cogscikid.bsky.social

cogscikid.bsky.social

@cogscikid.bsky.social

Kempner Institute research fellow @Harvard I want to build AI assistants for the human reinforcement learning algorithm. prev: deepmind, umich, msr https://cogscikid.com/

305 followers85 following103 posts

Pinned

cogscikid.bsky.social
@cogscikid.bsky.social

Task diversity is supposedly key to generalization in RL. But what does it do to continual RL, where agents face one new task distribution after another? We find that past a point, more diversity actually inhibits continual reinforcement learning 🧵

cogscikid.bsky.social reposted

RLDM
@rldmparis2027.bsky.social

📣 Save the date! RLDM 2027, the Multi-disciplinary Conference on Reinforcement Learning and Decision Making, is coming to Paris, July 6-9, 2027 🎉 Stay tuned for submission deadlines and more updates on our website rldm.org 👀

cogscikid.bsky.social reposted

Markus Brauer
@markusbrauer.bsky.social

The UW–Madison Department of Psychology is seeking to fill an open-rank faculty position in social psychology. If you know a great candidate, encourage them to apply! jobs.wisc.edu/jobs/assista...

jobs.wisc.eduAssistant Professor, Associate Professor, or Professor of Psychology - Madison, Wisconsin, United StatesCurrent Employees: If you are currently employed at any of the Universities of Wisconsin, log in to Workday to apply through the internal application process.Job Category:FacultyEmployment Type:Regula...

cogscikid.bsky.social reposted

Sam Nastase
@samnastase.bsky.social

We’re recruiting a full-time lab manager to join the Shared Minds Lab at USC! This will be a great opportunity for someone who wants to get hands-on experience with research before starting a PhD program in psychology or neuroscience. More below:

cogscikid.bsky.social reposted

Megan deBettencourt
@mdebettencourt.bsky.social

Come work with me at Google! We have an open 6-month temp role on my team - it’s a great opportunity for someone with a cog neuro or psych background interested in exploring big tech. Looking for comp modeling experience (e.g. RL, DDMs) and Python/JS. Reach out to learn more or pass along any names!

cogscikid.bsky.social reposted

Jamal A. Williams
@jayneuro.bsky.social

Our new paper is out in Nature Communications! 🎶🧠✨ If you're interested in how the brain represents meaning evoked by music, check it out. Shout out to my Princeton team for making this happen!🚀 @jalon.bsky.social, @margulisa.bsky.social, Karen Christianson, Cara Turnbull, Grace Simmons, Uri Hasson

Itamar Jalon@jalon.bsky.social

Play people instrumental music they've never heard, and they independently imagine the same story. Play them that story in words instead, and the same brain regions light up in the same pattern. Our new paper is out today in @natcomms.nature.com

cogscikid.bsky.social reposted

Sam Gershman
@gershbrain.bsky.social

@arthurpr4t.bsky.social has a new preprint with important results on a famous psychophysical law (Weber's law). It isn't, in fact, a law, because it can be broken. A more fundamental principle (efficient coding) shows when and why Weber's law holds true. www.biorxiv.org/content/10.6...

biorxiv.orgEfficient coding makes and breaks Weber's lawWeber's law is a rare quantitative regularity in psychology, yet its origins remain debated. Here we provide causal evidence that it arises from the more fundamental principle of efficient coding. Thi...

cogscikid.bsky.social reposted

Black In Neuro
@blackinneuro.bsky.social

Applications are open for the 2026–27 Black In Neuro Mentorship Cohort! 🧠✨ Apply as a mentor or mentee for 1:1 matching, application support, professional development & a global neuro community. 📲 Scan the QR code to apply! #BlackInNeuro

cogscikid.bsky.social reposted

Jack Mirkinson
@jackmirkinson.bsky.social

Today in @thenation.com: a harrowing piece written from inside the Dilley ICE detention center by Maryam Tahmasebi, a green card holder who's been locked up along with her husband and son for months despite none of them committing any crime. www.thenation.com/article/soci...

thenation.comOne Woman’s Message From the Hell of ICE DetentionMy husband, son, and I have been locked up for months—despite being permanent residents with no criminal record. It is an unendurable, unending nightmare.
cogscikid.bsky.social
@cogscikid.bsky.social

naturalistic cog comp neuro lab! sounds cool!

Hayoung Song@hayoungsong.bsky.social

The Naturalistic Cognitive Computational Neuroscience Lab is launching at UT Austin and is looking for founding members at all levels, including lab manager, PhD students, and postdocs! #NeuroJobs We will study human brain🧠 and memory & attention, using fMRI and modeling. thesonglab.github.io

cogscikid.bsky.social reposted

Natalia Vélez
@natvelali.bsky.social

It feels strange to share good news, but: I’m moving to UCSD in July!! 🥳 I'm very grateful for my time at Princeton and SO excited for what comes next. We're building multiplayer games to study how teams are shaped by the problems they solve. If that sounds fun to you, I'm accepting grad students!

cogscikid.bsky.social
@cogscikid.bsky.social

Excited to share this work has been selected for an Oral at the Reinforcement Learning Conference Continual RL workshop! @continual-rl.bsky.social bsky.app/profile/cogs...

bsky.appbsky.app
cogscikid.bsky.social@cogscikid.bsky.social

Task diversity is supposedly key to generalization in RL. But what does it do to continual RL, where agents face one new task distribution after another? We find that past a point, more diversity actually inhibits continual reinforcement learning 🧵

cogscikid.bsky.social reposted

Charley Wu
@thecharleywu.bsky.social

Look how far we've come in just a few years! We're fast as well! It turns out, recruiting reviewers is much easier when they're not being used as free labor for massively profitable publishers Support free-to-publish and Open Access science by submitting to Open Mind today! direct.mit.edu/opmi

direct.mit.eduOpen Mind | MIT Press
languagemit.bsky.social@languagemit.bsky.social

—— Open Mind is MIT Press’s Diamond Open Access Cognitive Science journal. We now have our first Impact Factor: 2.9 as of June 2026 Submit your papers to Open Mind direct.mit.edu/opmi

cogscikid.bsky.social reposted

Andrew Lampinen
@lampinen.bsky.social

Pleased to share that this work is now published in TMLR! openreview.net/forum?id=RuW...

openreview.netLatent learning: episodic memory complements parametric learning by...When do machine learning systems fail to generalize, and what mechanisms could improve their generalization? Here, we draw inspiration from cognitive science to argue that one weakness of...
Andrew Lampinen@lampinen.bsky.social

Why does AI sometimes fail to generalize, and what might help? In a new paper (arxiv.org/abs/2509.16189), we highlight the latent learning gap — which unifies findings from language modeling to agent navigation — and suggest that episodic memory complements parametric learning to bridge it. Thread:

arxiv.orgLatent learning: episodic memory complements parametric learning by enabling flexible reuse of experiencesWhen do machine learning systems fail to generalize, and what mechanisms could improve their generalization? Here, we draw inspiration from cognitive science to argue that one weakness of machine lear...

cogscikid.bsky.social reposted

Taylor Webb
@taylorwwebb.bsky.social

Very excited to share that my lab will be moving to Princeton (Neuroscience & Psychology) this fall. I'll be recruiting at all levels (more info to come soon), please share / get in touch if you're interested in joining!

cogscikid.bsky.social reposted

Deon T. Benton
@deontbenton.bsky.social

Newborns' preference for faces is learned rather than based on a innate face-like template academic.oup.com/chidev/artic...

academic.oup.comWhat can face preference in newborns tell us about their face representation? Contrasting learned familiarity with inherent bias accountsAbstract. Possible mechanisms behind newborn face preferences are debated, including innate templates, top-heavy bias, focus on eyes, and rapid learning ex
cogscikid.bsky.social
@cogscikid.bsky.social

Task diversity is supposedly key to generalization in RL. But what does it do to continual RL, where agents face one new task distribution after another? We find that past a point, more diversity actually inhibits continual reinforcement learning 🧵

View full thread
cogscikid.bsky.social
@cogscikid.bsky.social

Strikingly, higher diversity improves backward transfer to the very first distribution! So the agent keeps getting better on old tasks even as it stops improving on new ones. This suggests it's learning the shared structure, but losing the ability to specialize.

cogscikid.bsky.social
@cogscikid.bsky.social

If you want to learn more, or see the rest of our analysis, please check out our paper! 📄 Paper: arxiv.org/abs/2606.00880 🌐 Project page: cogscikid.com/banyan/ This was a fun collaboration with Purab Seth, Neil Shah, @gershbrain.bsky.social, @kjha02.bsky.social, & @maxkw.bsky.social !

arxiv.orgTask diversity produces systematic transfer but inhibits continual reinforcement learningContinual reinforcement learning aims to produce agents that learn not only to improve at their current tasks but also to adapt as task distributions change. Training an agent on many diverse tasks ca...
cogscikid.bsky.social
@cogscikid.bsky.social

Task diversity is supposedly key to generalization in RL. But what does it do to continual RL, where agents face one new task distribution after another? We find that past a point, more diversity actually inhibits continual reinforcement learning 🧵

View full thread
cogscikid.bsky.social
@cogscikid.bsky.social

Task diversity inhibits continual reinforcement learning! At medium diversity (16 layouts × 16 topologies), performance climbs with each new distribution. But past that point, it monotonically gets worse.

cogscikid.bsky.social
@cogscikid.bsky.social

Task diversity is supposedly key to generalization in RL. But what does it do to continual RL, where agents face one new task distribution after another? We find that past a point, more diversity actually inhibits continual reinforcement learning 🧵

View full thread
cogscikid.bsky.social
@cogscikid.bsky.social

We analyze backward transfer and find that varying the number of layouts or object assignments has no effect, but *varying the number of topologies has a big effect*! Again, this is true for both PPO and PQN.

cogscikid.bsky.social
@cogscikid.bsky.social

Task diversity is supposedly key to generalization in RL. But what does it do to continual RL, where agents face one new task distribution after another? We find that past a point, more diversity actually inhibits continual reinforcement learning 🧵

View full thread
cogscikid.bsky.social
@cogscikid.bsky.social

Interestingly, diversity has even been key to cognitive science results. Lake & Baroni showed that the right diverse task distribution induced systematicity in neural networks! A question open since Fodor & Pylyshyn. But nobody has studied what diversity does to continual learning.

cogscikid.bsky.social
@cogscikid.bsky.social

So we built Banyan: a procedurally-generated environment where you can systematically vary the components that define a task. A Banyan task is composed of a layout & a task tree, where each tree is defined by its topology and the objects assigned to its nodes.

cogscikid.bsky.social
@cogscikid.bsky.social

Task diversity is supposedly key to generalization in RL. But what does it do to continual RL, where agents face one new task distribution after another? We find that past a point, more diversity actually inhibits continual reinforcement learning 🧵

cogscikid.bsky.social
@cogscikid.bsky.social

In ML, diversity is the standard route to generalization. For example, domain randomization has shown that visual diversity enables zero-shot sim-to-real and XLand showed that agents trained on diverse tasks could then solve millions of new ones zero-shot.

cogscikid.bsky.social reposted

DuckDuckGo
@duckduckgo.com

Since Google revealed its plans for an AI search overhaul, visits to our "No AI" search page have tripled…and they’re still rising! Want to make it your default on Chrome or Firefox? Grab our No-AI extensions and banish AI-assisted answers, chat, and AI images. (1/3)

cogscikid.bsky.social reposted

Mike Frank
@mcxfrank.bsky.social

If a student reads a typical psychology paper, picks an experiment, and tries to replicate it, they have roughly a coin-flip chance of success. That's the punchline of a decade of metascience, and it's the focus of Ch 3 of Experimentology. 🧵 experimentology.io

cogscikid.bsky.social reposted

Andrew Lampinen
@lampinen.bsky.social

We've updated the preprint of our Naturalistic Computational Cognitive Science paper (arxiv.org/abs/2502.20349) — we've tried to clarify and streamline the arguments, and added some new examples: 1/5

arxiv.orgNaturalistic Computational Cognitive Science: Towards generalizable models and theories that capture the full range of natural behaviorHow can cognitive science build generalizable theories that span the full scope of natural situations and behaviors? We argue that progress in Artificial Intelligence (AI) offers timely opportunities ...