Reggie McLean
@reggiemclean.bsky.social
300 followers 1K following 89 posts
PhD Candidate at Toronto Metropolitan University. Reinforcement learning 🍒, machine learning. He/him. reggiemclean.ca
Posts Media Videos Starter Packs
Pinned
reggiemclean.bsky.social
Proud to announce that Meta-World+ was accepted to NeurIPs, Datasets and Benchmarks! Meta-World is a common benchmark for multi-task and meta-RL research! However, it was very difficult to do effective science with Meta-World as different versions produce different results.
Reposted by Reggie McLean
junlper.beer
it’s really good when “technology of the future” treats you leaving as if you were leaving a cult
Reposted by Reggie McLean
craigweekend.bsky.social
Ladies and gentlemen... the weekend. (also: you are important and are not alone 🧡)
Reposted by Reggie McLean
reggiemclean.bsky.social
I can also vouch for this plan (minus the work visa part) 😅
Reposted by Reggie McLean
simonwheeler.bsky.social
This is absolutely awesome. I am so pleased for everyone involved and particularly for people with Huntington's disease who will benefit.
Also, this is the first example I know of where gene therapy might well prevent dementia. Onwards to FTD!
www.independent.co.uk/news/health/...
Huntington’s treatment slows disease for first time giving hope to families
The condition, which currently has no cure, has been slowed in a groundbreaking trial
www.independent.co.uk
Reposted by Reggie McLean
mariusschneider.bsky.social
🚨Our NeurIPS 2025 competition Mouse vs. AI is LIVE!

We combine a visual navigation task + large-scale mouse neural data to test what makes visual RL agents robust and brain-like.

Top teams: featured at NeurIPS + co-author our summary paper. Join the challenge!

Whitepaper: arxiv.org/abs/2509.14446
Mouse vs. AI: A Neuroethological Benchmark for Visual Robustness and Neural Alignment
Visual robustness under real-world conditions remains a critical bottleneck for modern reinforcement learning agents. In contrast, biological systems such as mice show remarkable resilience to environ...
arxiv.org
Reposted by Reggie McLean
illumi.meme
me telling my grandkids what it was like to have vaccines
The Simpsons old man sitting on a stump telling a story but all the kids have been replaced with headstones
Reposted by Reggie McLean
firepile.bsky.social
Yes but many of us are women and people of color so it didn't count
vortexegg.com
Was this not… what critical AI researchers have been saying this entire time?
mikeyearworth.bsky.social
"In a landmark study, OpenAI researchers reveal that large language models will always produce plausible but false outputs, even with perfect data, due to fundamental statistical and computational limits."

www.computerworld.com/article/4059...
Reposted by Reggie McLean
eugenevinitsky.bsky.social
Meta-world is still very much the go to for multi task and meta-RL research but it has some real failings that Reggie fixes up here!
reggiemclean.bsky.social
Proud to announce that Meta-World+ was accepted to NeurIPs, Datasets and Benchmarks! Meta-World is a common benchmark for multi-task and meta-RL research! However, it was very difficult to do effective science with Meta-World as different versions produce different results.
Reposted by Reggie McLean
pcastr.bsky.social
Very happy this work got into #NeurIPS2025, it was a big effort that Reggie led, and I'm glad it was worth the work!
We've already been using Meta world+ in a lot of our research, and so should you!
Let's chat at #NeurIPS2025 !
reggiemclean.bsky.social
Proud to announce that Meta-World+ was accepted to NeurIPs, Datasets and Benchmarks! Meta-World is a common benchmark for multi-task and meta-RL research! However, it was very difficult to do effective science with Meta-World as different versions produce different results.
reggiemclean.bsky.social
reggiemclean.bsky.social
Proud to announce that Meta-World+ was accepted to NeurIPs, Datasets and Benchmarks! Meta-World is a common benchmark for multi-task and meta-RL research! However, it was very difficult to do effective science with Meta-World as different versions produce different results.
reggiemclean.bsky.social
To further improve the user experience when using Meta-World+, we streamline and improve the experience of creating environments such that users have both full control of the settings used to create their environments, while also ensuring that environments are created in the correct manner.
reggiemclean.bsky.social
To disambiguate the results from the different versions of the benchmark, we implement and re-run experiments across several key multi-task and meta-RL algorithms, on both the V1 and V2 reward functions.
reggiemclean.bsky.social
What has changed between V1 and V2…? The reward functions! Here we find that the per-timestep rewards in all tasks have wildly different scales across the V1 and V2 reward functions. This affects the ability of the value function to approximate the state-action value functions (right panel)
reggiemclean.bsky.social
We can highlight the performance difference between different versions of Meta-World by running a simple algorithm with both the V1 and V2 versions of Meta-World.
reggiemclean.bsky.social
Proud to announce that Meta-World+ was accepted to NeurIPs, Datasets and Benchmarks! Meta-World is a common benchmark for multi-task and meta-RL research! However, it was very difficult to do effective science with Meta-World as different versions produce different results.
Reposted by Reggie McLean
rl-agents-rg.bsky.social
📢 RL reading group TODAY @ 15:00 BST (in 2 hours!) 📢

Speakers: Olya Mastikhina and Dhruv Sreenivas (University of Montreal & Mila - Quebec AI Institute)

Title: Optimistic critics can empower small actors🦸

Details: edinburgh-rl.github.io/reading-group
Reposted by Reggie McLean
marloscmachado.bsky.social
This paper has now been accepted @neuripsconf.bsky.social !

Huge congratulations, Hon Tik (Rick) Tse and Siddarth Chandrasekar.
marloscmachado.bsky.social
📢 I'm happy to share the preprint: _Reward-Aware Proto-Representations in Reinforcement Learning_ ‼️

My PhD student, Hon Tik Tse, led this work, and my MSc student, Siddarth Chandrasekar, assisted us.

arxiv.org/abs/2505.16217

Basically, it's the SR with rewards. See below 👇
Reposted by Reggie McLean
alisabokulich.bsky.social
I haven't used "Academia .edu" for 15 yrs (only tried it when it 1st came out) but they just tried charge me ~$400 automatic on a card w/o any notification or me agreeing to renew/reactivate anything. This is such a predatory scam & they really shouldn't be allowed an ".edu" too Watch out for them
Reposted by Reggie McLean
amyko.phd
This is the experiment we're trying next: reciprocal.reviews. Earn tokens for reviews by writing adequate quality reviews, spend them on submissions. Earn the privilege to publish. We're at 70% on the beta, and should launch an ACM TOCE pilot next year some time.
reciprocal.reviews
Reposted by Reggie McLean
chrislhayes.bsky.social
Not really an overstatement to say that the test of a free society is whether or not comedians can make fun of the country's leader on TV without repurcussions.