Very excited to have contributed to this paper @icmlconf lead by @DikshantShehmar!
We explore if Laplacian representations could be a linchpin for planning with learned models. This work could unlock the data efficient future often promised from model-based planning!
Here's another paper we'll be presenting at ICML'26 - @icmlconf.
Laplacian Representations for Decision-Time Planning
https://t.co/0UEIuDxS47
This work was led by @DikshantShehmar, and he'll be presenting on 8th July at 2 pm, Korea time.
This video summarizes the idea well โ
At #NeurIPS2025 come check out our poster in the TCCAI workshop at 4:00pm on Sunday! In this preliminary work, we are thinking about how to apply various resolution-invariant models to power grid data to accelerate simulations with renewables!
https://t.co/OOPtivXe9k
Hey! @modl_ai is pretty cool and we are doing really cool things! This is a brilliant opportunity to do a PhD with Modl and work with some really fantastic researchers! If you have any questions feel free to reach out!
Happy to announce we have an industrial PhD position available at @ITUkbh in collaboration with @modl_ai on improving automated video game testing via machine learning ๐ฎ๐๐ค You'll be advised by me with support from @togelius and @yannakakis. Ping me if interested!
๐A warm welcome to Matthew to modl! He is our latest addition to our Technical Staff Team in Research with a wealth of knowledge in RL and Deep Learning, with a particular focus on agent perception. We're absolutely delighted to have you with us, @mattschleg! ๐
To be brief: The paper provides new ideas on how to evaluate recurrent agents in the reinforcement learning domain, future potential directions in designing recurrent architectures for continual decision making (i.e. RL), and a ton more including the appendix!
New paper published at TMLR in collaboration with Volodmyr Tkachuk, Adam White, and @white_martha! Done all in the @rlai_lab and @AmiiThinks labs!
Check it out: https://t.co/YowrwGF8KT
To be brief, we do a deep empirical investigation on a single axis of choice in RNNs for RL.
On the job market! Looking for Post Docs and Research Scientist positions that will broaden my views on RL and machine intelligence! My research is centered on understanding and developing multi-prediction systems and agent-state construction in RL.
CV: https://t.co/ICGtY5pVmw
just made a "decentralized" "alternative" to twitter; everyone should go "join" it
to make an account: fork https://t.co/vv271yPEMr
to tweet: git commit --allow-empty
to follow someone: git remote add <alias> <their fork url>
to retweet: git cherry-pick <their "tweet">
To those submitting papers to venues that allow for paper updates in response to reviews: Please clearly mark what edits are made from the original version so reviewers can easily spot changes and provide more feedback. I mark all text in red that is different from the original.
@SoloGen I've used Julia for all my projects for the past 4.5 years (previously C++ and some Python). Still a lot to improve on in terms of usability (startup time, package maturity, linting). But the low level control, multiple dispatch, and language focus on numerics is worth it to me.
I'm starting to write my actual thesis document. Awhile ago, I saw someone work on this in public (I can't remember who unfortunately, but do tag them in the comments!) and wanted to also do this. You will be able to see my progress at https://t.co/c9qtpwXXhX.