Reporting Uncertainty
Emphasizing the importance of reporting uncertainties, Rishabh discusses how confidence intervals complement distributions in reinforcement learning experiments. He introduces a novel metric that calculates the probability of one algorithm outperforming another, highlighting that this probability itself is a random variable subject to uncertainty. This approach encourages a deeper understanding of the variability in algorithm performance.In this clip
From this podcast
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
Deep Reinforcement Learning at the Edge of the Statistical Precipice with Rishabh Agarwal - #559
Related Questions