RL PPO Paper + What could we do if we haven't got enough data?
Details
I am very glad to announce the first DL meetup of the year 2019. There will be 3 presentations, and they are going to be purely technical, the first two of them will be mostly theory, and the last one will be practical (going through source code). We are going to be at a new location. This does not mean we leave Emarsys, we will just try the location of Millenaris Startup Campus (we are supported by IVSZ this time).
- presenter: Adam Budai, Ph.D. Student, ML researcher, Teaching Assistant at BME
https://www.linkedin.com/in/adambudai/
He will talk about the recent stage of OpenAI's new approach, OpenAI Five, to defeat the top players in Dota 2. He will introduce the core idea of the used reinforcement learning algorithm, the so-called PPO, and some further tweaks. Dota 2 is a very complex game but the proposed algorithm is surprisingly simple.
Paper: https://arxiv.org/pdf/1707.06347.pdf
- presenter: Levente Szabados, ML Consultant, Dozent at Frankfurt School of Finance & Management, Teacher at KÜRT, Mentor
https://www.linkedin.com/in/levente-szabados-76334728/
He will talk about some tricks what to do if we don't have enough data in our ML project. These tricks can be cost sensitive Learning, different sampling techniques, data generation and data augmentation, transfer learning. etc.
- presenter: Miklos Toth, ML Engineer, Freelancer IT Consultant, Big Data (Spark) Instructor at Datapao
https://www.linkedin.com/in/miklostoth/
He will build on the top of the second presentation and will show the theory in practice. With other words, he will introduce some of the above-mentioned techniques in practice with Python/Tensorflow/Keras. The source code will be available on GitHub after the meetup.
Location: Millenaris Startup Campus
https://www.millenaris-startupcampus.hu/
Budapest, Kis Rókus u., 1024, D Building
Networking starts at 18:00, the presentations will be started at 18:30.
We will have pizza and beverages as always.
The event is supported by IVSZ. ( http://ivsz.hu/ )
The event will be recorded on video.