SAILabSiena Artificial Intelligence Lab

[Jul 27th 2021] LabMeeting: Continuous Action Spaces vs Discrete Action Spaces in Reinforcement Learning: a Practical Example

, Seminars

Luca Pasqualini (University of Siena)

When: Jul 27th, 2021 – 11:00 – 11:45 AM

Where: Google meet link

Description

In reinforcement learning, with the exception of some control problems, de-facto continuous action spaces are rare. They have a powerful feature though: they can be used to reduce the complexity of combinatorial discrete action spaces. Through a pratical example, I show how to approach this kind of problems and their drawbacks, more often then not deterimental to the overall learning capabilities of the model. Finally, a hierarchical approach is shown in place of the continuous one.

All seminars