The Tea Time Talks 2020: Week Six

Now that the 2020 Tea Time Talks are on Youtube, you can always have time for tea with Amii and the RLAI Lab! Hosted by Amii’s Chief Scientific Advisory Dr. Richard S. Sutton, these 20-minute talks on technical topics are delivered by students, faculty and guests. The talks are a relaxed and informal way of hearing leaders in AI discuss future lines of research they may explore, with topics ranging from ideas starting to take root to fully-finished projects.

Week six of the Tea Time Talks features:

Parash Rahman: Stochastic Gradient Descent in a Changing World

In this talk, Parash discussed the online learning setting where a predictor learns to predict from a stream of data -- a problem setting which is important to handle changes found in real-world applications. He further discusses the surprising mediocre adaptability of multilayer networks updated with stochastic gradient descent when the data distribution is changing, which have otherwise been successful in modern applications. Finally, he considers the use of generate-and-test algorithms in this problem setting.

Yufeng Yuan: Multimodal Observation Space for Robot Learning

In this talk, Yufeng explains multimodal observation space for robot learning, and how it’s different from other more commonly-used observation spaces, such as pixel observation. He then introduces useful tricks for learning completely from pixels and investigates their effectiveness in the multimodal setting.

Han Wang: Emergent Representations in Reinforcement Learning and their Properties

Representation learning remains one of the central challenges in reinforcement learning. Earlier representation learning work has focused on designing fixed-basis architectures to achieve desirable properties. However, recent work suggests that representations emerge under appropriate training schemes, and properties should thus be determined by the data stream. In this talk, Han explores properties of representations trained end-to-end with different auxiliary tasks, provides novel insights regarding the auxiliary task effect, and investigates the relationship between properties and transfer learning performance.

Martha Steenstrup: Control of Communications Networks - Tryin' to Make it RL Compared to What?

To design effective algorithms for controlling the performance of a communications network, one must confront several challenges relating to environment dynamics, fidelity of observations, responsiveness, appropriate credit assignment and cost-benefit trade-offs. In this talk, Martha argues that RL algorithms are well-suited to surmount these challenges, with supporting evidence garnered from her group’s application of RL to the problem of congestion control.

The Tea Time Talks have now concluded for the year, but stay tuned as we will be uploading the remaining talks in the weeks ahead. In the meantime, you can rewatch or catch up on previous talks on our Youtube playlist.