Differences

This shows you the differences between two versions of the page.

--- teaching:mfe:ia [2016/03/14 16:03]
mdorigo [Localisation and tracking of components in self-assembling systems]
+++ teaching:mfe:ia [2020/04/16 09:07]
bersini [Etude de l'algorithme du Deep Learning]
@@ Line 1: / Line 1: @@
-====== MFE 2015-2016 : Intelligence Artificielle ======
+====== MFE 2018-2019 : Intelligence Artificielle ======
 ===== Introduction =====
-Le laboratoire IRIDIA aborde des problèmes dans le domaine de l'Intelligence Artificielle. Si l'on reprend les dires d'un de ses pioniers Marvin Lee Minsky, l'Intelligence Artificielle est définit comme "la construction de programmes informatiques qui s'adonnent à des tâches qui sont, pour l'instant, accomplies de façon plus satisfaisante par des êtres humains car elles demandent des processus mentaux de haut niveau tels que : l'apprentissage perceptuel, l'organisation de la mémoire et le raisonnement critique". L'IA a beaucoup évolué depuis et s'inspire largement de phénomènes biologiques, physiques, cognitifs ou encore écologiques. C'est donc définitivement une approche transdisciplinaire qui s'accorde principalement à traîtrer des problèmes très complexes. Les domaines principaux de compétence d'IRIDIA sont : l'intelligence en essaim, les métaheuristiques, l'étude des réseaux biologiques et l'application de Business Intelligence. C'est dans cette perspective que les sujets de MFE présentés ci-après s'inscrivent.
+Le laboratoire IRIDIA aborde des problèmes dans le domaine de l'Intelligence Artificielle. Si l'on reprend les dires d'un de ses pioniers Marvin Minsky, l'Intelligence Artificielle est définit comme "la construction de programmes informatiques qui s'adonnent à des tâches qui sont, pour l'instant, accomplies de façon plus satisfaisante par des êtres humains car elles demandent des processus mentaux de haut niveau tels que : l'apprentissage perceptuel, l'organisation de la mémoire et le raisonnement critique". L'IA a beaucoup évolué depuis et s'inspire largement de phénomènes biologiques, physiques, cognitifs ou encore écologiques. C'est donc définitivement une approche transdisciplinaire qui s'accorde principalement à traîter des problèmes très complexes. Les domaines principaux de compétence d'IRIDIA sont : l'intelligence en essaim, les métaheuristiques, l'étude des réseaux biologiques et l'application de Business Intelligence et du Big Data. C'est dans cette perspective que les sujets de MFE présentés ci-après s'inscrivent.
+Ces sujets sont prêt à être encadrer, mais il va s'en dire qu'ils ne sont pas uniques. Les étudiants sont vivement encouragés à prendre contact avec Hugues Bersini (bersini AT ulb.ac.be), Marco Dorigo (mdorigo AT ulb.ac.be) ou Mauro Birattari (mbiro AT ulb.ac.be) afin de discuter de l'une ou l'autre initiative inspirée pouvant faire l'objet dun autre sujet de MFE ou de préciser le cadres, le contenu et les attentes relatives au sujets présentés.
+---
+===== Automatic design of communication protocols in swarm robotics =====
+Automatic design methods are a promising approach to the development of control software of robot swarms. In previous research, we have developed AutoMoDe, a method that automatically generate a finite state machine to control each individual robot of the swarm. AutoMoDe automatically assembles the finite state machine starting from pre-defined behavioral modules and transition criteria.  In this project, the goal is to extend AutoMoDe so as to enable the automatic generation of communication protocols.
+Required skills: The candidate should have good programming skills and previous experience with C++ programming under UNIX.
+* Contact: [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]], Lorenzo Garattoni, Gianpiero Francesca (IRIDIA)
+===== Intelligent interactive console for swarm of robots =====
+The goal of this project is to design and implement an interactive tool for monitoring, debugging and controlling experiments in swarm robotics. Through the interface of this tool, the user can pause the experiment, monitor the state of the robots, select a robot to check the values of sensors and actuators, and modify them if needed. The tool will be integrated in ARGoS (the simulator for robot swarms developed at IRIDIA) and will feature the existing debug facilities featured by ARGoS, which currently work only in simulation. The tool will also use the IRIDIA arena's tracking system, which is equipped with 16 ceiling-mounted cameras. Finally, the tool will integrate an existing console software to monitor and control the state of the robots.
+* Contact: [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]], Lorenzo Garattoni (IRIDIA)
+===== Efficient chain formation in a robot swarm =====
+The goal of this project is to study and design techniques to efficiently create and maintain robust chains of robots. Chain formation is a known collective behavior in swarm robotics. In chain formation, robots place themselves in the environment to create a chain that connects two locations. The chain can be used by other robots as navigation support. Chain formation behaviors are often inspired by ants, which form chains of individuals that connect their nest to foraging sites. Although chain formation has been implemented in several different configurations (e.g., chains of moving robots, chains of aerial robots that aid the navigation of ground robots, directional chains, etc.), the definition of efficient methods to build, use, and maintain chains of robots is still missing. The ultimate goal of the project is therefore the definition of a efficient and robust chain formation behaviour.
+* Contact: [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]], Lorenzo Garattoni (IRIDIA)
-Ces sujets sont prêt à être encadrer, mais il va s'en dire qu'ils ne sont pas uniques. Les étudiants sont vivement encouragés à prendre contact avec Hugues Bersini (bersini AT ulb.ac.be) ou Marco Dorigo (mdorigo AT ulb.ac.be) afin de discuter de l'une ou l'autre initiative inspirée pouvant faire l'objet dun autre sujet de MFE ou de préciser le cadres, le contenu et les attentes relatives au sujets présentés.
@@ Line 13: / Line 43: @@
   * Contact : [[http://code.ulb.ac.be/iridia.people.php?id=1|Hugues Bersini (IRIDIA)]]
 ===== Développer un programme informatique permettant une analyse statistique en vue de  l'évaluation d'un module psychothérapeutique. =====
 Ce mémoire se fera en collaboration avec l'équipe médicale du centre pour l'anorexie et la boulimie de l'hôpital Erasme. Il consistera en l'analyse informatisée des données récoltées lors d'entretiens avec le patient et sa famille au cours du traitement. Les données sont actuellement stockées dans dans une base de données SPSS.  Le mémoire consistera pour l'essentiel au traitement de ces données par des approches "Machine Learning" et "Data Mining" dans une perspective de Quality Management.
+  * Contact : [[http://code.ulb.ac.be/iridia.people.php?id=1|Hugues Bersini (IRIDIA)]]
+===== Développer des solutions de text mining pour la société Procter & Gamble ou dans le domaine juridique =====
+Ce mémoire se fera en collaboration avec des responsables marketing de cette société ou avec des reponsables du monde juridique. Il consistera en l'analyse informatisée des appréciations récoltées sur les produits de la firme et d'une analyse comparative avec les appréciations de concurrents. Il s'agira pour l'essentiel d'études de type "clustering", "classification" et "sentiment analysis" au départ de nombreuses librairies déjà existantes.
+Dans le cas du monde juridique, il s'agira de processus d'organisation et d'indexation automatique des documents juridiques
+  * Contact : [[http://code.ulb.ac.be/iridia.people.php?id=1|Hugues Bersini (IRIDIA)]]
+===== Le data mining en tant qu’outil d’aide à la gestion active d’actifs cotés  =====
+L’idée de ce mémoire serait d’avoir un travail en plusieurs parties dont notamment :
+•	une partie théorique qui reprendrait les fondamentaux du data mining,
+•	un screening, une présentation et comparaison des outils existants utilisés dans le cadre de la gestion d’actifs cotés,
+•	une partie pratique destinée à construire un data set et à le modéliser.
+  * Contact : [[http://code.ulb.ac.be/iridia.people.php?id=1|Hugues Bersini (IRIDIA)]]
+===== Speed-up Techniques for Shortest Path Problems =====
+The shortest path problem is one of the most important in combinatorial optimization and has multiple real-world applications.
+The well-known Dijkstra algorithm solves the shortest path problem in any weighted graph with non-negative weights. However, with the recent increase of the amount of data, we feel the need to drastically increase the execution time of the algorithm.
+In the latest years, we observe the emergence of a lot of new speed-up techniques; among them there are: heuristic search, bidirectional search, landmarks, reaches, short-cuts and contractions. The efficiency of these techniques heavily depends on the topology of the graph. Moreover, graphs with more complex structures (as user-adapted graphs) do not support some of these techniques, at least as they are currently presented.
+From these observations, several questions arise, among them we propose:
+Which optimization methods are the most efficient depending on the graph topology?
+How can we adapt these techniques to more complex graph structures?
+Other research questions in this topic can be proposed and discussed.
+These questions can be studied from both theoretical and empirical approach.
+  * Contact : [[http://code.ulb.ac.be/iridia.people.php?id=1|Hugues Bersini (IRIDIA)]]
+===== Développer des solutions de data mining pour la société BMW Group Belux =====
+Ce mémoire se fera en collaboration avec des responsables marketing et techniques de cette société.
+Les sujets pourraient couvrir tout ce qui concerne les data, tant les data dites de campagne, communication, que celles liées à la production ou aux véhicules connectés.
+  * Contact : [[http://code.ulb.ac.be/iridia.people.php?id=1|Hugues Bersini (IRIDIA)]]
+===== Transposition en logique floue des réseaux de neurones récurrents de type GRU =====
+Ce mémoire consistera en la conception et l'analyse de systèmes de logiques floues, capables d'apprentissage automatique, et très inspirés des réseaux de neurones récurrents de type GRU, utilisés avec succès ces jours-ci pour la traduction automatique, la labélisation automatique d'images, la génération de musique et la conduite de véhicules autonomes.
   * Contact : [[http://code.ulb.ac.be/iridia.people.php?id=1|Hugues Bersini (IRIDIA)]]
@@ Line 33: / Line 111: @@
-===== Etude pratique et expérimentale du langage de programmation F# =====
+===== Etude de l'algorithme du Deep Learning et de l'optimisation des architectures des ces réseaux de neurones =====
-Depuis quelques années, Microsoft met en avant un nouveau langage de programmation F# créé dans le sillage des langages dits déclaratifs ou fonctionnels. Il semblerait que ce langage soit idéal pour le traitement des données. Le MFE consistera en une étude expérimentale de ce langage et un comparatif avec les langages de programmation aujourd'hui les plus usités.
- * Contact : [[http://code.ulb.ac.be/iridia.people.php?id=1|Hugues Bersini (IRIDIA)]]
-===== Etude de l'algorithme du Deep Learning =====
 Les réseaux de neurones multicouches sont redevenus très à la mode depuis que Google les utilise massivement pour le traitement automatique d'images et de vidéos. Nous avons à IRIDIA étudié et réalisé plusieurs algorithmes d'apprentissage de ces réseaux multicouches. Le mémoire consistera en une comparaison des algorithmes de deep learning tels ceux utilisés chez Google et les alternatives que nous proposons ici à IRIDIA depuis plusieurs années.
+Plus particulièrement, nous nous intéressons à des mécanismes automatiques d'optimisation de l'architecture de ces réseaux: Combien de couches, combien de filtres convolutifs par couche...
  * Contact : [[http://code.ulb.ac.be/iridia.people.php?id=1|Hugues Bersini (IRIDIA)]]
@@ Line 162: / Line 236: @@
   * Contact : [[http://code.ulb.ac.be/iridia.people.php?id=1|Hugues Bersini (IRIDIA)]]
+===== Utilisation de réseaux de neurones récurrents pour la détection d'arythmies cardiaques =====
+Ce mémoire s'effectuera en collaboration avec le service de cardiologie d'Erasme et consistera en la détection automatique d'apparition d'arythmies cardiaques à l'aide des réseaux de neurones récurrents, qui semblent être très efficaces pour le traitement des séries chronologiques complexes.
-===== Software framework for Ant Colony Optimization =====
+  * Contact : [[http://code.ulb.ac.be/iridia.people.php?id=1|Hugues Bersini (IRIDIA)]]
-Ants have inspired a number of computational techniques and among the most successful is ant colony optimization (ACO). ACO is an optimization technique that can be applied to tackle a wide variety of computational problems that arise in computer science, telecommunications, and engineering. While ACO has a very wide applicability, the development times for effective ACO algorithms can be relatively high. This is due to the fact that each time a new problem is to be tackled by an ACO algorithm, a researcher needs to implement the algorithms almost from scratch.
-The goal of the project is to provide a software framework to support the application and the implementation of ACO algorithms to new problems. The software framework will offer all the standard procedures that are used in ACO algorithms and will allow for the rapid prototyping of ACO algorithms.
+===== Text Categorisation and quality control through automatic language processing =====
-The application of this software framework will be tested on a number of optimization problems.
-Required skills: The candidate should be well acquainted with   programming in object oriented languages.
+This thesis is developed in collaboration with the Energy Efficiency in Industrial Processes (EEIP) company. EEIP is a global industry information network. As part of their activities, they disseminate case studies to various network groups. The goal of the project is to develop an automatic language processing algorithm capable to evaluate the quality (accept / reject) of the proposed case studies and to allocate them to single/multiple categories. Testing and training the algorithm is a key part as it not only requires development and testing of concepts such as how to evaluate quality or definition of requirements for multiple category allocation but the project also has
+to be developed in a limited data environment (+/- 1000 case studies as training set).
+Required skills: A background in machine learning would be helpful.
+  * Contacts :
+    * [[http://iridia.ulb.ac.be/~stuetzle|Thomas Stützle (IRIDIA)]]
+    * [[https://www.ee-ip.org/|Jürgen Ritzek (EE-IP)]]
+===== Automated configuration of multi-objective continuous optimizers =====
+Many problems arising in real-world applications involve the optimization of various, often conflicting objectives. While the design of algorithms for tackling multi-objective problems has usually done manually, over the recent years automated design methodologies have been established and proved to be very powerful.
+The goal of this project is to extend the automated design to multi-objective continuous optimization problems. As the basis of the approach, a framework based on the two-phase plus Pareto local search approach will be developed into which basic search techniques for continuous optimization will be integrated. The goal is to build first a flexible framework from which then in a second step effective multi-objective optimizers will be generated exploiting automated algorithm design techniques. The final goal of this work is to participate in algorithm competitions with the goal of challenging the methodology.
+Required skills: The candidate should have very good analytical as well as programming skills.
   * Contacts :
-    * [[http://iridia.ulb.ac.be/~mdorigo|Marco Dorigo (IRIDIA)]]
     * [[http://iridia.ulb.ac.be/~stuetzle|Thomas Stützle (IRIDIA)]]
@@ Line 218: / Line 309: @@
     * [[http://code.ulb.ac.be/iridia.people.php?id=1393|Alberto Franzin (IRIDIA)]]
-===== Formal verification of a swarm robotics behavior through statistical model checking =====
-The goal of this thesis is to apply statistical model
-checking to formally verify properties of a collective behavior of a
-robot swarm. Verifying that a system behaves as desired in all
-possible situations is necessary when autonomous robots are involved.
-This is particularly true in swarm robotics systems, where the
-interactions of large number of individuals can result in behaviors
-difficult to predict. Model checking is a common technique to formally
-prove properties of a system. However, its results are limited to
-small systems, because medium-sized or large systems are
-computationally impossible to analyze.
-This thesis is will explore the application of a novel model checking technique, called statistical model checking, to formally verify a swarm robotics system. A collective behavior will be firstly implemented in
-simulation and then analyzed through statistical model checking.
-Required skills: the candidates should be acquainted with C/C++
-programming and have a working knowledge of the English language.
-  * Contacts : [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]] and Manuele Brambilla (IRIDIA)
-===== ROBO-POINTER -- A New Device for Human-Robot Interaction =====
-In this project, the student will design and implement a novel human-robots interaction system.
-The human uses a special laser pointer (called robo-pointer), and by simply pointing to a specific robot he can select it, check its status on a dedicated monitor and send to the robot instructions.
-Now, our arena is equipped with a multi-camera tracking system that allows to detect the 3D position of the robots.
-The student will design a radio sensor network to detect the 3D position of the robo-pointer.
-The two pieces of information will be combined to trigger the selection of the pointed robot.
-* Contacts : [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]] and Andreagiovanni Reina (IRIDIA)
-===== Let a robot swarm get around home! =====
-Robot swarms have very limited knowledge of their environment and therefore, moving around a complex environment (i.e., room with obstacles, building, ...) is a complex task. The idea of this project is to allow an overhead multi-camera tracking system -- that has a complete vision and knowledge of the environment -- to interact with a robot swarm in order to help the swarm move to a specific goal in the environment (for instance, moving from the kitchen to a bedroom of a house).
-The multi-camera tracking system's software is already developed. The student will be supposed to (i) develop a GUI that allows a human operator to specify a goal in the environment, (ii) design and implement a communication protocol (between the tracking system and the robot swarm) that guarantees efficiency and scalability and finally, (iii) test the interaction system in a simulated environment.
-* Contacts : [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]], Gaëtan Podevijn, and Andreagiovanni Reina (IRIDIA)
-===== Online Framework to Compose Robotic Solutions =====
-The design of swarm robotics systems is a challenging task due to difficulties to predict and model swarm systems.
-The most common approach is a trial-and-error process where the developer tries solutions, tests them in simulation and corrects the code to eventually achieve the desired global outcome.
-The project aims to ease this design process: the developer have access to the code of already solved problems (called code-modules), and he can graphically combine the code-modules through a graphical interface to create his own solution.
-The student is supposed to design this novel online framework that offers two main functionalities: (1) to compose new solutions combining the existing code-models to solve new open problems, and (2) to code new code-modules and submit them to the online system.
-* Contacts : [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]], Andreagiovanni Reina, and Gianpiero Francesca (IRIDIA)
-===== UML for Swarm robotics: formal specification of a collective behavior =====
-Swarm robotics is an interesting approach to the
-coordination of hundreds of robots as it promotes the realization of
-systems which are scalable, robust and flexible. However, up to now,
-swarm robotics application has been quite limited, also due to the
-lack of an engineering approach to its development.
-In particular, formal specification has not been applied yet to swarm
-robotics systems.
-In this thesis, we will explore possible ways to formally specify
-swarm robotics systems. As a starting point we will consider UML and
-UML extensions like AUML and UML for multi-agent systems. If
-necessary, we will develop a specific extension for swarm robotics
-systems. Once the preliminary work is done we will consider an
-example, perform formal specification of a task and then implement the
-system in simulation.
-Required skills: the candidates should be acquainted with C/C++
-programming, have a good knowledge of formal specification and UML,
-and have a working knowledge of the English language.
-  * Contacts : [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]] and Manuele Brambilla (IRIDIA)
-===== A GUI for debugging the behavior of a robot swarm =====
-Debugging a robot swarm is a complex and difficult task.
-The desired behavior of the swarm is the result of the complex
-non-linear interactions of tens or hundreds of robots. When
-implementing a swarm robotics system, very often it is necessary to
-analyze individually the output of the execution of each robot, a very
-long and boring process. Since the goal of the developer is to obtain
-a specific collective behavior, it would be better to debug the system
-at the collective level and, only if necessary, at the individual
-level.
-In this thesis we will analyze a possible way to debug the collective
-behavior of a swarm of robots, using macroscopic and microscopic
-modeling. The goal is to develop a GUI that shows the state of the
-collective behavior of the system, and if the user requires it, the
-state of a single robot. We will start with a version of the debugging
-GUI that interface with the ARGoS simulator and eventually one that
-interfaces with the real robots.
-Required skills: the candidates should be acquainted with C/C++
-programming, GUI programming (QT/C++ or QT/Python or Java) and have a
-working knowledge of the English language.
-  * Contacts : [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]] and Manuele Brambilla (IRIDIA)
-===== A virtual machine for mobile code in a swarm of robots =====
-Mobile code is a technology whereby nodes in a network of
-computing nodes exchange code. In other words, code migrates from
-machine to machine like an agent navigating an environment. Mobile
-code is a promising technology for swarm robotics because it would
-enable a new, novel type of robot-to-robot interaction. The aim of this
-project is produce a simple, yet high-performance virtual machine to
-support code exchange in a swarm of robots. A simple experiment with
-the robots demonstrating the capabilities of the VM will be performed.
-Required Skills: Good knowledge of C
-  * Contact: [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]] and Carlo Pinciroli (IRIDIA)
-===== Swarmscope =====
-One the main problems in the development of swarm robotics
-systems is the difficulty of producing, analyzing and debugging code for
-large distributed systems. The aim of this project is to produce a set of
-innovative tools to aid the development of complex swarm robotics
-systems. The produced tools will involve new, creative visualization
-methods and media, novel human-robot swarm interaction and effective
-debugging tools.
-Required Skills: Good knowledge of C++ and Qt4
-  * Contact: [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]] and Carlo Pinciroli (IRIDIA)
-===== Swarm construction: development of remote monitoring software for intelligent structures =====
-S-blocks are dynamically reconfigurable blocks used for swarm-based autonomous construction applications. When two or more S-blocks are assembled they are capable of communicating with each other over a near field communication (NFC) wireless interface. The goal of this master thesis is to develop software to monitor (and control) the blocks in an intelligent structure remotely over the auxiliary Zigbee-based wireless interface. As only one block in the structure is fitted with this wireless interface, it is required that the other blocks communicate with the PC, via routing messages through the block-to-block NFC interfaces. This will require the software on the S-Blocks to be enhanced to use preemptive task swapping, to allow multiple blocks to communicate with each other simultaneously.
-Required skills: The candidates should understand low level computer concepts such as: interrupts, timers, and registers, have some experience with C/C++ programming, and have a working knowledge of the English language.
-* Contact: [[http://iridia.ulb.ac.be/~mdorigo|Marco Dorigo]] (IRIDIA)
-===== Design of a holonomic drive system for autonomous robots in a swarm =====
-Unlike a differential drive system, a holonomic drive system has the advantage of being able to move in any direction at a given instant. The goal of this master thesis is to design and evaluate the performance of a holonomic drive system. The drive system will be assembled from a combination of off-the-shelf components and 3D printed parts. In order to evaluate the drive system, close loop controllers need to be designed and evaluated in C/C++.
-Required skills: The candidates should have some experience with programming in C/C++, and some experience with 3D modelling (preferably Solidworks). The candidates should be able to use basic kinematics to solve simple physics problems, and have a working knowledge of the English language.
-* Contact: [[http://iridia.ulb.ac.be/~mdorigo|Marco Dorigo]] (IRIDIA)
@@ Line 464: / Line 392: @@
-===== Applications of a hybrid SLS algorithm framework =====
+===== Automated configuration of hybrid algorithms =====
-We have recently developed a software framework from which hybrid stochastic local search algorithms can be designed automatically. This framework has only been applied to a few problems. The goal of this project would be to extend this framework to other problems and compare its results with the methods proposed in the literature. The student will learn to solve combinatorial optimization problems with SLS algorithms, automatic configuration of optimization algorithms, and analysis and comparison of optimization algorithms.
+We have recently developed a software framework from which hybrid local search algorithms can be designed automatically. This framework has only been applied to a few problems. The goal of this project would be to extend this framework to other problems, in particular, vehicle routing problems and to compare the results that can be obtained with the methods proposed in the literature. The student will learn to solve combinatorial optimization problems with heuristic algorithms, automatic configuration of optimization algorithms, and the analysis and comparison of optimization algorithms.
   * Contacts :
     * [[http://iridia.ulb.ac.be/~stuetzle|Thomas Stützle (IRIDIA)]]
     * [[http://code.ulb.ac.be/iridia.people.php?id=1388|Federico Pagnozzi (IRIDIA)]]