The seminar will introduce students to a selected area of the algorithms of deep Reinforcement Learning. At the end of the seminar course, the student will be able to
- Understand the advanced Reinforcement Learning algorithms.
- Understand the application areas.
- Find a preference algorithm to solve Reinforcement Learning problems.
- Basic knowledge of reinforcement learning
- Familiar with supervised learning methods
- Familiar with basic matrix algebra and optimization algorithms
- Familiar with deep neural networks
- Familiar with fundamentals of control-theory
- Familiar with Python and its libraries
Lectures and presentations will be given via Zoom every Wednesday from 10.00-12.00. The link is the same for each seminar: https://aalto.zoom.us/j/65356356800
- The grading scale is pass/fail.
- Participation in every seminar is compulsory.
- One page (250-300 words) summary of each session.
- One presentation based on the paper listed or can choose a related topic.
- Become an opponent in at least one session.
- Grading assessment depends on technical correctness, writing quality, and language.
- Total teaching hours 12 hours
- Independent study 15 hours
- Preparation to presentation work, reading 15 pages, 30~40h
- Written work 15 hours (5 sessions x ~3 h)
- DQN [Mnih et al., 2013]
- Double DQN [Van Hasselt et al., 2016]
- PER [Schaul et al., 2015]
- QT-OPT [Kalashnikov et al., 2018]
- AlphaGO [Silver et al., 2016]
- TRPO [Schulman et al., 2015]
- PPO [Schulman et al., 2017]
- Deep Dyna-Q [Peng et al., 2018]
- SAC [Haarnoja et al., 2018]
- DDPG [Lillicrap et al., 2015]
- I2A [Racaniere et al., 2017]
- Inverse RL [Choi and Kim, 2012]
- MBPO [Janner et al., 2019]
- CQL [Kumar et al., 2020]
Sign-up for the topic and reading resources: Link
Presentation 1 File
Presentation 3 File
Presentation 4 File
Presentation 5 File