Speaker: Tamas Endrei.

Abstract: Reinforcement learning (RL) has emerged as one of the most fascinating fields of machine learning, providing solutions to challenging problems ranging from complex robotics behaviors to optimizing neural network architectures. Despite its immense potential, RL’s complex nature and the difficulty of implementation relative to other deep learning fields leave this field underutilized in many cases. In this seminar, we aim to demystify RL by exploring its fundamental concepts, the intuitions behind it, while showcasing the promising prospects this field could yield. Furthermore, we explain the intuitions behind the most commonly found algorithms and tricks in their implementation, elucidating the chain of thought behind the evolution of reinforcement learning algorithms while also mentioning current limitations and caveats. Finally, we conclude the seminar by proposing a reinforcement learning framework for speech enhancement, aiming to directly optimize audio quality as a fine-tuning step of current state-of-the-art deep learning methods.