This talk covers the systems and algorithmic challenges behind making globally distributed RL work at scale. We will discuss how we decouple training from inference, propagate compressed weight updates across datacenters, control policy staleness, and design the infrastructure to tolerate frequent hardware failures. We will also examine the training–inference mismatches that emerge in asynchronous RL and the techniques we use to preserve policy entropy and maintain stable learning over long training runs.
Speakers: Carlo Baronio, Research Engineer, Cognition
