| old_uid | 11847 |
|---|
| title | Journée Evolution artificielle Thématique |
|---|
| start_date | 2016/06/24 |
|---|
| schedule | 09h30-17h30 |
|---|
| online | no |
|---|
| summary | 10:00-11:00 – SESSION 1
a) Pascal Seppecher, "Is the market really a good teacher? Market selection, collective adaptation and financial instability"
This paper proposes to model market mechanisms as a collective learning process for firms in a complex adaptive system, namely Jamel, an agent-based, stock-flow consistent macroeconomic model. Inspired by Alchian’s (1950) “blanketing shotgun process” idea, our learning model is an ever-adapting process that puts a significant weight on exploration vis-à-vis exploitation. We show that decentralized market selection allows firms to collectively adapt their overall debt strategies to the changes in the macroeconomic environment so that the system sustains itself, but at the cost of recurrent deep downturns. We conclude that, in complex evolving economies, market processes do not lead to the selection of optimal behaviors, as the characterization of successful behaviors itself constantly evolves as a result of the market conditions that these behaviors contribute to shape. Heterogeneity in behavior remains essential to adaptation in such an ever-changing environment. We come to an evolutionary characterization of a crisis, as the point where the evolution of the macroeconomic system becomes faster than the adaptation capabilities of the agents that populate it, and the so far selected performing behaviors suddenly cease to be, and become instead undesirable.
b) Laurence Devillers, "Dimensions affectives et sociales dans les interactions homme robot : éthique, évaluation et usages"
summary not yet available.
11:15-12:15 – SESSION 2
a) Jing Yang, "Optimal Parameter Choices via Precise Black-Box Analysis "
In classical runtime analysis it has been observed that certain working principles of an evolutionary algorithm cannot be understood by only looking at the asymptotic order of the running time, but that more precise estimates are needed. In this work we demonstrate that the same, possibly even more pronounced, applies to black-box complexity analysis. We prove that the unary unbiased black-box complexity of the classic OneMax testfunction class is $n \ln(n) - cn \pm o(n)$ for a constant $c$ between $0.254$ and $0.266$. Our analysis yields a simple (1+1) evolutionary algorithm achieving this runtime bound via a fitness-dependent mutation strength. When translated into a fixed-budget perspective, our algorithm with the same budget computes a solution that is 13\% closer to the optimum (given that the budget is at least $0.4n$).
b) Arthur Bernard, "Mechanistic Constraints in the Evolution of Coordination "
La thématique est centrée autour de l'étude de l'évolution de la coopération à l'aide d'outils de robotique évolutionniste. J'aborde donc une question posée par la biologie (comment la coopération a-t-elle pu évoluer) avec une approche de simulations robotiques. Le but est de montrer que, à l'aide de ce genre d'approches (et par rapport à des approches plus classiques de théorie des jeux), nous pouvons montrer le rôle de l'évolution de la coordination dans l'apparition de la coopération. Pour cela, j'utilise une expérience de chasse collective avec une modélisation individu-centré.
13:30-15:00 – SESSION 3
a) Carola Doerr, "Improving Applicability of Nature-Inspired Optimisation by Joining Theory and Practice (ImAppNIO)"
Nature-inspired search and optimisation heuristics are easy to implement and apply to new problems. However, in order to achieve good performance it is usually necessary to adjust them to the problem at hand. Theoretical foundations for the understanding of such approaches have been built very successfully in the past 20 years but there is a huge disconnect between the theoretical basis and practical applications. The development of powerful analytical tools, significant insights in general limitations of different types of nature-inspired optimisation methods and the development of more practically relevant perspectives for theoretical analysis have brought impressive advances to the theory-side of the field. However, so far impact on the application-side has been limited and few people in the diverse potential application areas have benefitted from these advances. The main objective of the COST Action is to bridge this gap and improve the applicability of all kinds of nature-inspired optimisation methods. It aims at making theoretical insights more accessible and practical by creating a platform where theoreticians and practitioners can meet and exchange insights, ideas and needs; by developing robust guidelines and practical support for application development based on theoretical insights; by developing theoretical frameworks driven by actual needs arising from practical applications; by training Early Career Investigators in a theory of nature-inspired optimisation methods that clearly aims at practical applications; by broadening participation in the ongoing research of how to develop and apply robust nature-inspired optimisation methods in different application areas.
b) Julio Navarro-Lara, "Morwilog, an ant-hill algorithm for detecting security anomalies"
Threat detection is one of the basic mechanisms for protecting a network. Finding an attack is difficult because the most harmful ones are targeted against an specific victim and crafted for the first time. Moreover, in many occasions intrusions can only be efficiently detected by analyzing its effects on more than one network elements. Log recollection offers a way to centralize event information from an heterogeneous set of sources, which can be normalized to a common language and analyzed as a whole by a security system. In this paper we propose Morwilog, an ant-inspired method for detecting unknown attacks represented by a sequence of logs.
c) Cédric Pupka, "Présentation de Mosgo, la plateforme de benchmark d'algorithmes de classification pour la détection d'anomalies"
La détection d’anomalies est un sujet de premier plan dans de nombreuses applications, telles que la détection de fraudes, d’intrusions dans un système informatique ou encore à destination du domaine médical. De nombreuses techniques ont été développées, certaines pour des cas bien précis d’autres encore fournissent des outils plus génériques. Le plus souvent, on retrouve des approches statistiques sous le prisme de l’apprentissage automatique ou de l’informatique bio-inspirée. Une des grandes difficultés vient du nombre important d’algorithmes pour la détection d’anomalies. Dans le domaine de l’apprentissage automatique, ce sont plusieurs milliers d’algorithmes, sans compter les centaines qui sont publiées chaque année dans diverses publications scientifiques. Il convient alors de structurer les approches en définissant une typologie, permettant de discerner l’algorithme le plus adéquat dans un cas donné. Bien que cette démarche permette de fournir un cadre de décision quant au choix d’un algorithme de détection d’anomalies, il reste néanmoins difficile de prédire la performance des algorithmes dans un contexte. Notre objectif est de réaliser un « Benchmark » d’algorithmes de classification pour la détection d’anomalies, avec comme enjeu de fournir des données quantitatives et qualitatives sur les performances des d'algorithmes face à différents jeux de données liés à la sécurité. Les algorithmes qui sont évalués sont : les systèmes immunitaires artificiels pour la reconnaissance (AIRS), les tables de décision, les arbres de décision (J48), le perceptron, la classification bayésienne naïve et les machines à vecteurs de support (SVM). Au cours de cette présentation, il sera question de présenter nos premiers résultats ainsi que la plateforme Mosgo développée dans le cadre de nos recherches.
15:20-17:00 – Meeting of the association “Evolution Artificielle” |
|---|
| responsibles | Tonda, Lutton |
|---|
| |