Deep Reinforcement Learning that Matters · HackerPapers discussions

Deep Reinforcement Learning that Matters — mentioned by MasterScrat

Wed, 15 Apr 2020 14:18:54 GMT

This is an important direction. For now, training reinforcement learning agents mean that you need to simulate an environment, for example a car simulation, or a robotic simulation, or a video game. The simulation can be quite slow. In the worst case, your environment may actually not be a simulation, but a real-world experiment - in which case each interaction is even slower. For now, this is something RL researchers have to deal with. If we could get "offline" reinforcement learning to work, ie learning from pre-recorded experiences only, this would bring a considerable boost to the field…

Paper and discussion history. A mention is not an endorsement.

Deep Reinforcement Learning that Matters — mentioned by nealeratzlaff

Tue, 02 Jan 2018 22:53:55 GMT

I think for most people that hurdle of understanding is immense, and isn't likely to get any better. ML/DL infrastructure will continue to improve and deploying AI systems will become more accessible as time goes on. But I don't see familiarity with (non) convex optimization, statistical learning, topology, etc. becoming mainstream. Without these its hard to see a reasonable path forward for AI. For me this year in DL was a lot narrower. it mostly revolved around systems becoming more human in capability. I think most of this is in the article already: We can now hopefully say that ImageNet…

Paper and discussion history. A mention is not an endorsement.

Deep Reinforcement Learning that Matters — HN submission

Thu, 21 Sep 2017 01:22:16 GMT

Deep Reinforcement Learning That Matters

2 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.