TalkRL: The Reinforcement Learning Podcast

David Silver, a principal research scientist at DeepMind and a professor at UCL, dives deep into the evolution of reinforcement learning. He discusses the fascinating transition of AlphaFold from RL to supervised learning for protein folding and highlights RL's potential in protein design. Silver also reflects on how personal health impacts research output and shares insights on AlphaZero's learning strategies in various games. He encourages aspiring researchers to embrace boldness in their endeavors and sketches his journey towards advancing artificial general intelligence.

Vincent Moens, Applied ML Research Scientist at Meta and author of TorchRL, discusses the design philosophy and challenges in creating a versatile reinforcement learning library. He also shares his research journey from medicine to ML, evolution of RL perceptions in the AI community, and encourages active engagement in the open-source community.

Arash Ahmadian discusses preference training in language models, exploring methods like PPO. The podcast dives into reinforced leave one out method, reinforced vs vanilla policy gradient in deep RL, and token-level actions. Reward structures and optimization techniques in RLHF are also explored, emphasizing the importance of curated reward signals.

Glen Berseth is an assistant professor at the Université de Montréal, a core academic member of the Mila - Quebec AI Institute, a Canada CIFAR AI chair, member l'Institute Courtios, and co-director of the Robotics and Embodied AI Lab (REAL).  Featured Links  Reinforcement Learning Conference  Closing the Gap between TD Learning and Supervised Learning--A Generalisation Point of View Raj Ghugare, Matthieu Geist, Glen Berseth, Benjamin Eysenbach

A Research scientist at OpenAI discusses information theory and RL, joint predictions, and Epistemic Neural Networks. They explore challenges in reinforcement learning, handling uncertainty, and balancing exploration vs exploitation. The podcast delves into the importance of joint predictive distributions, Thompson sampling approximation, and uncertainty frameworks in Large Language Models (LLMs).

Sharath Chandra Raparthy, an AI Resident at FAIR at Meta, discusses in-context learning for sequential decision tasks, training models to adapt to unseen tasks and randomized environments, properties of data for in-context learning, burstiness and trajectories in transformers, and the use of G flow nets in sampling from complex distributions.

Pierluca D'Oro and Martin Klissarov discuss their recent work on 'Motif, Intrinsic Motivation from AI Feedback' and its application in NetHack. They also explore the similarities between RL and Learning from Preferences, the challenges of training an RL agent for NetHack, the gap between RL and language models, and the difference between return and loss landscapes in RL.

Martin Riedmiller, a research scientist and team lead at DeepMind, discusses using reinforcement learning to control the magnetic field in a fusion reactor. They explore challenges in the TOCOMAC project, reward design, designing actor and critic networks, DQN and NFQ algorithms, the importance of explainability in RL systems, and the horde architecture for collecting experience.

Max Schwarzer is a PhD student at Mila, with Aaron Courville and Marc Bellemare, interested in RL scaling, representation learning for RL, and RL for science.  Max spent the last 1.5 years at Google Brain/DeepMind, and is now at Apple Machine Learning Research.   Featured References Bigger, Better, Faster: Human-level Atari with human-level efficiency  Max Schwarzer, Johan Obando-Ceron, Aaron Courville, Marc Bellemare, Rishabh Agarwal, Pablo Samuel Castro  Sample-Efficient Reinforcement Learning by Breaking the Replay Ratio Barrier Pierluca D'Oro, Max Schwarzer, Evgenii Nikishin, Pierre-Luc Bacon, Marc G Bellemare, Aaron Courville  The Primacy Bias in Deep Reinforcement Learning Evgenii Nikishin, Max Schwarzer, Pierluca D'Oro, Pierre-Luc Bacon, Aaron Courville  Additional References   Rainbow: Combining Improvements in Deep Reinforcement Learning, Hessel et al 2017  When to use parametric models in reinforcement learning? Hasselt et al 2019 Data-Efficient Reinforcement Learning with Self-Predictive Representations, Schwarzer et al 2020  Pretraining Representations for Data-Efficient Reinforcement Learning, Schwarzer et al 2021  

Julian Togelius is an Associate Professor of Computer Science and Engineering at NYU, and Cofounder and research director at modl.ai  Featured References  Choose Your Weapon: Survival Strategies for Depressed AI AcademicsJulian Togelius, Georgios N. YannakakisLearning Controllable 3D Level GeneratorsZehua Jiang, Sam Earle, Michael Cerny Green, Julian TogeliusPCGRL: Procedural Content Generation via Reinforcement LearningAhmed Khalifa, Philip Bontrager, Sam Earle, Julian TogeliusIlluminating Generalization in Deep Reinforcement Learning through Procedural Level GenerationNiels Justesen, Ruben Rodriguez Torrado, Philip Bontrager, Ahmed Khalifa, Julian Togelius, Sebastian Risi