Streaming Deep Reinforcement Learning Finally Works · HackerPapers discussions

Streaming Deep Reinforcement Learning Finally Works — mentioned by AndrewKemendo

Thu, 28 Aug 2025 14:47:23 GMT

Too long and complex for my morning but ultimately we’re at a point of hardware/computing commodification and ubiquitous data that online RL can start to provide meaningful efficiencies across any possible real task Here’s two good papers to start with: https://arxiv.org/abs/2410.14606 https://storage.googleapis.com/deepmind-media/Era-of-Experie...

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

Streaming Deep Reinforcement Learning Finally Works — mentioned by AndrewKemendo

Sun, 15 Jun 2025 02:34:48 GMT

Q learning isn’t scalable because of the stream barrier, however streaming DRL (TD-Lambda) is scalable: https://arxiv.org/abs/2410.14606 Note that this is from Turing award winner Richard Sutton’s lab at UofA RL works

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

Streaming Deep Reinforcement Learning Finally Works — mentioned by AndrewKemendo

Fri, 16 May 2025 13:02:07 GMT

I’m very excited to see that cybernetic thinking is finding its way into a more mainstream audience as people are understanding the generality of computing systems in the longest term I have not read the book in question yet so I will reserve judgment however based on this post it would seem that the majority of its claims have already been captured in much more specific detail in the 2018 Sutton and Barto reinforcement learning textbook (1). Specifically, if you look through section 3 chapters 14 and 15, they specifically address the psychology and neuroscience congruence with the TD(L)…

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

Streaming Deep Reinforcement Learning Finally Works — HN submission

Sun, 01 Dec 2024 05:20:02 GMT

Streaming Deep Reinforcement Learning Works

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