Log inSign up
Greg Farquhar
25 posts
@greg_far

Greg Farquhar

@greg_far
Joined October 2017
95
Following
455
Followers
RepliesRepliesRepostsRepostsMediaMedia

Log in or sign up for X

See what’s happening and join the conversation

Continue with phone
or
Log in with username or email
Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
  • Pinned
    @greg_far
    Greg Farquhar
    @greg_far
    Jul 15, 2021
    There’s huge potential in using ‘demonstrations’ from other agents with different goals: to understand which features & dynamics of the environment *might* be important to you; and to borrow from others' behaviours only where they are useful for you.
    @filangelos
    Angelos Filos
    @filangelos
    Jul 15, 2021
    👽 PsiPhi-learning 👽 (long talk #ICML) sites.google.com/view/psiphi-le… shows how an agent can use data from the behavior of other agents with diverse goals: to infer their intentions and fulfill its own! 🧵
    1
  • @greg_far
    Greg Farquhar
    @greg_far
    Dec 3, 2025
    This was a great project to work on. Happy to have it published now in @Nature! Meta-learning is important.
    @junh_oh
    Junhyuk Oh
    @junh_oh
    Dec 2, 2025
    Excited to announce that our work on “Discovering state-of-the-art RL algorithms” is finally published in @Nature! In this work, we meta-learned RL algorithms at scale. Paper: nature.com/articles/s4158… Blog: google-deepmind.github.io/disco_rl/ See thread 👇
  • @greg_far
    Greg Farquhar
    @greg_far
    Jun 22, 2020
    Permanent damage to generalisation from early updates in non-stationary training -- really enjoyed looking into this intriguing problem and trying to solve it for deep RL agents!
    @MaxiIgl
    Maximilian Igl
    @MaxiIgl
    Jun 22, 2020
    Really excited about our new work: In deep RL, we typically collect new data using a non-stationary policy that gets updated as we learn and improve. We show this can impact the learning dynamics of our deep policy and lead to worse generalization arxiv.org/abs/2006.05826 (1/7)
    Image
    Image
  • @greg_far
    Greg Farquhar
    @greg_far
    May 5, 2020
    This is awesome, but I'm a little scared of how much time I might spend playing it myself...
    @_rockt
    Tim Rocktäschel
    Recursive
    @_rockt
    May 5, 2020
    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/
    Image
    GIF
  • @greg_far
    Greg Farquhar
    @greg_far
    Mar 20, 2020
    I particularly enjoyed visualising & analysing the learned mixing functions that combine per-agent utilities into joint values!
    Image
    @_samvelyan
    Mikayel Samvelyan
    @_samvelyan
    Mar 20, 2020
    Happy to share the extended version of our #QMIX paper “Monotonic Value Function Factorisation for Deep Multi-Agent RL” We include further analysis and ablation studies that investigate how monotonic factorisation of joint Q-val helps QMIX outperform VDN arxiv.org/abs/2003.08839
Advertisement
Advertisement