The 2019 edition of NeurIPS, the leading conference in machine learning and AI, has come and gone. The conference took place from December 8th to the 14th at the Vancouver Convention Center. The week was filled with insightful papers, workshops and panels/talks that Sportlogiq’s innovation team, SLiQ Labs, was thrilled to attend.
Sportlogiq Presents Panel on Explainable Reinforcement Learning
Sportlogiq has been a proud sponsor of the NeurIPS conference since 2015. This year marked the first time Sportlogiq hosted a panel at the conference. Moderated by our CTO and Co-Founder, Mehrsan Javan, the panel discussed approaches for making reinforcement learning transparent, interpretable, and accepted by industry clients.
The highlight of the panel was about turning models into something that can be understood by the end user. Although different users demand different levels of explanations from the models (doctors and nurses vs patients), at the end when we design the RL systems we need to make sure that the given optimal policy is compatible with the users expectations. This means that the first step for designing explainable RL and inverse RL models is to understand the end user’s utility function and optimize for the same function. This comes down to actively involve the domain experts in the development process which is usually ignored when building AI models.