Just like last year, I had the pleasure (indeed a privilege) of delivering another lecture on the current state of deep reinforcement learning (deep RL) on Nov 15, 2018, at NTNU, Trondheim. The field is moving very fast, so I get to talk about something new every time, which is fun.
The lecture was open to the public and streamed live. It was mostly attended by students taking courses offered by Prof. Keith Downing. It was also wonderful to see some of my immediate colleagues from Telenor Research and academic friends from NTNU. Some of the material I covered will likely be stale by the time you reach here. Nevertheless, below is the recording, and slides are available here. A shout out to BRAIN NTNU for taking care of the local arrangements.
What does it cover and who is it for?
The lecture attempts to draw a broad and intuitive picture of the field as it stands. The historical development of the field is examined, building up to current frontiers. One aim of this lecture is to spread awareness of current advances and some of the key technical challenges towards scaling deep RL methods to real world control/sequential decision making problems. It also aims at inspiring curiosity to tackle these challenges amongst students at University, data scientists, and academic and industry researchers intending to work in the field. A lot is going on to address each one of these challenges. Lecture favours covering enough conceptual ground to being exhaustive.
Why another lecture on the topic?
As is perhaps obvious, exposing fresh minds to the frontier and the technical challenges therein, is one way to make some of them curious about advancing the field. When these minds get to work, the field invariably matures, so does its applicability, trickling those advances into the industry. Lectures like these are necessary to keep the momentum going. Since the research effort in the field is moving quite rapidly, it also helps to checkpoint the current state every now and then.
Keep calm and advance RL!
It is indeed of great help that my work at Telenor Research revolves around applying and advancing RL. Although it takes some effort to keep up, it is fun to do so, and discussions with colleagues and students with similar interests keeps things under control. It has indeed been a great pleasure for me to get to work with a few in the recent past, in part due to supervisory activities at the Norwegian Open AI Lab, and through this reading group. I hope to keep learning from them and the wider community in the foreseeable future!