Past Meetup

Reinforcement Learning: All you need to know

This Meetup is past

83 people went

Location image of event venue

Details

Join us to learn all you need to know about Reinforcement Learning: a thorough introduction, examples, coding, and Deep Reinforcement Learning!

Agenda:
19:15: 'Reinforcement Learning: An Introduction', Anestis Fachantidis
20:15: 'Practical Deep Reinforcement Learning', Nikolaos Passalis
21:00: Networking and socializing
---

'Reinforcement Learning: An Introduction', Anestis Fachantidis

Abstract: Reinforcement Learning (RL) is one of the most ambitious fields of Machine Learning (ML) today. It has attracted a lot of publicity in the last decade, especially after its recent successes such as that of AlphaGo defeating Lee Sedol in the game of Go. Apart from games, what are the other applications of RL ? How does it work and what differentiates it from the rest of the ML approaches ? From designing reward and state signals to the exploration/exploitation problem and the basic RL algorithms we will try to gain an intuitive understanding of the fundamental aspects of RL and we will also take a quick look on how we can represent and solve an RL problem.

Bio: Anestis Fachantidis is Lead Data Scientist in the Intelligent Systems Lab at the Department of Informatics, Aristotle University of Thessaloniki (AUTH) and a Machine Learning postdoctoral researcher at the same department. He has been an Adjunct Faculty member at the Department of Informatics, AUTH, teaching Business Intelligence, Operational Research and Machine Learning [masked]). He holds a PhD in Machine Learning from the Department of Informatics, AUTH, a MSc degree in Information Systems from the Department of Applied Informatics, University of Macedonia, and a Bachelor Degree in Mathematics from the Aristotle University of Thessaloniki. As a Data Scientist, he has designed and developed ML systems for major Greek Businesses including systems for demand forecasting, customer segmentation and fraud detection. As a researcher, his interests focus on Reinforcement Learning, Transfer Learning and Business Intelligence. He has published several articles in refereed journals and conference proceedings and served as a Program Committee member to some of the most significant AI and Machine Learning conferences. He has been a visiting researcher in the Center For Robotics and Neural Systems (CRNS), Plymouth, U.K (2012) and is a member of the Association for Computing Machinery (ACM) and the IEEE Computational Intelligence society since 2013.
---

'Practical Deep Reinforcement Learning', Nikolaos Passalis

Abstract: Combining Deep Learning models with Reinforcement Learning (RL) techniques led to the development of powerful algorithms for solving various problems, ranging from playing Atari to developing self-driving cars, often outperforming humans! However, the large number of different Deep RL methods that have been recently proposed, together the vast amount of different hyper-parameters that must be tuned, leads to a large number of design choices that must be taken before even attempting to train a Deep RL agent. Therefore, it is not always straightforward to directly use Deep RL to tackle the problem at hand and a significant amount of experimentation and fine-tuning might be required. In this hands-on tutorial we will go through the most important deep RL algorithms and we will implement them from scratch using PyTorch! We will examine various ways to debug the algorithms, solve various issues that might arise and apply them in different problems.

Bio: Nikolaos Passalis obtained his Ph.D. in Informatics, specializing in Deep Learning, from the Department of Informatics, AUTh in 2018. Starting from December 2018, he will be postdoctoral researcher at the Signal Processing Laboratory of the Technical University of Tampere, Finland. He has (co-)authored more than 30 papers published in international journals and conference proceedings. His research interests include deep learning, computational intelligence and information retrieval.
---