Date: Friday, October 31
Time: 12-1pm
Location: Soda 510
Title: RL at Databricks
Abstract: In this talk, we will take a journey through reinforcement learning research at Databricks. RL plays many different roles at Databricks for our 20,000 enterprise customers, powering our in-house models, providing custom training, and – through its constituent parts – enabling a new generation of AI products. We will discuss how our customers’ practical RL challenges differ from academic framing and the novel research we have conducted to address them. After all, the only true eval is whether AI works in production for real people. We hope you will take away a new set of practically-relevant fundamental research questions and insight into what it takes for AI to reliably perform economically valuable work. (*No Mateis were harmed in the making of this research.)
Bio: Jonathan Frankle is Chief AI Scientist at Databricks, where he leads research on reinforcement learning. He arrived at Databricks through the $1.3B acquisition of MosaicML, where he was co-founder and Chief Scientist. He completed his PhD at MIT, focusing on his “lottery ticket hypothesis” (ICLR best paper award, MIT dissertation award, ACM SIGAI dissertation award). He lives in New York, and he forgives Californians for their second-rate bagels.
Ashutosh Baheti is a senior research scientist at Databricks, where he has performed novel reinforcement learning research culminating in multiple product launches. He completed his PhD at Georgia Tech, studying RL for LLMs before it was cool. He aspires to someday create JARVIS, a dream that grows nearer to reality by the day.