Français Anglais
Accueil Annuaire Plan du site
Home > Research results > Dissertations & habilitations
Research results
Ph.D de RIMMEL Arpad
RIMMEL Arpad
Ph.D
Group : Learning and Optimization

"Improvements and Evaluation of the Monte Carlo Tree Search Algorithm".

Starts on 01/10/2006
Advisor : SEBAG, Michèle
[CORNUEJOLS Antoine]

Funding : A
Affiliation : Université Paris-Saclay
Laboratory : LRI

Defended on 15/12/2009, committee :
Jean-Yves Audibert (examinateur), Université Paris Est
Tristan Cazenave (rapporteur), Université Paris Dauphine
Antoine Cornuéjols (directeur de thèse), AgroParisTech
Phillipe Dague (examinateur), Université Paris Sud
Martin Müller (rapporteur de thèse), University of Alberta
Olivier Teytaud (directeur de thèse), Université Paris Sud

Research activities :

Abstract :
"My thesis deals with planification in a discrete environment with finite
horizon and with a number of states too large to be explored entirely. The
goal is to maximize a reward function that associates a value to final
states. This thesis focuses on particular on improving and evaluating a
new algorithm: bandit-based Monte Carlo tree search. After presenting the
state of the art (Minimax and Alphabeta for the two-players case; nested
Monte Carlo and Dynamic Programing for the one-player case), I describe
the principle of the algorithm. Then, I propose an efficient
parallelization method for the case of separated memories. This method can
be combined with classical parallelization methods for shared memories. I
propose also a way of constructing an opening book and show its efficiency
in the concrete case of the game of Go. I introduce also several ways of
using expert knowledge, in the part concerning bandits as well as in the
Monte Carlo part. Finally, I show that this algorithm that gives very good
results in the context of two-players applications is also efficient in a
one-player context. I propose an adaptation of the algorithm in order to
handle graphs and use a different bandit formula in order to solve the
problem of the automatic generation of linear transforms libraries. I
obtain results much better than by using a classical dynamic programming
algorithm."

Ph.D. dissertations & Faculty habilitations
MICRO VISUALIZATIONS: DESIGN AND ANALYSIS OF VISUALIZATIONS FOR SMALL DISPLAY SPACES
The topic of this habilitation is the study of very small data visualizations, micro visualizations, in display contexts that can only dedicate minimal rendering space for data representations. For several years, together with my collaborators, I have been studying human perception, interaction, and analysis with micro visualizations in multiple contexts. In this document I bring together three of my research streams related to micro visualizations: data glyphs, where my joint research focused on studying the perception of small-multiple micro visualizations, word-scale visualizations, where my joint research focused on small visualizations embedded in text-documents, and small mobile data visualizations for smartwatches or fitness trackers. I consider these types of small visualizations together under the umbrella term ``micro visualizations.'' Micro visualizations are useful in multiple visualization contexts and I have been working towards a better understanding of the complexities involved in designing and using micro visualizations. Here, I define the term micro visualization, summarize my own and other past research and design guidelines and outline several design spaces for different types of micro visualizations based on some of the work I was involved in since my PhD.

A NEW GENERATION OF GRAPH NEURAL NETWORKS TO TACKLE AMORPHOUS MATERIALS


SPOTTING NEURAL NETWORK BOTTLENECKS AND FIXING THEM BY ARCHITECTURE GROWTH