On Tuesday 12/06 at 14:30 we will have an interesting invited talk by Prof. Olivier Pietquin (Google Brain, Paris) on the topic: «Guesswhat?! - Learning strategies for visually grounded dialogue»
Abstract: In this talk we will present the Guesswhat?! game and the associated database. Guesswhat!? is a language-based game supported by an image in which one player selects an object in the image and the other has to discover that object by asking yes-no questions in natural language. A database of 150k dialogues have been collected and is freely available for research. Code for supervised learning baselines is also available. In addition, we will present recent work on Reinforcement Learning applied to that environment and some improvements brought to the supervised learning approach based on conditioning on language a feature-wise modulation of convnets.