TECoSA Research Seminar: Explainable Reinforcement Learning for Telecom

Speakers:  Industrial PhD students Ahmad Terra (TECoSA, Ericsson) and Franco Ruggeri (Ericsson)
Location: C435, Brinellvägen 85 (plan 4)
Zoom link and sign-up link circulated to members
Please email vickid@kth.se if you have any questions.

ABSTRACT: Explainable Reinforcement Learning (XRL) primarily to provide explainability for black-box models to enable trust. XRL includes multiple methods applied to different elements of RL agent (state, rewards, policy explanations, etc.). This seminar will present multiple existing state-of-the-art XRL methods, their application to different elements especially to the remote electrical antenna tilt (RET) use case. Additionally, the seminar includes a new method “Both Ends Explanations for RL (BEERL)” that goes one-step further to connect and compare input and output explanations, identify and mitigate bias, and generates two different levels of explanation which in turn allows RL agent re-configurations when unwanted behaviours are observed.