Taylor W. Killian

CoCoMARL 2025 Stimulated by the increasing complexity of real-world problems, multi-agent reinforcement learning (MARL) has emerged as a promising approach for enabling autonomous agents to learn and adapt in dynam...

There has been a multi-agent workshop the past two years!

Here's last year's workshop website: sites.google.com/view/cocomar...

The workshops are maybe the best part of RLC. Bring us your workshops that could never happen anywhere else!

3 weeks to get your RLC papers in shape! And to get the word out to those who have yet to experience an RLC review process.

🚀 Excited to share REPPO, a new on-policy RL agent!

TL;DR: Replace PPO with REPPO for fewer hyperparameter headaches and more robust training.

REPPO, led by @cvoelcker.bsky.social, will be presented at ICLR 2026. How does it work? 🧵👇

Please don't hesitate to reach out with questions to me directly or via email to workshops@rl-conference.cc

The OpenReview submission site can be found at:
openreview.net/group?id=rl-...

Workshops, held on the first day, are a primary feature of
@rl-conference.bsky.social and have set a delightfully inquisitive tone. Discussions started at both workshops I've been a part of in the past persisted through the week, with some still ongoing!

CfW here: rl-conference.cc/call_for_wor...

We're thrilled to share that the Call for Workshops for this year's @rl-conference.bsky.social is now live!

As Workshop co-chair (alongside the wonderful Raksha Kumaraswamy and @claireve.bsky.social) we are looking forward to seeing the proposals for workshops that we receive.

LINK IN NEXT POST

One paper accepted to ICML with one paper rejected as well. It’s called balance (and measured frustration) #icml2025

Shipping a copy of Kuhn's "Structure of Scientific Revolutions" to every ICML meta-reviewer

Great question! To be honest, no idea. Part of the research design is to identify the objectives we want to hit and these aren’t fully fleshed out just yet.

Multi-agent in terms of mixed autonomy type of environments (or training a LLM to gather information from other agents, human or artificial) is somewhat along the path of our future strategic directions for the US-SV office.

Large behavioral sims or specific applied focus areas are likely not.

We're hiring at all levels for engineering and research roles with focus areas in CV, LLMs, and RL (broadly defined). We are firmly committed to open science and have extensive computational resources. Check us out, see you in Singapore! ifm.mbzuai.ac.ae

We're putting in final planning and preparation steps this week to be ready for #ICLR2025. I'm excited to be attending, in part to help represent @mbzuai.bsky.social and @llm360.bsky.social as well as recruit Research Scientists and Engineers to our new lab in the Bay Area! (see next for more info)

Thank you for the kind consideration. I am however no longer in Canada (left during the COVID pandemic to follow my advisor to MIT, and now find myself in the SF Bay Area and missing Massachusetts and Ontario a whole lot...)

I love this idea. Really and truly. Let's do this @icmlconf.bsky.social! (Maybe also something @rl-conference.bsky.social could look into?)

Wow! Congratulations to both BU and you!

I used to dream of days like this.

Mowed the lawn, trimmed some hedges, pruned a few trees to let more sun into our yard, picked the rest of our orange tree (we yielded 658 🍊this year!) had the kids help clean up the branches, etc. #suburbandadsaturdays

Real-world RL for the public good! Love to see it!

Real-World Deployment and Assessment of a Multi-Agent Reinforcement Learning-Based Variable Speed Limit Control System This article presents the first field deployment of a multi-agent reinforcement learning (MARL) based variable speed limit (VSL) control system on Interstate 24 (I-24) near Nashville, Tennessee. We de...

A year later of a multi-agent RL controlled variable speed limit system:
arxiv.org/abs/2503.01017
Fewer crashes, quicker responses, quicker warnings

Reviewing a paper right now and I'm having lots of "I wish that I'd written this!" feelings. Great sign for the authors tbh.

Was about to ask if this was how you were doing your ICML reviewing ;)

Congratulations Andy and Rich! You've given RL yet another place in the history books. I can't imagine how you're feeling right now, but it must be amazing to reflect on how far the ideas have come, especially after all the care and dedication you gave them.

awards.acm.org/about/2024-t...

Think of LLM Applications as POMDPs — Not Agents · TensorZero Think of LLM Applications as POMDPs — Not Agents

As I've started down the RL+LLM rabbit hole (for reals this time), this blog post by TensorZero is absolutely compelling. It's nice to see clear thinking conceptual framing that at once open the doors to new research but also leverages existing insights:
www.tensorzero.com/blog/think-o...

a man with a beard is saying he can 't keep getting away with it ! ALT: a man with a beard is saying he can 't keep getting away with it !

Struggled to find a paper to recommend to authors that would strengthen their claims while writing a review. Used @GeminiApp and @youdotcom to no success. Reluctantly pulled up @ChatGPTapp, nailed it in the first returned result. #HesitantAboutLLMsButTryingToLearn

"Scaling laws" tries to piggyback on physics where the scaling laws are resultant from statistical limits. To call them "laws" rather than "empirical scaling curves" is real iffy

I’ve got a lot of thoughts about the current market as an ML researcher, not sure I’ll share them in total.

But… it was more difficult than anticipated and I felt hopeless a lot of the time. If you’re feeling similarly; know you’re not worthless, it just may take more time. 🤗

Some additional information (and job postings) can be found here: mbzuai.ac.ae/institute-of-f…

Let me know if you want to learn more about IFM, why I chose this opportunity, or chat about open access research.

Day 1 with @mbzuai.bsky.social as we start building out a new non-profit, open research lab in the Bay Area. Lots to learn and my head is spinning but I’m excited to get started pushing the boundaries of what we know about reasoning and decision making under uncertainty.

I’m so glad that you found this paper. Lots of really fun implications and avenues to explore, both in applied and foundational directions.

Oh man, i wish that I had a concrete answer here. I do recall discussions about evaluations over all aspects of nuPlan but don’t remember what results we had.