Our next BeNeRL Reinforcement Learning Seminar (Nov 16) is coming: 

Speaker: Cansu Sancaktar (https://csancaktar.github.io), PhD student at Max Planck Institute. 

Title: Playful Exploration in Reinforcement Learning

Date: November 16, 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 also maintain a page with summary of advice of previous speakers: https://www.benerl.org/seminar-series/summary-advice-on-reinforcement-learning-experimentation

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 November 16!

Kind regards,

Zhao Yang & Thomas Moerland

Leiden University

——————————————————————

Upcoming talk: 

Date: November 16, 16.00-17.00 (CET)

Speaker: Cansu Sancaktar (https://csancaktar.github.io)

Title: Playful Exploration in Reinforcement Learning

Zoom: https://universiteitleiden.zoom.us/j/65411016557?pwd=MzlqcVhzVzUyZlJKTEE0Nk5uQkpEUT09

Abstract: Designing artificial agents that explore their environment efficiently via intrinsic motivation, akin to children's curious play, remains a challenge in complex environments. In this talk, she will present CEE-US which combines multi-horizon planning, epistemic uncertainty reduction, and structured world models for sample-efficient and interaction-rich exploration, particularly in multi-object manipulation environments. She will showcase how the self-reinforcing cycle between good models and good exploration opens up another avenue: zero-shot generalization to downstream tasks via model-based planning. In the second part of the talk, she will introduce regularity as an intrinsic reward signal, which we proposed in recent work. She will present how taking inspiration from developmental psychology, they inject the notion of creating regularities/symmetries during free play and achieve superior success rates in assembly tasks in a multi-object manipulation environment.

Bio: Cansu Sancaktar is a PhD student at the Max Planck Institute for Intelligent Systems, supervised by Prof. Georg Martius. Her research focuses on curiosity and intrinsically-motivated Reinforcement Learning (RL). Taking inspiration from developmental psychology, she is developing methods to help robots explore their environment efficiently without extrinsic rewards, similar to how children perform free play. Before starting her PhD, she completed her Bachelor’s and Master’s degrees in Electrical Engineering and Information Technology at TU Munich. Her specialization was in Robotics and Automation. During her studies, she worked on various machine learning projects spanning diverse fields such as robotics, signal processing, communications and neuroscience.