This talk does not have an abstract.

This talk does not have an abstract.

This talk does not have an abstract.

This talk does not have an abstract.

This talk does not have an abstract.

In reinforcement learning, an agent chooses actions in order to maximize the rewards given by a dynamic environment. As the environment is initially unknown, the agent has to interact with it to gather information. Moreover, only the reward of the chosen actions is revealed. That is why the agent faces the exploration/exploitation dilemma: she has to explore loosely estimated actions in order to build a better estimate, and she would like to maximize her cumulated reward by playing the empirically best actions. The mathematical framework which handles the exploration/exploitation dilemma is called the multi-armed bandit problem. In comparison to the general reinforcement framework, the main difference is that the state of the environment does not depend on the actions chosen previously. This simplification makes it possible to build efficient algorithms that can be proven optimal. Although the multi-armed bandits have been designed for clinical trials, they are used in many fields such as advertising, recommendation, marketing optimization, web site optimization, and since a decade in telecommunication networks. The first use case was the Opportunistic Spectrum Access (OPA), where a Secondary User aims to find a channel free from Primary User. With the development of IoTs networks and 5G networks, new use cases have emerged for instance for configuring Self Organized Network, or for optimizing energy consumption in IoT and sensor networks.

Biography

Raphaël Féraud has obtained is PhD in 1997 at University of Rennes I. During his PhD, he has worked on Neural Networks applied to Face Detection in images and video. Then at France-Télécom R&D, he has worked on different application of Neural Networks for Telecommunication including churn, profiling, fraud detection, resources allocation in ATM networks. As a project leader he has lead several projects on data mining and Big Data at Orange Labs with applications on marketing optimization and Ad targeting. Then as researcher at Orange Labs, he worked on reinforcement learning and in particular on bandit algorithms. His research on contextual bandits, non-stationary bandits and multi-player bandits are applied for optimizing communication in IoT networks.

This talk does not have an abstract.

This talk does not have an abstract.

This talk does not have an abstract.

This talk does not have an abstract.

This talk does not have an abstract.