Deep reinforcement learning methods traditionally struggle with tasks  where environment rewards are particularly sparse. In this video we  discuss two-stage method that overcomes these limitations by relying on  noisy, unaligned footage without access to such data.

Paper: https://arxiv.org/abs/1805.11592

Deep Learning book: https://amzn.to/2FTO77C

Connect with me on Twitter: https://twitter.com/aijournalyt

Please support me on Patreon : https://patreon.com/aijournal