Efficient Episodic Control and Virtualization in Reinforcement Learning Agents



Efficient Episodic Control and Virtualization in Reinforcement Learning Agents

16:15 CET, 23 April 2024 | Duration: 2 hours - Online



In this course, participants will delve into the challenge of sample efficiency in Reinforcement Learning (RL), beginning with introducing the issue and exploring innovative solutions grounded in the organisation and function of the mammalian hippocampus and the overall architecture in which it is embedded. We will look specifically at state-of-the- art solutions such as Sequential Episodic Control and Prioritized Replay methods. The curriculum merges theoretical insights with hands-on coding exercises, emphasising the importance of long-term memory and virtualisation mechanisms in decision-making. Learners will creatie RL agents that incorporate advanced concepts such as epistemic reward valuation and episodic control. This course provides a distinctive blend of brain and cognitive science and artificial intelligence, equipping participants with tools to address the sample efficiency problem in RL effectively.



  1. Introductory talk: The sample efficiency problem in Reinforcement
  2. Contextual control: Sequential Episodic Control for sample-efficient decision-
  3. Introductory talk: Hippocampal replay methods in Reinforcement
  4. Virtual control: Model-based Reinforcement learning with hippocampal replay and epistemic reward



Scientific team of CAVAA (EIC Awareness Inside project: Counterfactual Assessment and Valuation for Awareness Architecture) coordinated by Prof. Paul Verschure, Donders Centre for Neuroscience – Neurophysics, Radboud University.

Co-funded by the European Commission. Further info on the CORDIS EU research platform.



Prof. Paul Verschure, (Donders Centre for Neuroscience – Neurophysics, Radboud University)
Ismael T. Freire (Ph.D. candidate in NeuroAI, Donders Centre for Neuroscience – Neurophysics, Radboud University)
Erik Németh (Ph.D. student in NeuroAI, Institut des Systèmes Intelligents et de Robotique – Sorbonne Université)



Researchers (postgraduate); undergraduate students (3rd year or higher) with a background relevant to the topics: Cognitive Sciences, Computational Neurosciences, Cognitive Robotics, Neuro AI, or similar.



  • University degree (or enrollment) in an affiliated field
  • Basic level of Python programming
  • Access to Google Colab.
  • Intermediate proficiency in the English language
  • Interest in the topic and the desire to actively engage during the session


Fill out the registration form and upload your CV and short motivation letter HERE before 30.03.2024. Participation is open (free of charge).

Based on the CV and motivation evaluation, selected participants will be informed via mail within 07.04.2024.



  • Verschure, F. (2016). Synthetic consciousness: the distributed adaptive control perspective. Philosophical Transactions of the Royal Society B: Biological Sciences, 371(1701), 20150448.
  • Freire, T., Amil, A. F., & Verschure, P. F. (2021). Sequential episodic control. arXiv preprint arXiv:2112.14734.
  • Massi, , Barthélemy, J., Mailly, J., Dromnelle, R., Canitrot, J., Poniatowski, E., ... & Khamassi, M. (2022). Model-Based and Model-Free Replay Mechanisms for Reinforcement Learning in Neurorobotics. Frontiers in Neurorobotics, 16, 864380.


For any inquiries, please contact Ismael T. Freire at ismael.freire@donders.ru.nl

Go back