 14h30, salle S1-23 (1er sous-sol), EHESS, 54 boulevard Raspail, 75006 Paris.

Janet Pierrehumbert
Professor of Language Modelling,  Dept. of Engineering Science, University of Oxford, UK
Remembering and generalizing from examples of words.
People learn words from experienced examples of them. The words in turn provide the statistical foundation for learning word-formation patterns. What information about words is encoded and remembered? How are generalisations formed from stored examples? When people encounter variable input, do they simply remember and reproduce this variation, or do they systematise it in their own outputs?
In this talk, I will present results from corpus analyses and online game-like experiments that address these questions. These indicate that mental representations of words include much detail, including contextual detail such as social-indexical associations. At the same time, there is a great amount of individual variation in what information is encoded and the way abstract generalizations are formed from remembered examples. The basic mechanism for learning word formation is not probability-matching, as assumed in recent Bayesian models. People have a propensity to systematise the input, but differ in what contextual associations they notice, how open they are to unexpected input, and how aggressively they systematise. These results support claims that heterogeneity in the speech community plays a strong role in language variation and change.

  • Vendredi 22 décembre, 14h30, salle A4-47, EHESS, 54 boulevard Raspail, 75006 Paris.

Julien Perret
Chargé de recherche du Ministère de la Transition écologique et solidaire (MTES)
LaSTIG (Laboratoire des Sciences et Technologies de l’Information Géographique)
IGN (Institut National de l’Information Géographique et Forestière)
Simplu3D : un modèle de simulation des droits à bâtir
Simplu3D est un modèle de simulation des droits à bâtir permettant de générer des configurations bâties qui respectent les règles locales du droit de l’urbanisme (PLU/POS/RNU). Il s’agit de résoudre un problème d’optimisation d’une fonction objectif (le volume totale de la configuration bâtie par exemple) avec des contraintes géométriques (hauteur de construction, distances aux limites séparatives, etc.). La méthode de résolution choisie utilise un recuit simulé et un échantillonnage stochastique par Reversible-jump Markov chain Monte Carlo (RJMCMC).
Dans cette présentation, nous présentons le fonctionnement du modèle, son utilisation, mais aussi sa validation et les perspectives qu’ouvrent un tel modèle pour une meilleure compréhension des réglementations d’urbanisme, la concertation autour des projets d’urbanisme et, plus généralement, une meilleure prise de décision en matière d’urbanisme et d’aménagement du territoire.


  • Vendredi 8 décembre, 14h30, salle A4-47, EHESS, 54 boulevard Raspail, 75006 Paris.

Paul Pezanis-Christou
School of Economics, The University of Adelaide, Australia
Exploration vs Exploitation, Impulse Balance Equilibrium, and the El Farol bar problem
Many social and economic activities require some degree of coordination in people’s actions to be enjoyed. Congestion games, in which a group of individuals contemplates participating in an event that is enjoyed only if a few participate, characterize well such situations. Such games have been widely investigated in the social sciences, and laboratory experiments suggest that despite the existence of multiple equilibria in pure strategies, participants somehow manage to behave almost optimally: overall, their behaviour is found to converge to the game’s (unique) symmetric Nash mixed-strategy equilibrium. This pattern was first coined as ‘magic’, and has subsequently been explained in terms of reinforcement learning processes, probability distortions or overconfidence. Yet, some investigations report persistent deviations which can be rationalised in terms of Quantal Response Equilibrium: a stochastic version of the Nash equilibrium which is remarkably successful in organizing the behaviour observed in numerous economic experiments.
The paper provides two new rationales of behaviour in a class of congestion games that is widely used in economics to study market-entry decisions and often referred to as the El Farol bar problem (Arthur, AmEconRev, 1994). The first model, EvE, uses an ‘Exploration vs Exploitation’ argument and happens to be structurally equivalent to that of the Quantal Response Equilibrium (McKelvey and Palfrey, GamesEconBehav, 1996). The second model, IBE, uses an Impulse Balance Equilibrium argument inspired from Selten et al. (ExpEcon, 2005) that makes a heuristic prediction as to how agents play games and assumes no best-responding behaviour: agents only balance the foregone payoffs associated to the options available.
Both EvE and IBE are parsimonious stationary models that involve only one parameter. To assess their explanatory power, we conduct market-entry experiments that manipulate both payoff structures and payoff levels, and that involve groups of ten participants playing for 150 rounds. The data indicate that these two models explain equally well the data in terms of goodness-of-fit whenever the observed probability of entry is less than the Nash equilibrium prediction; otherwise IBE marginally outperforms EvE. The estimates are also constant throughout the experiments which suggests no obvious learning pattern in participants’ behaviour. When assuming symmetric agents (as is usually done in the experimental economics literature) and when estimating the models with group rather than pooled data, IBE yields more consistent estimates than EvE. However, the opposite occurs when the symmetry assumption is relaxed, and the conduct of a specification test overwhelmingly rejects the null of consistency of the data with either model. The pros and cons of modelling heterogeneity in such games is briefly reviewed.
Overall, our study indicates that the conclusions drawn from the usual approach of assuming symmetric agents when estimating such models crucially depends on whether ones uses pooled or group data and that in the latter case, these models have in fact little empirical support.
Work in collaboration with Alan Kirman and François Laisney.

 


  • Séance double, mercredi 22 novembre à 14h30, à l’Institut Henri Poincaré, Amphithéâtre Hermite, 11 rue Pierre et Marie Curie, 75005 Paris.

Cette séance spéciale est organisée dans le cadre de la conférence de clôture de l’ERC ReaDi (« Reaction-Diffusion Equations, Propagation and Modelling »), qui se tient du 20 au 23 novembre à l’IHP. Elle remplace les deux séances du séminaire, initialement prévues pour les 10 et 24 novembre.

14h30 :
José Scheinkman
Columbia University
Supply and Shorting in Speculative Markets
This is joint work with Marcel Nutz (Columbia University). We propose a continuous-time model of trading among risk-neutral agents with heterogeneous beliefs. Agents face quadratic costsof-carry on their positions and as a consequence, their marginal valuation of the asset decreases when the magnitude of their position increases, as it would be the case for risk-averse agents. In the equilibrium models of investors with heterogeneous beliefs that followed the original work by Harrison and Kreps, investors are risk-neutral, short-selling is prohibited and agents face a constant marginal cost of carrying positions. The resulting resale option guarantees that the equilibrium price exceeds the price of the asset in a static buy-and-hold model where speculation is ruled out. Our model features three main novelties. First, increasing marginal costs entail that the price depends on the exogenous supply. Second, in addition to the resale option, agents may also value an option to delay, and this may cause the market to equilibrate below the static buy-and-hold price. Third, we introduce the possibility of short-selling; then the resale option for agents with short positions partly compensates the resale option for long agents. We characterize the unique equilibrium of our model through a Hamilton–Jacobi–Bellman equation of a novel form and use it to derive several comparative statics results.

15h30 :
Jean-Pierre Nadal
CNRS et EHESS
The 2005 French riots: a data-driven epidemiological modeling reveals a spreading wave of contagious violence
During autumn 2005, after two youth died while trying to escape a police control, riots started in a poor suburb of Paris, spread around and then in all France, hitting more than 800 municipalities and lasting about 3 weeks. Thanks to an access to a detailed, day by day, account of these events, we analyzed the dynamics of these riots. In this talk I will show that a parsimonious data-driven epidemic like model, taking into account both local (within city) and non-local (through geographic proximity or media) contagion, allows to reproduce the full time course of the riots at the scale of the country. I will make explicit the specificity of the model as compared to the modeling of the spread of infectious diseases. I will also discuss the methodology in the broader context of modeling social phenomena, showing in particular how modeling provides a kind of regularization of the data. In the case of the 2005 French riots, it allows to visualize the wave propagation around Paris in a way not described before.
Ref: Laurent Bonnasse-Gahot, Henri Berestycki, Marie-Aude Depuiset, Mirta B. Gordon, Sebastian Roché, Nancy Rodrez and Jean-Pierre Nadal, « Epidemiological modeling of the 2005 French riots: a spreading wave and the role of contagion « , submitted, arXiv:1701.07479


Deux séances exceptionnelles se tiendront à l’Institut des Systèmes Complexes (ISC-PIF, 113 rue Nationale 75013 Paris, M° Olympiades/Nationale), co-organisées avec le laboratoire SAMM (Université Paris 1 Panthéon Sorbonne) :

*Please note the following access codes for the ISC-PIF building: 2605 and 2606.*
The seminar room is located on the first floor.

  • Vendredi 6 octobre 2017à 10 h 30.

Modeling Archeological Interactions in Space, avec Barbara Mills (Univ. Arizona) et Oliver Nakoinz (Univ. Kiel).

Plus d’informations ici.

Il s’agit aussi d’une demi-journée satellite du colloque Interactions du SAMM, et qui sera (pour)suivie par une table-ronde aux 20e Rendez-vous de l’Histoire de Blois le lendemain (voir ici).

  • Mardi 10 octobre 2017, 14h30.

Première séance d’une série autour des travaux de Thomas Schelling, décédé en décembre dernier, avec une conférence de William Clark (UCLA, professeur invité à Paris 1).

William A. V. Clark
Selection, sorting and social distance : the tegacy and continuing relevance of Thomas Schelling.
Although Thomas Schelling was awarded the Nobel prize for “having enhanced our understanding of conflict and cooperation though game theory analysis”, for most social scientists his legacy is his work on the “tipping point” and how small differences in preferences can have macro effects on spatial organization. That thinking has been the basis for popular works like the Big Sort. The central contribution of Micromotives and Macrobehavior was to show how behavioral actions at the individual level create aggregate outcomes across social structures. A combination of agent based and mathematical modelling, and survey analyses of mobility behavior, continues to inform our understanding of how social distance creates residential patterns. Because most residential mobility involves short distances the process of initial selection and continuing adjustment creates clustering by social distance. Sorting by race, ethnicity, income and education is likely to continue despite social intervention. At the same time there is some survey evidence of changing preferences in US contexts which may alter ethnic but not economic sorting.

Plus d’informations ici.


Archives :