Proud of the work from hmc-lab.com & collaborators @ #CogSci2025 this year, but sad I cant be there myself
@hanqizhou.bsky.social @davidnagy.bsky.social @alexthewitty.bsky.social @stepalminteri.bsky.social @brendenlake.bsky.social @kefang.bsky.social @rdhawkins.bsky.social @meanwhileina.bsky.social

Looking forward to sharing our work at #cogsci2025! We aim at getting one step closer to a domain-general formulation of mental costs via policy compression.

Come see my presentation at "Talks 35: Reasoning". It is scheduled at 16:44 PST, August 1 at Nob Hill C!

gershmanlab.com/pubs/LiuGers...

The shift from pensions to 401ks played a big role in fueling Americans' hesitance to tax big wealth. Because now, even Americans who have only a tiny retirement account feel invested in the success of an institution that overwhelmingly benefits those much wealthier than them.

a paradox I'm currently grappling with: choosing actions valued by your "group" increases your morale, but being able to choose actions that go against the norms of your group increases your agency 🤔

Maggie Boden and Murray Shanahan on a panel in 2018

Very sad to learn of the death on 18th July of Margaret (Maggie) Boden, a titan of cognitive science and AI. I met her many times, and respected her greatly.
www.theargus.co.uk/memorials/de...

Fast efficient coding and sensory adaptation in gain-adaptive recurrent networks As the statistics of sensory environments often change, neural sensory systems must adapt to maintain useful representations. Efficient coding prescribes that neuronal tuning curves should be optimize...

Efficient coding theories often implicitly or explicitly assume slow changes in tuning (e.g., through synaptic plasticity). Arthur Prat-Carrabin has collected psychophysical data showing that it can be fast, and this can be explained by a gain-adaptive RNN:
www.biorxiv.org/content/10.1...

excited to have you join us!

Excited to announce the first workshop on CogInterp: Interpreting Cognition in Deep Learning Models @ NeurIPS 2025! 📣

How can we interpret the algorithms and representations underlying complex behavior in deep learning models?

🌐 coginterp.github.io/neurips2025/

1/4

Really pumped for my Oral presentation on this work today!!! Come check out the RL session from 3:30-4:30pm in West Ballroom B

You can also swing by our poster from 4:30-7pm in West Exhibition Hall B2-B3 # W-713

See you all there!

THIS WEEK the @cogscisociety.bsky.social Minds in the Making workshop brings you Vanessa Bermudez and @togelius.bsky.social in conversation about LEARNING 🧠💪 x DESIGN 🛠️! And what makes GAMES awesome for learning. Wednesday July 16th, 12pm-1pm PT. Register here: stanford.zoom.us/meeting/regi....

New published paper with Ham Huang and @actlab.bsky.social! @cognitionjournal.bsky.social
authors.elsevier.com/sd/article/S...

Project page: cogscikid.com/preplay

Code + Data: github.com/KempnerInsti...

Preprint: arxiv.org/abs/2507.05561

4. Finally, Multitask Preplay might also provide a computational account for "preplay"---a phenomenon where hippocampal place cells activate during rest for locations an animal hasn't yet visited but will in the future.

3. humans and robots face the same bottleneck where collecting new experience in the world is expensive. Multitask Preplay might inspire future work where robots can learn about many goals from each real-world experience.

Some reasons Multitask Preplay is cool

1. it shows that fast, reactive behavior can be surprisingly adept to novel tasks when trained with the right learning algorithm.

2. it is capable of predicting human behavior in natural domains that share task oc-occurence structure like real-world homes.

4. Finally, Multitask Preplay might also provide a computational account for "preplay"---a phenomenon where hippocampal place cells activate during rest for locations an animal hasn't yet visited but will in the future.

3. humans and robots face the same bottleneck where collecting new experience in the world is expensive. Multitask Preplay might inspire future work where robots can learn about many goals from each real-world experience.

Some reasons Multitask Preplay is cool

1. it shows that fast, reactive behavior can be surprisingly adept to novel tasks when trained with the right learning algorithm.

2. it is capable of predicting human behavior in natural domains that share task oc-occurence structure like real-world homes.

Main result 3: we speculate that this may have unexpected benefits for human generalization.

We present AI simulations where Multitask Preplay improves generalization of complex, long-horizon behaviors to 10,000 unique new environments when they share subtask co-occurrence structure.

Main result 2: we generalize these predictions to Craftax, a partially observable, 2D minecraft domain.

Here, we once again find evidence that people preplay completion of tasks that were accessible but unpursued, but now in a much larger world where generalization to new tasks is much harder.

Main result 1: Across 4 experiments, we find evidence that people preplay completion of tasks that were accessible but unpursued, even if they don't know those tasks will come up later on.

Key idea behind Multitask Preplay: when people engage in replay of one task, they might "preplay" completion of another task and leverage temporal-difference learning to cache the results into a neural network, enabling fast, automatic behavior for that task later on.

arxiv.org/abs/2507.05561

Excited to share a new project spanning cognitive science and AI where we develop a novel deep reinforcement learning model---Multitask Preplay---that explains how people generalize to new tasks that were previously accessible but unpursued.

Hello world! This is the RL & Agents Reading Group

We organise regular meetings to discuss recent papers in Reinforcement Learning (RL), Multi-Agent RL and related areas (open-ended learning, LLM agents, robotics, etc).

Meetings take place online and are open to everyone 😊

Considering What We Know and What We Don’t Know: Expectations and Confidence Guide Value Integration in Value-Based Decision-Making Abstract. When making decisions, we often have more information about some options than others. Previous work has shown that people are more likely to choose options that they look at more and those t...

Our newest paper, led by Romy Froemer
and @fredcallaway.bsky.social‬, is now out in Open Mind: “Considering What We Know and What We Don’t Know: Expectations and Confidence Guide Value Integration in Value-Based Decision-Making”

direct.mit.edu/opmi/article...

Preemptive Solving of Future Problems: Multitask Preplay in Humans and Machines Humans can pursue a near-infinite variety of tasks, but typically can only pursue a small number at the same time. We hypothesize that humans leverage experience on one task to preemptively learn solu...

@cogscikid.bsky.social has cool new work, in collaboration with Honglak Lee and Sam Hall-McMaster, on how humans solve multitask reinforcement learning problems by preemptively simulating paths to counterfactual goals:
arxiv.org/abs/2507.05561
He also shows that it works on challenging ML tasks.

gotta say that the US decimating its own academic research ecosystem is making it a bit easier to recruit students who would have otherwise probably stayed here and pursued US PhDs

Music is an incredibly powerful retrieval cue. What is the neural basis of music-evoked memory reactivation? And how does this reactivation relate to later memory for the retrieved events? In our new study, we used Eternal Sunshine of the Spotless Mind to find out. www.biorxiv.org/content/10.1...

Beyond Statistical Learning: Exact Learning Is Essential for General Intelligence Sound deductive reasoning -- the ability to derive new knowledge from existing facts and rules -- is an indisputably desirable aspect of general intelligence. Despite the major advances of AI systems ...

First position paper I ever wrote. "Beyond Statistical Learning: Exact Learning Is Essential for General Intelligence" arxiv.org/abs/2506.23908 Background: I'd like LLMs to help me do math, but statistical learning seems inadequate to make this happen. What do you all think?

Discovering cognitive strategies with tiny recurrent neural networks - Nature Modelling biological decision-making with tiny recurrent neural networks enables more accurate predictions of animal choices than classical cognitive models and offers insights into the underlying cog...

Thrilled to see our TinyRNN paper in @nature! We show how tiny RNNs predict choices of individual subjects accurately while staying fully interpretable. This approach can transform how we model cognitive processes in both healthy and disordered decisions. doi.org/10.1038/s415...