Sam Gershman

This is one of the most creative and interesting games I've ever played. You get dunked head first into a conceptual art project.

If you work at the intersection of computational neuroscience and machine learning, consider applying for this postdoc position (January 2027 start date):
academicpositions.harvard.edu/postings/15868
An opportunity to work with a great group of people across Harvard, MIT, and UC Berkeley.

"data available upon reasonable request"

The lesson for all the students out there is that science is a community project. Most of us make individually small contributions to this project. Success is measured at the collective level. Many of our professional (and personal) dysfunctions could be fixed by more fully embracing this view.

In fairness to Grossberg, he made valuable contributions to computational neuroscience. The problem is that his grievances alienated all the people who would otherwise have appreciated his contributions. I see this as a cautionary tale about what happens when scientists are obsessed with credit.

When I started grad school, my advisor counseled me that I'd know that I'd made it in neuroscience when Grossberg accuses me of stealing his ideas. Probably his advisor told him the same thing!

Connectionists mailing list.

If I were him, what I would be worried about is other people getting called "the Stephen Grossberg of X" which would probably mean someone whose genuine scientific contributions are overshadowed by their narcissism and shameless self-aggrandizement.

Stephen Grossberg explaining why he is both the Newton *and* the Einstein of the mind. Why not also the Lavoisier, Curie, and Darwin of the mind? 🙄

We are hiring next year in semantics- one year positions of course less exciting and good for the field than straight up tenure track postings, but on the other hand hiring in something that isn’t computational feels like a win anymore in linguistics. Come bring some innovative takes on meaning!

🚨 Hiring in Munich 🇩🇪: 2 open-topic PhD positions in human & machine learning (TVöD E13 80%).
Start ~June 2026 (flexible). Deadline: March 2, 2026.
Apply/info: hcai-munich.com/PhD_Job_Ad.pdf
Reposts appreciated 🙏

a close up of a person 's face with the words in the computer below ALT: a close up of a person 's face with the words in the computer below

🤣 totally!

Given how secretive tech companies are these days (supposedly only publishing the stuff that doesn't work well), can we infer the most cutting-edge algorithms from what is *not* published?

Some background reading on what this research is about:
gershmanlab.com/pubs/cell_le...
gershmanlab.com/pubs/memory_...
www.cell.com/current-biol...

My lab is looking to recruit 1-2 paid summer interns to do wet lab work on learning in a unicellular organism (Stentor coeruleus). You can apply here:
forms.gle/b47WpSobjFjo...

We are hiring a research specialist, to start this summer! This position would be a great fit for individuals looking to get more experience in computational and cognitive neuroscience research before applying to graduate school. #neurojobs Apply here: research-princeton.icims.com/jobs/21503/r...

This plot is a nice illustration of what the system (blue diamond) can accomplish relative to throwing LLMs directly at the problem: high success rate and much lower computational cost. A reasoning LLM (red diamond) achieves an okay (still worse) success rate, but requires twice as much computation.

Learning Abstractions for Hierarchical Planning in Program-Synthesis Agents Humans learn abstractions and use them to plan efficiently to quickly generalize across tasks -- an ability that remains challenging for state-of-the-art large language model (LLM) agents and deep rei...

A new and improved version of TheoryCoder, which learns to play video games in a human-like way by synthesizing both high-level abstractions and a low-level model of game mechanics:
arxiv.org/abs/2602.00929

A real-world, high-stakes case study of policy compression:
osf.io/preprints/ps...
Clinicians in the emergency department exhibit near-optimal adaptation to varying cognitive demands.

#JNeurosci: Results from Hall-McMaster, @nicoschuck.bsky.social, @gershbrain.bsky.social et al suggest the entorhinal cortex might highlight aspects of past experiences that can be generalized, allowing us to make effective decisions in new environments https://doi.org/10.1523/JNEUROSCI.1492-25.2025

RA job alert. We are looking for a fearless experimentalist to join our experiments on cortical neuromodulators during visually guided learning, lead by @michael-lynn.bsky.social . Opportunity to learn 2p imaging, design and conduct experiments. RT plz. Apply here: my.corehr.com/pls/uoxrecru...

now accepted at ICLR! 🐺🥳🐺

arxiv.org/abs/2506.20666

Thanks, Nancy!! I'm teaching it this semester, so your students can take my class if they're interested.

I'm gradually populating this page with additional resources: lecture slides, projects, code. As always, please let me know if you have any feedback!

And of course please correct me if I've misunderstood anything!

I'm really glad Jeong et al did this work. It's a valuable empirical contribution and will stimulate new theoretical work, even if I'm not yet completely persuaded by their interpretation.

The paper posits that latent cause models should treat G1 and G3 the same. Some do, but I could imagine some (capable of retrospective revaluation) that don't.

This alternative account might also offer a different interpretation of the DA results. Inhibition during extinction could induce a prediction error that drives latent cause inference.

An alternative explanation of the behavioral findings is that G2 gets more evidence for the acq latent cause in extinction than G1, thereby overwriting the original memory.

I'd be curious if ANCCR can explain the gradual extinction effect:

www.frontiersin.org/journals/beh...

elifesciences.org/articles/95849

Here the retrospective contingency is held fixed across conditions, but there are large differences in later memory recovery.