Greg Farquhar retweeted
I am proud to announce the release of the NetHack Learning Environment (NLE)! NetHack is an extremely difficult procedurally-generated grid-world dungeon-crawl game that strikes a great balance between complexity and speed for single-agent reinforcement learning research. 1/
Greg Farquhar retweeted
Tuomas Sandholm and I are doing a Reddit AMA now on the #Pluribus poker AI!
reddit.com/r/MachineLearning…
Greg Farquhar retweeted
How can RL agents exploit the compositional, relational and hierarchical structure of the world? A growing number of authors propose learning from natural language. We are excited to share our @IJCAIconf survey of this emerging field! arxiv.org/abs/1906.03926
TL;DR:🤖+📖=📈🎯🏆🥳
Greg Farquhar retweeted
I had the pleasure to co-supervise outstanding MSc students jointly with Jakob Foerster (@j_foerst) and Greg Farquhar (@greg_far) at @CompSciOxford this year. Together, we compiled our advice for embarking on short-term machine learning research projects: rockt.github.io/2018/08/29/m…
Greg Farquhar retweeted
I am very excited to share our ICML paper “Deep Variational Reinforcement Learning (DVRL) for POMDPs”: Our agent learns a model of the environment and acts based on its belief state in this model. w/ @zinmalu @tuananhle7 @frankdonaldwood @shimon8282 arxiv.org/abs/1806.02426
Greg Farquhar retweeted
Our latest paper: how to learn complex joint value functions for teams of agents whose greedy policies can be computed and executed in a decentralised fashion. The trick is a new monotonic value function factorisation. With results on StartCraft 2! arxiv.org/abs/1803.11485
Greg Farquhar retweeted
Excited to share "DiCE: The Infinitely Differentiable Monte Carlo Estimator": arxiv.org/abs/1802.05098 Try this one weird objective for correct any-order gradient estimators in all your stochastic graphs ;) With fantastic Oxford/CMU team: @greg_far @alshedivat @_rockt @shimon8282
