O. Teytaud – Monte-Carlo Tree Search : a new paradigm for computational intelligence

Carte non disponible

Date(s) - 14 octobre 2010

Catégories Pas de Catégories

(Séminaire joint avec l’équipe BDAA du LIF.) We will introduce the recent Monte-Carlo Tree Search algorithm and present some of its main successes (including the game of Go, but also industrial applications). We will also present some extensions (continuous domains, partially observable cases leading to interesting decidability issues), and the main directions of research (introducing supervised learning, mixing with handcrafted value functions) – and also the main limitations of the approach – we are still very far from the abstraction power of the human brain.