Section: Overall Objectives
One major breakthrough of TAO regarding optimal control in uncertain environments was illustrated by the MoGo program, first ever Computer-Go program to win a professional human player (more in section 6.5 ). This achievement, acknowledged to be a “notable success of AI” by The Economist (jan. 2007) is based on original tree-structured extensions of Multi-Armed Bandit algorithms (Fig. 1). Its parallelization was undertaken in collaboration with the Parall team and Bull; collaborations with Microsoft started in 2008, and a branch of MoGo has been developped jointly with the National University of Tainan (NUTN, Taiwan). Notably, the approach involves little specific domain knowledge; it directly tackles optimal policy learning under uncertain/incomplete information.
Recent developments of MoGo include the use of patterns (off-line learning) and the on-line exploitation of information gathered through the search, referred to as RAVE. The combined use of patterns and RAVE pioneered by TAO is now widely used in the litterature. Among the applications of the general MoGo principles are the optimization of the mathematical Spiral library (coll. CMU).
This work has been widely disseminated to the general public: IA-Go Challenge in Paris in March; Gold medal Olympiads 2008, first ever win against a pro in 9x9 Go in 2007, first ever win against a pro in 9x9 Go in 2008 without blitz, first ever win against a pro in 19x19 Go with handicap 9 in 2008. Three more world successes (still at the top of the state of the art as of Nov. 2009) have been featured by TAO in 2009, win against i) a top professional player with handicap 7 (Tainan 2009); ii) a professional player with handicap 6 (Tainan 2009); and iii) a top professional player as black in 9x9 Go(With the standard komi 7.5, 9x9 Go is more difficult for the black; this result is the first win ever in such a situation.) (Taipei 2009).
Other fundamental results regard:
advancing the theory of stochastic optimization, based on the adaptation of statistical learning theory tools to evolutionary optimization. Specifically, Monte Carlo Markov Chains were used for proving the convergence of Evolution Strategies (ES) and analysing their convergence rate. Other results regard the optimization of computationally expensive functions or the use of quasi-random algorithms. The robustness of ES compared to standard, Newton-based approaches in convex settings with high condition number has been demonstrated experimentally.
investigating design-oriented representations, enforcing stability, compacity, and versatility in the framework of EC-based applications in Numerical Engineering. Various search spaces have been considered, ranging from developmental representations to Echo State Networks. Architectural applications in collaboration with EZCT, opened promising avenues for research (exhibitions at the Beaubourg Museum of Modern Art; primed contribution to the Seroussi contest, Fig. 2).
pioneering ML-based Autonomic Computing, a major field of applications for both learning and optimization. This research theme benefitted from the arrival and expertise of Cécile Germain in 2005 (Pr. UPS) and Balász Kégl in 2007 (CR1 CNRS), strengthening the link between LRI and Laboratoire de l'Accélérateur Linéaire. Within EGEE(The Enabling Grid for E-SciencE infrastructure project, spanning 2004-2010 (FP6 and FP7).), a Grid Observatory (http://www.grid-observatory.org/ ) was launched to gather and edit traces of grid activity, supporting the behavioural modelling of the grid and ultimately aimed at its self-management.
investigating Automatic parameter tuning, in the framework of the joint lab Microsoft-INRIA. This goal has been tackled through two applications, respectively on-line tuning of evolutionary parameters (best paper award LION 2009) and on-line heuristics selection in Constraint Programming.
During the year 2009, TAO has confirmed its position as a key research group in Evolutionary Computation (invited talks at 8th ISEA in June 2009, at GECCO 2009, at 3rd Workshop TRSH; Editor in Chief of MIT Press Evolutionary Computation journal; Best Paper Awards at LION'09 conference in January, at EvoBIO'09 in April, and in the Continuous Optimization track at ACM-GECCO in July; as well as in Machine Learning (steering committee of PASCAL and PASCAL-2 NoEs, 2003-2013; chair of ECML/PKDD 2010). It has expanded its activities to Data Mining (invited talk at IEEE Forum on DM 2008) and Complex Systems (chair of ANR SYSCOMM Evaluation Committee; invited talk at Eur. Conf. on Complex Systems 2009). TAO activities have also been recognized as a source of applicative breakthroughs (OMD RNTL; MoGo; coll. with EZCT nominated for Seroussi contest). Following earlier recommendations, TAO 's robotics activities are now supported by an active collaboration with hardware roboticists (SYMBRION IP), witnessed by the organization of two robotics events (collocated workshops at IROS and ECML/PKDD 2009).