TalkRL: The Reinforcement Learning Podcast

Ben Eysenbach is a PhD student in the Machine Learning Department at Carnegie Mellon University.  He was a Resident at Google Brain, and studied math and computer science at MIT. He co-founded the ICML Exploration in Reinforcement Learning workshop.  Featured References Diversity is All You Need: Learning Skills without a Reward Function Benjamin Eysenbach, Abhishek Gupta, Julian Ibarz, Sergey Levine Search on the Replay Buffer: Bridging Planning and Reinforcement Learning Benjamin Eysenbach, Ruslan Salakhutdinov, Sergey Levine Additional References Behaviour Suite for Reinforcement Learning, Ian Osband, Yotam Doron, Matteo Hessel, John Aslanides, Eren Sezener, Andre Saraiva, Katrina McKinney, Tor Lattimore, Csaba Szepesvari, Satinder Singh, Benjamin Van Roy, Richard Sutton, David Silver, Hado Van Hasselt Learning Latent Plans from Play, Corey Lynch, Mohi Khansari, Ted Xiao, Vikash Kumar, Jonathan Tompson, Sergey Levine, Pierre Sermanet Finale Doshi-Velez Emma Brunskill Closed-loop optimization of fast-charging protocols for batteries with machine learning,  Peter Attia, Aditya Grover, Norman Jin, Kristen Severson, Todor Markov, Yang-Hung Liao, Michael Chen, Bryan Cheong, Nicholas Perkins, Zi Yang, Patrick Herring, Muratahan Aykol, Stephen Harris, Richard Braatz, Stefano Ermon, William Chueh CMU 10-703 Deep Reinforcement Learning, Fall 2019, Carnegie Mellon University ICML Exploration in Reinforcement Learning workshop 

Thank you to all the presenters that participated.  I covered as many as I could given the time and crowds, if you were not included and wish to be, please email talkrl@pathwayi.com More details on the official NeurIPS Deep RL Workshop site. 0:23  Approximating two value functions instead of one: towards characterizing a new family of Deep Reinforcement Learning algorithms; Matthia Sabatelli (University of Liege); Gilles Louppe (University of Liège); Pierre Geurts (University of Liège); Marco Wiering (University of Groningen) [external pdf link] 4:16  Single Deep Counterfactual Regret Minimization; Eric Steinberger (University of Cambridge). 5:38  On the Convergence of Episodic Reinforcement Learning Algorithms at the Example of RUDDER; Markus Holzleitner (LIT AI Lab, Institute for Machine Learning, Johannes Kepler University Linz, Austria); José Arjona-Medina (LIT AI Lab, Institute for Machine Learning, Johannes Kepler University Linz, Austria); Marius-Constantin Dinu (LIT AI Lab / University Linz ); Sepp Hochreiter (LIT AI Lab, Institute for Machine Learning, Johannes Kepler University Linz, Austria). 9:33  Objective Mismatch in Model-based Reinforcement Learning; Nathan Lambert (UC Berkeley); Brandon Amos (Facebook); Omry Yadan (Facebook); Roberto Calandra (Facebook). 10:51  Option Discovery using Deep Skill Chaining; Akhil Bagaria (Brown University); George Konidaris (Brown University). 13:44  Blue River Controls: A toolkit for Reinforcement Learning Control Systems on Hardware; Kirill Polzounov (University of Calgary); Ramitha Sundar (Blue River Technology); Lee Reden (Blue River Technology). 14:52  LeDeepChef: Deep Reinforcement Learning Agent for Families of Text-Based Games; Leonard Adolphs (ETHZ); Thomas Hofmann (ETH Zurich). 16:30  Accelerating Training in Pommerman with Imitation and Reinforcement Learning; Hardik Meisheri (TCS Research); Omkar Shelke (TCS Research); Richa Verma (TCS Research); Harshad Khadilkar (TCS Research). 17:27  Dream to Control: Learning Behaviors by Latent Imagination; Danijar Hafner (Google); Timothy Lillicrap (DeepMind); Jimmy Ba (University of Toronto); Mohammad Norouzi (Google Brain) [external pdf link]. 20:48  Adaptive Temperature Tuning for Mellowmax in Deep Reinforcement Learning; Seungchan Kim (Brown University); George Konidaris (Brown). 22:05  Meta-learning curiosity algorithms; Ferran Alet (MIT); Martin Schneider (MIT); Tomas Lozano-Perez (MIT); Leslie Kaelbling (MIT). 24:09  Predictive Coding for Boosting Deep Reinforcement Learning with Sparse Rewards; Xingyu Lu (Berkeley); Stas Tiomkin (BAIR, UC Berkeley); Pieter Abbeel (UC Berkeley). 25:44   Swarm-inspired Reinforcement Learning via Collaborative Inter-agent Knowledge Distillation; Zhang-Wei Hong (Preferred Networks); Prabhat Nagarajan (Preferred Networks); Guilherme Maeda (Preferred Networks). 26:35  Multiplayer AlphaZero; Nicholas Petosa (Georgia Institute of Technology); Tucker Balch (Ga Tech) [external pdf link]. 27:43  Prioritized Sequence Experience Replay; Marc Brittain (Iowa State University); Joshua Bertram (Iowa State University); Xuxi Yang (Iowa State University); Peng Wei (Iowa State University) [external pdf link]. 29:14  Recurrent neural-linear posterior sampling for non-stationary bandits; Paulo Rauber (IDSIA); Aditya Ramesh (USI); Jürgen Schmidhuber (IDSIA - Lugano). 29:36  Improving Evolutionary Strategies With Past Descent Directions; Asier Mujika (ETH Zurich); Florian Meier (ETH Zurich); Marcelo Matheus Gauy (ETH Zurich); Angelika Steger (ETH Zurich) [external pdf link]. 31:40  ZPD Teaching Strategies for Deep Reinforcement Learning from Demonstrations; Daniel Seita (University of California, Berkeley); David Chan (University of California, Berkeley); Roshan Rao (UC Berkeley); Chen Tang (UC Berkeley); Mandi Zhao (UC Berkeley); John Canny (UC Berkeley) [external pdf link]. 33:05  Bottom-Up Meta-Policy Search; Luckeciano Melo (Aeronautics Institute of Technology); Marcos Máximo (Aeronautics Institute of Technology); Adilson Cunha (Aeronautics Institute of Technology) [external pdf link]. 33:37  MERL: Multi-Head Reinforcement Learning; Yannis Flet-Berliac (University of Lille / Inria); Philippe Preux (INRIA) [external pdf link]. 35:30  Emergen...

Scott Fujimoto is a PhD student at McGill University and Mila. He is the author of TD3 as well as some of the recent developments in batch deep reinforcement learning.  Featured References Addressing Function Approximation Error in Actor-Critic Methods Scott Fujimoto, Herke van Hoof, David Meger Off-Policy Deep Reinforcement Learning without Exploration Scott Fujimoto, David Meger, Doina Precup Benchmarking Batch Deep Reinforcement Learning Algorithms Scott Fujimoto, Edoardo Conti, Mohammad Ghavamzadeh, Joelle Pineau Additional References Striving for Simplicity in Off-Policy Deep Reinforcement Learning Rishabh Agarwal, Dale Schuurmans, Mohammad Norouzi Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, Sergey Levine Way Off-Policy Batch Deep Reinforcement Learning of Implicit Human Preferences in Dialog Natasha Jaques, Asma Ghandeharioun, Judy Hanwen Shen, Craig Ferguson, Agata Lapedriza, Noah Jones, Shixiang Gu, Rosalind Picard Continuous control with deep reinforcement learning Timothy P. Lillicrap, Jonathan J. Hunt, Alexander Pritzel, Nicolas Heess, Tom Erez, Yuval Tassa, David Silver, Daan Wierstra Distributed Distributional Deterministic Policy Gradients Gabriel Barth-Maron, Matthew W. Hoffman, David Budden, Will Dabney, Dan Horgan, Dhruva TB, Alistair Muldal, Nicolas Heess, Timothy Lillicrap 

Dr. Jessica Hamrick is a Research Scientist at DeepMind. She holds a PhD in Psychology from UC Berkeley. Featured References Structured agents for physical construction Victor Bapst, Alvaro Sanchez-Gonzalez, Carl Doersch, Kimberly L. Stachenfeld, Pushmeet Kohli, Peter W. Battaglia, Jessica B. Hamrick Analogues of mental simulation and imagination in deep learning Jessica Hamrick Additional References Metacontrol for Adaptive Imagination-Based Optimization Jessica B. Hamrick, Andrew J. Ballard, Razvan Pascanu, Oriol Vinyals, Nicolas Heess, Peter W. Battaglia Surprising Negative Results for Generative Adversarial Tree Search  Kamyar Azizzadenesheli, Brandon Yang, Weitang Liu, Zachary C Lipton, Animashree Anandkumar Metareasoning and Mental Simulation Jessica B. Hamrick Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm David Silver, Thomas Hubert, Julian Schrittwieser, Ioannis Antonoglou, Matthew Lai, Arthur Guez, Marc Lanctot, Laurent Sifre, Dharshan Kumaran, Thore Graepel, Timothy Lillicrap, Karen Simonyan, Demis Hassabis Object-oriented state editing for HRL Victor Bapst, Alvaro Sanchez-Gonzalez, Omar Shams, Kimberly Stachenfeld, Peter W. Battaglia, Satinder Singh, Jessica B. Hamrick FeUdal Networks for Hierarchical Reinforcement Learning Alexander Sasha Vezhnevets, Simon Osindero, Tom Schaul, Nicolas Heess, Max Jaderberg, David Silver, Koray Kavukcuoglu PILCO: A Model-Based and Data-Efficient Approach to Policy Search Marc Peter Deisenroth, Carl Edward Rasmussen Blueberry Earth Anders Sandberg 

Dr Pablo Samuel Castro is a Staff Research Software Engineer at Google Brain.  He is the main author of the Dopamine RL framework. Featured References A Comparative Analysis of Expected and Distributional Reinforcement Learning Clare Lyle, Pablo Samuel Castro, Marc G. Bellemare  A Geometric Perspective on Optimal Representations for Reinforcement Learning Marc G. Bellemare, Will Dabney, Robert Dadashi, Adrien Ali Taiga, Pablo Samuel Castro, Nicolas Le Roux, Dale Schuurmans, Tor Lattimore, Clare Lyle Dopamine: A Research Framework for Deep Reinforcement Learning Pablo Samuel Castro, Subhodeep Moitra, Carles Gelada, Saurabh Kumar, Marc G. Bellemare Dopamine RL framework on github  Tensorflow Agents on github Additional References Using Linear Programming for Bayesian Exploration in Markov Decision Processes Pablo Samuel Castro, Doina Precup Using bisimulation for policy transfer in MDPs Pablo Samuel Castro, Doina Precup Rainbow: Combining Improvements in Deep Reinforcement Learning Matteo Hessel, Joseph Modayil, Hado van Hasselt, Tom Schaul, Georg Ostrovski, Will Dabney, Dan Horgan, Bilal Piot, Mohammad Azar, David Silver Implicit Quantile Networks for Distributional Reinforcement Learning Will Dabney, Georg Ostrovski, David Silver, Rémi Munos A Distributional Perspective on Reinforcement Learning Marc G. Bellemare, Will Dabney, Rémi Munos 

Dr. Kamyar Azizzadenesheli is a post-doctorate scholar at Caltech.  His research interest is mainly in the area of Machine Learning, from theory to practice, with the main focus in Reinforcement Learning.  He will be joining Purdue University as an Assistant CS Professor in Fall 2020. Featured References Efficient Exploration through Bayesian Deep Q-Networks Kamyar Azizzadenesheli, Animashree Anandkumar Surprising Negative Results for Generative Adversarial Tree Search Kamyar Azizzadenesheli, Brandon Yang, Weitang Liu, Zachary C Lipton, Animashree Anandkumar Maybe a few considerations in Reinforcement Learning Research? Kamyar Azizzadenesheli  Additional References Model-Based Reinforcement Learning for Atari  Lukasz Kaiser, Mohammad Babaeizadeh, Piotr Milos, Blazej Osinski, Roy H Campbell, Konrad Czechowski, Dumitru Erhan, Chelsea Finn, Piotr Kozakowski, Sergey Levine, Afroz Mohiuddin, Ryan Sepassi, George Tucker, Henryk Michalewski Near-optimal Regret Bounds for Reinforcement Learning Thomas Jaksch, Ronald Ortner, Peter Auer Curious Model-Building Control Systems Jürgen Schmidhuber Rainbow: Combining Improvements in Deep Reinforcement Learning  Matteo Hessel, Joseph Modayil, Hado van Hasselt, Tom Schaul, Georg Ostrovski, Will Dabney, Dan Horgan, Bilal Piot, Mohammad Azar, David Silver Schema Networks: Zero-shot Transfer with a Generative Causal Model of Intuitive Physics Ken Kansky, Tom Silver, David A. Mély, Mohamed Eldawy, Miguel Lázaro-Gredilla, Xinghua Lou, Nimrod Dorfman, Szymon Sidor, Scott Phoenix, Dileep George Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm David Silver, Thomas Hubert, Julian Schrittwieser, Ioannis Antonoglou, Matthew Lai, Arthur Guez, Marc Lanctot, Laurent Sifre, Dharshan Kumaran, Thore Graepel, Timothy Lillicrap, Karen Simonyan, Demis Hassabis 

Antonin Raffin is a researcher at the German Aerospace Center (DLR) in Munich, working in the Institute of Robotics and Mechatronics. His research is on using machine learning for controlling real robots (because simulation is not enough), with a particular interest for reinforcement learning. Ashley Hill is doing his thesis on improving control algorithms using machine learning for real time gain tuning. He works mainly with neuroevolution, genetic algorithms, and of course reinforcement learning, applied to mobile robots.  He holds a masters degree in Machine learning, and a bachelors in Computer science from the Université Paris-Saclay. Featured References stable-baselines on github Ashley Hill, Antonin Raffin primary authors. S-RL Toolbox Antonin Raffin, Ashley Hill, René Traoré, Timothée Lesort, Natalia Díaz-Rodríguez, David Filliat Decoupling feature extraction from policy learning: assessing benefits of state representation learning in goal based robotics Antonin Raffin, Ashley Hill, René Traoré, Timothée Lesort, Natalia Díaz-Rodríguez, David Filliat Additional References Learning to Drive Smoothly in Minutes, Antonin Raffin Multimodal SRL (best paper at ICRA): Making Sense of Vision and Touch: Self-Supervised Learning of Multimodal  Representations for Contact-Rich Tasks,  Michelle A. Lee, Yuke Zhu, Krishnan Srinivasan, Parth Shah, Silvio Savarese, Li Fei-Fei, Animesh Garg, Jeannette Bohg Benchmarking Model-Based Reinforcement Learning, Tingwu Wang, Xuchan Bao, Ignasi Clavera, Jerrick Hoang, Yeming Wen, Eric Langlois, Shunshi Zhang, Guodong Zhang, Pieter Abbeel, Jimmy Ba TossingBot: Learning to Throw Arbitrary Objects with Residual Physics Andy Zeng, Shuran Song, Johnny Lee, Alberto Rodriguez, Thomas Funkhouser Stable Baselines roadmap OpenAI baselines stable-baselines github pull request 

Michael L Littman is a professor of Computer Science at Brown University.  He was elected ACM Fellow in 2018 "For contributions to the design and analysis of sequential decision making algorithms in artificial intelligence". Featured References Convergent Actor Critic by Humans James MacGlashan, Michael L. Littman, David L. Roberts, Robert Tyler Loftin, Bei Peng, Matthew E. Taylor People teach with rewards and punishments as communication, not reinforcements Mark Ho, Fiery Cushman, Michael L. Littman, Joseph Austerweil Theory of Minds: Understanding Behavior in Groups Through Inverse Planning Michael Shum, Max Kleiman-Weiner, Michael L. Littman, Joshua B. Tenenbaum Personalized education at scale Saarinen, Cater, Littman Additional References Michael Littman papers on Google Scholar, Semantic Scholar Reinforcement Learning on Udacity, Charles Isbell, Michael Littman, Chris Pryby  Machine Learning on Udacity, Michael Littman, Charles Isbell, Pushkar Kolhe  Temporal Difference Learning and TD-Gammon, Gerald Tesauro Playing Atari with Deep Reinforcement Learning, Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou, Daan Wierstra, Martin Riedmiller Ask Me Anything about MOOCs, D Fisher, C Isbell, ML Littman, M Wollowski, et al Reinforcement Learning and Decision Making (RLDM) Conference Algorithms for Sequential Decision Making, Michael Littman's Thesis Machine Learning A Cappella - Overfitting Thriller!, Michael Littman and Charles Isbell feat Infinite Harmony Turbotax Ad 2016: Genius Anna/Michael Littman 

Natasha Jaques is a PhD candidate at MIT working on affective and social intelligence.  She has interned with DeepMind and Google Brain, and was an OpenAI Scholars mentor.  Her paper “Social Influence as Intrinsic Motivation for Multi-Agent Deep Reinforcement Learning” received an honourable mention for best paper at ICML 2019. Featured References Social Influence as Intrinsic Motivation for Multi-Agent Deep Reinforcement Learning Natasha Jaques, Angeliki Lazaridou, Edward Hughes, Caglar Gulcehre, Pedro A. Ortega, DJ Strouse, Joel Z. Leibo, Nando de Freitas Tackling climate change with Machine Learning David Rolnick, Priya L. Donti, Lynn H. Kaack, Kelly Kochanski, Alexandre Lacoste, Kris Sankaran, Andrew Slavin Ross, Nikola Milojevic-Dupont, Natasha Jaques, Anna Waldman-Brown, Alexandra Luccioni, Tegan Maharaj, Evan D. Sherwin, S. Karthik Mukkavilli, Konrad P. Kording, Carla Gomes, Andrew Y. Ng, Demis Hassabis, John C. Platt, Felix Creutzig, Jennifer Chayes, Yoshua Bengio Additional References MIT Media Lab Flight Offsets,  Caroline Jaffe, Juliana Cherston, Natasha Jaques Modeling Others using Oneself in Multi-Agent Reinforcement Learning, Roberta Raileanu, Emily Denton, Arthur Szlam, Rob Fergus Inequity aversion improves cooperation in intertemporal social dilemmas,  Edward Hughes, Joel Z. Leibo, Matthew G. Phillips, Karl Tuyls, Edgar A. Duéñez-Guzmán, Antonio García Castañeda, Iain Dunning, Tina Zhu, Kevin R. McKee, Raphael Koster, Heather Roff, Thore Graepel Sequential Social Dilemma Games on github, Eugene Vinitsky, Natasha Jaques  AI Alignment newsletter, Rohin Shah Paired Open-Ended Trailblazer (POET): Endlessly Generating Increasingly Complex and Diverse Learning Environments and Their Solutions, Rui Wang, Joel Lehman, Jeff Clune, Kenneth O. Stanley The social function of intellect, Nicholas Humphrey Autocurricula and the Emergence of Innovation from Social Interaction: A Manifesto for Multi-Agent Intelligence Research, Joel Z. Leibo, Edward Hughes, Marc Lanctot, Thore Graepel A Recipe for Training Neural Networks, Andrej Karpathy Emotionally Adaptive Intelligent Tutoring Systems using POMDPs, Natasha Jaques Sapiens, Yuval Noah Harari 

August 2, 2019 Transcript The idea with TalkRL Podcast is to hear from brilliant folks from across the world of Reinforcement Learning, both research and applications.  As much as possible, I want to hear from them in their own language.  I try to get to know as much as I can about their work before hand.  And Im not here to convert anyone, I want to reach people who are already into RL.  So we wont stop to explain what a value function is, for example.  Though we also wont assume everyone has read the very latest papers. Why am I doing this? Because it’s a great way to learn from the most inspiring people in the field!  There’s so much happening in the universe of RL, and there’s tons of interesting angles and so many fascinating minds to learn from. Now I know there is no shortage of books, papers, and lectures, but so much goes unsaid. I mean I guess if you work at MILA or AMII or Vector Institute, you might be having these conversations over coffee all the time, but I live in a little village in the woods in BC, so for me, these remote interviews are like a great way to have these conversations, and I hope sharing with the community makes it more worthwhile for everyone. In terms of format, the first 2 episodes were interviews in longer form, around an hour long.  Going forward, some may be a lot shorter, it depends on the guest. If you want want to be a guest or suggest a guest, goto talkrl.com/about, you will find a link to a suggestion form. Thanks for listening!