Dear colleagues,

Our next BeNeRL Reinforcement Learning Seminar (Oct. 10) is coming: Speaker: Ademi Adeniji (https://ademiadeniji.github.iohttps://ademiadeniji.github.io/), PhD student from UC Berkeley. Title: Reinforcement Learning Behavioral Generalists - Top-Down and Bottom-Up Date: October 10, 16.00-17.00 (CET) Please find full details about the talk below this email and on the website of the seminar series:https://www.benerl.org/seminar-series

The goal of the online BeNeRL seminar series is to invite RL researchers (mostly advanced PhD or early postgraduate) to share their work. In addition, we invite the speakers to briefly share their experience with large-scale deep RL experiments, and their style/approach to get these to work.

We would be very glad if you forward this invitation within your group and to other colleagues that would be interested (also outside the BeNeRL region). Hope to see you on October 10!

Kind regards, Zhao Yang & Thomas Moerland VU Amsterdam & Leiden University

——————————————————————

Upcoming talk:

Date: October 10, 16.00-17.00 (CET) Speaker: Ademi Adeniji (https://ademiadeniji.github.iohttps://ademiadeniji.github.io/) Title: Reinforcement Learning Behavioral Generalists - Top-Down and Bottom-Up Zoom: https://universiteitleiden.zoom.us/j/65411016557?pwd=MzlqcVhzVzUyZlJKTEE0Nk5... Abstract: The success of training large foundation models with scalable, self-supervised objectives has led to significant advancements in AI, particularly in vision and language. In this talk, I argue that while many challenges in general-purpose agentic learning can be mitigated by using these models as black boxes, there remain valuable opportunities for scalable pretraining tailored specifically to the reinforcement learning domain. I will present work from both perspectives: first, showcasing how foundation models trained via conventional methods can enhance decision-making, and second, exploring novel, scalable pretraining approaches that are native to control and hold promise for endowing artificial agents with stronger forms of behavioral generalization. Bio: Ademi Adeniji is a Computer Science PhD student at UC Berkeley advised by Pieter Abbeel. Ademi’s research interests lie in creating agents capable of developing general-purpose intelligent behaviors through data and experience. To this end, his work has focused on self-supervised reinforcement learning algorithms for enabling agents to discover broad control strategies without human supervision for efficiently solving new and unseen tasks. He previously interned at NVIDIA where he worked on reinforcement learning and robotics. He completed his BS and MS at Stanford University conducting research in the Stanford Vision and Learning Lab advised by Fei-Fei Li. He is supported by the Berkeley Chancellors Fellowship.