RL Israel Meetup: Value-Based Methods & Beating Human in Games with the Alphas

After the success of the first event, I'm very excited to organize the second Reinforcement Learning Israel meetup! The event is for everyone who's interested to hear more about Reinforcement Learning and its applications.
The meetup will include a teachable talk and an application talk, so everyone is welcome, beginners and experienced practitioners alike.

The meetup is sponsored by Microsoft and will take place at the Microsoft Reactor offices in Tel-Aviv.

Looking forward to seeing you all!
Shani

Schedule:
18:00-18:30 - Gathering with Snacks and Beer
18:30-18:40 - Opening Remarks
18:40-19:25 - Deep RL: Value-Based Methods, Barak Or
19:25-19:30 - A short break
19:30-20:15 - Beating Human in Games with the Alphas, Roi Reshef
20:15-20:30 - Networking

Abstracts:

Title: Deep RL: Value-Based Methods

The topic of value-based methods in Reinforcement Learning will be reviewed. We'll explore and demonstrate some basic value-based methods in the model-free framework, such as Deep-Q-Learning (DQN), dueling and double DQN and experience replay tricks. In addition, some practical applications will be discussed.

Short Bio: Barak Or deals with AI and is interested in its applications to the industry. He is particularly interested in reinforcement learning for solving dynamic problems and computational geometry. Or is the winner of the 2018 Space defense Gemunder prize for his work on "Satellite Interception by Game Theory Approach". Holds a bachelor's and master's degrees in Aerospace engineering in the field of optimal control and estimation for missiles guidance based game theory approaches. Both from Technion. In addition, holds a bachelor's degree in economics and management - from the Technion. Currently, algorithm engineer at Qualcomm.

Title: Beating Human in Games with the Alphas

Go is a rather complicated head-to-head board game invented in China more than 2,500 years ago and is believed to be the oldest board game still played. Due to its complexity, for years it was considered impossible for computers to beat top human players in that game. Hence, it is only natural to use it for benchmarking artificial intelligence as they evolve and become better and better, with the years go by.
2016 was the first year artificial intelligence software surpassed human-level at Go, undoubtedly beating one of the best players at Go (Lee Sedol, 18 world titles) in a best-of-five-game competition that was watched by hundreds of millions of people around the world.
This talk is an introduction to DeepMind's approaches behind AlphaGo, AlphaZero, and AlphaStar - the AI agent that was used to beat world-champion at Go, followed by even further achievements of the same approach on other benchmarks, such as Chess and StarCraft. We will cover the MCTS search algorithm and the reinforcement learning techniques that complement it, and what makes this approach so powerful.

Short Bio: Roi is the Planning and Decision Making team lead for General Motor's autonomous driving flagship project. Prior to that, he worked for companies such as Microsoft, Intel, Outbrain and Viber. Roi holds an MSc in engineering from Ben Gurion University, and his research interests include Robotics, Reinforcement Learning, Deep Learning, and Search.