Gabriel Synnaeve

photo de Gabriel Synnaeve

firstname.lastname AT gmail.com

Main address:
LPPA, Collège de France
11 place Marcelin Berthelot
75231 Paris Cedex 05, France

Secondary address:
Team E-Motion, INRIA Rhône-Alpes
655 avenue de l'Europe
38334 Montbonnot, France

PhD student in the E-Motion team at INRIA Rhône-Alpes, in Grenoble. I am working under the supervision of Pierre Bessière at Collège de France (team LPPA), in Paris.

Publications

A Dataset for StarCraft AI & an Example of Armies Clustering, Gabriel Synnaeve, Pierre Bessière, AIIDE Workshop on AI in Adversarial Real-time games 2012
Special Tactics: a Bayesian Approach to Tactical Decision-making, Gabriel Synnaeve, Pierre Bessière, CIG (IEEE) 2012 [title inspired by White-Ra]
A Bayesian Tactician, Gabriel Synnaeve, Pierre Bessière, ECAI Computer Games Workshop 2012
A Bayesian Model for Plan Recognition in RTS Games applied to StarCraft, Gabriel Synnaeve, Pierre Bessière, AIIDE (AAAI) 2011 (slides)
A Bayesian Model for Opening Prediction in RTS Games with Application to StarCraft, Gabriel Synnaeve, Pierre Bessière, CIG (IEEE) 2011 (slides)
A Bayesian Model for RTS Units Control applied to StarCraft, Gabriel Synnaeve, Pierre Bessière, CIG (IEEE) 2011 (slides)
Optimizing Micro-Management in RTS Games, Gabriel Synnaeve, Pierre Bessière, Paris Game AI Conference 2011 (Poster)
Kinetic Models and Qualitative Abstraction for Relational Learning in Systems Biology, Gabriel Synnaeve, Katsumi Inoue, Andrei Doncescu, Hidetomo Nabeshima, Yoshitaka Kameya, Masakazu Ishihata, Taisuke Sato, BIOSTEC Bioinformatics 2011 (Best student paper award) (slides)
A Bayesian Hybrid Approach to Unsupervised Time Series Discretization, Yoshitaka Kameya, Gabriel Synnaeve, Andrei Doncescu, Katsumi Inoue, Taisuke Sato, TAAI-2010
Modeling of a Human MMORPG Player, Gabriel Synnaeve, Pierre Bessière, MaxEnt 2010
Kinetic Models for Logic-Based Hypothesis Finding in Metabolic Pathways, Gabriel Synnaeve, Andrei Doncescu, Katsumi Inoue, ILP 2009

Thesis: Bayesian Programming Applied to Multiplayer Video Games

We adapt Bayesian robotic programming techniques to video games. This work follows the one of Ronan Le Hy that dealt with FPS (first person shooters). Multiplayer video games are a convenient in-between of the real world and simulators, to develop and benchmark AI techniques. Indeed, they are simulated worlds: no sensors problems, finite worlds, limited parameters; and yet, the other players are human beeings (or advanced robots in the case of AI competitions).

Bayesian Programming of Robots Applied to Real Time Strategy Games

Current RTS AI are not enjoyable (too predictible, not hard enough) to play against once you figured out the basics of the gameplay, as opposed to Chess AI or even First Person Shooters AI. Research in this area is using many different AI techniques: from good old fashioned (and working) AI to very recent machine learning techniques. Also, it involves people from all horizons: academics and engineers from the industy. Bayesian programming allows us to have a structured approach of modeling the human player and learn the parameters of this model from available data (replays), or model the units as sensory motor robots.

The purposes of this work are three folds:

Player viewpoint: The goal is to achieve "fun to play against" AI, by having more competitive (stronger) and less predictible (adaptative) AIs. Also, pass the Turing test of video game AI. ;-)
Game developers viewpoint:
- Have additional tools to benchmark/balance the game and explore the meta-game.
- Have an easy framework for easily developing complex/adaptative AI through machine learning instead of "more LOC".
Theoretical viewpoint: The goal is to use robotics/sensory-motor systems designed algorithms and techniques (Bayesian programming) to test their validity with different problems (games AI), developing further what we need along the way.

Bayesian Programming of Robots Applied to MMO Games

Bayesian programming is a machine learning technique consisting of building a probabilistic model of a human player to be used by the robot. The robot will be able to learn play the game by looking at some human-played game sequences. I am currently searching for a MMORPG that would grant us a special permission given the fact that this breaks the normal EULAs. The goal is not to build farmer bots nor to cheat, but to do research in a simulated environment while taking advantage of the fact that interaction with other players are interactions with real humans and not simulated ones.

The applicative purposes of such a work is two folds:

Player viewpoint: The goal is to give the ability to the player to train avatars that could play autonomously in order to improve the game play by having, for instance:
- Intelligent pets
- Intelligent servants and followers
- Persistent characters
Game developers viewpoint: The goal is to give the possibility to the developers to train mobs instead of programming them. This, of course, should drastically reduce development time and hopefully leads to more intelligent and more human like mobs.

Released Software

BroodwarBotQ a bot for Starcraft: Broodwar, using BWAPI.

Teaching

2009-2010: C programming (License 3) and Unix system programming (Master 1) at Polytech Grenoble.
2010-2011: LI105 (initiation to CS), LI223 (Scheme) and LI214 (Logic, automatas, practical work only) at UPMC (Jussieu, Paris 6).
2011-2012: LI105 (initiation to CS, practical work only) and LI220 (Ocaml) at UPMC (Jussieu, Paris 6).

Work related links

7708 StarCraft: Broodwar gosus 1v1 replays, BWTA (terrain analysis) files for some of them
Dataset ready for analysis (Warning: it is not up-to-date with the latest version of bwrepdump), from above replays, documentation (and producing code)
Additional dataset of 1029 Broodwar ICCUP users 1v1 (Warning: it is not up-to-date with the latest version of bwrepdump), with the corresponding ICCUP users 1v1 replays
Results of a survey on players' core characterics to master different gameplay genres.
bayesian-programming.org
Bayesian programming at E-Motion presentation, StarCraft AI presentation and visit report given at SKKU (South Korea).
Summer 2010 project presentation
StarCraft AI competition

Some past and old works

Master Thesis (ILP applied to systems biology) under the supervision of Katsumi Inoue and Taisuke Sato
Biological models for inductive logic programming
I-Terms unification and implementation (C++) Bs.C work under the supervision of Nicolas Peltier

Others

My policy on reviewing and open-access

I reviewed for ICDM 2011, GECCO 2012, and AIIDE 2012 workshop. Now, my policy on reviewing is that I won't accept to review for venues which are not open-access. I firmly believe in open-access research. Considering the very low cost of editing computer science digital papers (basically, a LaTeX macro for page numbering and compiling the PDF), it is unacceptable that conferences proceedings and journals are not freely available. Science has to be spread, read, evaluated, replicated, or it is not science.

My interests

Solve strong AI, use it to solve immortality, use this time to revert entropy. Seriously now, I have interest in all things about intelligence, probabilities, models. I am particularly fond of understanding how we learn (so fast, we have so many priors, in a large sense, and generalize abstract structures and concepts that we re-use). I think experimental psychology and NLP, as well as (not only molecular) biology and medecine will blend in the next 15 years. I hope AI will help science, research, society and the humanity. I enjoy passionate people. I like food (I'm French...), wine, beers, cinema, music (sometimes I even try to play), sailing, and skiing.

My public PGP key

A sticker-printing ready QR code to a Wikipedia random page (English)

A sticker-printing ready QR code to a Wikipedia random page (French)

QR code vers Wikipedia article al�atoire

Education

"Les diplômes ne servent qu'à faire taire les cons."

Scientific Baccalauréat, options engineering and mathematics, summa cum laude (TB), 2004
Lycée Saint Louis, PCSI/PSI (Physics, Engineering, Mathematics), 2004-2006
Grenoble INP, ENSIMAG, French engineer (Master), majored in Computer Science, magna cum laude, 2006-2009
Université Joseph Fourier (Grenoble), Master in Artificial Intelligence and Web, magna cum laude, 2008-2009
Rushed/skimmed online stuff: ML-class (Andrew Ng), Robotic Car (Sebastian Thrun), NLP class (Dan Jurafsky, Chris Manning, without turning in one assignment :))

Raptor