Sam Gershman
@gershbrain.bsky.social
Professor, Department of Psychology and Center for Brain Science, Harvard University
https://gershmanlab.com/
https://gershmanlab.com/
In his Treatise on Instrumentation and Orchestration, the composer Hector Berlioz envisioned an ideal orchestra that included 30 grand pianos, 30 harps, 360 singers, 120 violins!! As Gertrude Stein said, "If it can be done, why do it?"
www.hberlioz.com/Scores/Berli...
www.hberlioz.com/Scores/Berli...
November 25, 2025 at 4:28 PM
In his Treatise on Instrumentation and Orchestration, the composer Hector Berlioz envisioned an ideal orchestra that included 30 grand pianos, 30 harps, 360 singers, 120 violins!! As Gertrude Stein said, "If it can be done, why do it?"
www.hberlioz.com/Scores/Berli...
www.hberlioz.com/Scores/Berli...
Reposted by Sam Gershman
I am recruiting graduate students for the experimental side of my lab @mcgill.ca for admission in Fall 2026!
Get in touch if you're interested in how brain circuits implement distributed computation, including dopamine-based distributed RL and probabilistic representations.
Get in touch if you're interested in how brain circuits implement distributed computation, including dopamine-based distributed RL and probabilistic representations.
November 19, 2025 at 6:04 PM
I am recruiting graduate students for the experimental side of my lab @mcgill.ca for admission in Fall 2026!
Get in touch if you're interested in how brain circuits implement distributed computation, including dopamine-based distributed RL and probabilistic representations.
Get in touch if you're interested in how brain circuits implement distributed computation, including dopamine-based distributed RL and probabilistic representations.
Reposted by Sam Gershman
Storing long-lived memories via molecular error correction: a minimal mathematical model of Crick's memory switch https://www.biorxiv.org/content/10.1101/2025.11.13.688304v1
November 14, 2025 at 8:08 AM
Storing long-lived memories via molecular error correction: a minimal mathematical model of Crick's memory switch https://www.biorxiv.org/content/10.1101/2025.11.13.688304v1
If funders wanted to make a huge positive impact on scientific practice, they would mandate that all publications appear first as registered reports, that APCs are only paid for RRs, and that grant applications only require preliminary data for RR sample size determination / power analysis.
November 12, 2025 at 11:55 AM
If funders wanted to make a huge positive impact on scientific practice, they would mandate that all publications appear first as registered reports, that APCs are only paid for RRs, and that grant applications only require preliminary data for RR sample size determination / power analysis.
Reposted by Sam Gershman
Don’t count US science out just yet. The largest economy on the planet will rebound in a very big way. It’s already correcting…..
episteme.com
episteme.com
Episteme
We are a new type of R&D organization that recruits exceptional researchers across disciplines, who want to pursue important translational science for humanity that cannot be enabled efficiently withi...
episteme.com
November 12, 2025 at 9:25 AM
Don’t count US science out just yet. The largest economy on the planet will rebound in a very big way. It’s already correcting…..
episteme.com
episteme.com
My reviewing style has changed over time. Rather than litigate every little thing, and pushing my own ideas, I focus only on 2 things:
(1) Are the claims interesting/important?
(2) Does the evidence support the claims?
Most of my reviews these days are short and focused.
(1) Are the claims interesting/important?
(2) Does the evidence support the claims?
Most of my reviews these days are short and focused.
November 8, 2025 at 11:22 AM
My reviewing style has changed over time. Rather than litigate every little thing, and pushing my own ideas, I focus only on 2 things:
(1) Are the claims interesting/important?
(2) Does the evidence support the claims?
Most of my reviews these days are short and focused.
(1) Are the claims interesting/important?
(2) Does the evidence support the claims?
Most of my reviews these days are short and focused.
Reposted by Sam Gershman
New #J2C Certification:
Synthesizing world models for bilevel planning
Zergham Ahmed, Joshua B. Tenenbaum, Chris Bates, Samuel J. Gershman
https://openreview.net/forum?id=m9V4JHLJrD
#planning #reinforcement #games
Synthesizing world models for bilevel planning
Zergham Ahmed, Joshua B. Tenenbaum, Chris Bates, Samuel J. Gershman
https://openreview.net/forum?id=m9V4JHLJrD
#planning #reinforcement #games
November 3, 2025 at 9:24 PM
New #J2C Certification:
Synthesizing world models for bilevel planning
Zergham Ahmed, Joshua B. Tenenbaum, Chris Bates, Samuel J. Gershman
https://openreview.net/forum?id=m9V4JHLJrD
#planning #reinforcement #games
Synthesizing world models for bilevel planning
Zergham Ahmed, Joshua B. Tenenbaum, Chris Bates, Samuel J. Gershman
https://openreview.net/forum?id=m9V4JHLJrD
#planning #reinforcement #games
Kazuki Irie has a forthcoming paper in NeurIPS that studies the following idea:
Linear attention has cheap, unbounded memory but low precision, whereas softmax attention has expensive, bounded memory but high precision. These can be combined to build better transformers.
arxiv.org/abs/2506.00744
Linear attention has cheap, unbounded memory but low precision, whereas softmax attention has expensive, bounded memory but high precision. These can be combined to build better transformers.
arxiv.org/abs/2506.00744
Blending Complementary Memory Systems in Hybrid Quadratic-Linear Transformers
We develop hybrid memory architectures for general-purpose sequence processing neural networks, that combine key-value memory using softmax attention (KV-memory) with fast weight memory through dynami...
arxiv.org
November 4, 2025 at 10:33 AM
Kazuki Irie has a forthcoming paper in NeurIPS that studies the following idea:
Linear attention has cheap, unbounded memory but low precision, whereas softmax attention has expensive, bounded memory but high precision. These can be combined to build better transformers.
arxiv.org/abs/2506.00744
Linear attention has cheap, unbounded memory but low precision, whereas softmax attention has expensive, bounded memory but high precision. These can be combined to build better transformers.
arxiv.org/abs/2506.00744
Reposted by Sam Gershman
John J. Vastola, Samuel J. Gershman, Kanaka Rajan: Gradient Descent as Loss Landscape Navigation: a Normative Framework for Deriving Learning Rules https://arxiv.org/abs/2510.26997 https://arxiv.org/pdf/2510.26997 https://arxiv.org/html/2510.26997
November 3, 2025 at 6:32 AM
John J. Vastola, Samuel J. Gershman, Kanaka Rajan: Gradient Descent as Loss Landscape Navigation: a Normative Framework for Deriving Learning Rules https://arxiv.org/abs/2510.26997 https://arxiv.org/pdf/2510.26997 https://arxiv.org/html/2510.26997
Reposted by Sam Gershman
“They’re earning huge profit margins because we’ve all kind of acquiesced to a system in which we provide free services to corporations. Part of the open access movement is an attempt to break that,” Gershman said.
www.thecrimson.com/article/2025...
www.thecrimson.com/article/2025...
Academic Publishing Keeps Getting More Expensive. Some Harvard Scholars Want to Make It Free. | News | The Harvard Crimson
The high cost of publishing open access has plagued researchers for years, but a dedicated group of Harvard scientists and librarians are fighting to alleviate the costs of publishing.
www.thecrimson.com
October 30, 2025 at 2:00 PM
“They’re earning huge profit margins because we’ve all kind of acquiesced to a system in which we provide free services to corporations. Part of the open access movement is an attempt to break that,” Gershman said.
www.thecrimson.com/article/2025...
www.thecrimson.com/article/2025...
Gonna wear this to the next faculty meeting to see if they think I’m Steve Pinker.
October 31, 2025 at 9:38 PM
Gonna wear this to the next faculty meeting to see if they think I’m Steve Pinker.
What's with machine learning researchers always reporting standard deviation instead of standard error? My understanding is that the error bars are typically used to back up inferential claims about significant differences between sample means (although statistical tests are rare, another problem).
October 26, 2025 at 10:14 AM
What's with machine learning researchers always reporting standard deviation instead of standard error? My understanding is that the error bars are typically used to back up inferential claims about significant differences between sample means (although statistical tests are rare, another problem).
In case you don't know already, the journal Open Mind has a Bluesky account that automatically posts new papers:
@openmindjournal.bsky.social
The journal is diamond open access (free to read, free to publish) thanks to the support of MIT Press, Harvard Library, & MIT Library.
@openmindjournal.bsky.social
The journal is diamond open access (free to read, free to publish) thanks to the support of MIT Press, Harvard Library, & MIT Library.
October 24, 2025 at 12:03 PM
In case you don't know already, the journal Open Mind has a Bluesky account that automatically posts new papers:
@openmindjournal.bsky.social
The journal is diamond open access (free to read, free to publish) thanks to the support of MIT Press, Harvard Library, & MIT Library.
@openmindjournal.bsky.social
The journal is diamond open access (free to read, free to publish) thanks to the support of MIT Press, Harvard Library, & MIT Library.
Reposted by Sam Gershman
Applying to do a postdoc or PhD in theoretical ML or neuroscience this year? Consider joining my group (starting next Fall) at UT Austin!
POD Postdoc: oden.utexas.edu/programs-and... CSEM PhD: oden.utexas.edu/academics/pr...
POD Postdoc: oden.utexas.edu/programs-and... CSEM PhD: oden.utexas.edu/academics/pr...
October 23, 2025 at 9:36 PM
Applying to do a postdoc or PhD in theoretical ML or neuroscience this year? Consider joining my group (starting next Fall) at UT Austin!
POD Postdoc: oden.utexas.edu/programs-and... CSEM PhD: oden.utexas.edu/academics/pr...
POD Postdoc: oden.utexas.edu/programs-and... CSEM PhD: oden.utexas.edu/academics/pr...
Reposted by Sam Gershman
We're excited to announce that Cognitive Science at Dartmouth is recruiting PhD students to work collaboratively with me, Steven Frankland, and Fred Callaway. Come study the principles and mechanisms that enable us to understand, plan, and act in the world! Info: sites.dartmouth.edu/cogscigrad/
Cognitive Science Graduate Admissions – Information about graduate admissions from the cognitive science faculty
sites.dartmouth.edu
October 23, 2025 at 5:30 PM
We're excited to announce that Cognitive Science at Dartmouth is recruiting PhD students to work collaboratively with me, Steven Frankland, and Fred Callaway. Come study the principles and mechanisms that enable us to understand, plan, and act in the world! Info: sites.dartmouth.edu/cogscigrad/
Is there any evidence that the brain uses something like a target network in deep reinforcement learning (i.e., a slowly updating copy of the "online" network used to control the policy)?
October 22, 2025 at 10:16 AM
Is there any evidence that the brain uses something like a target network in deep reinforcement learning (i.e., a slowly updating copy of the "online" network used to control the policy)?
Everything would be much simpler/cheaper/efficient if scientists could apply for funding by just sending their CV to the funder. If the funder likes what's on the CV, they send money. Instead of progress reports, the scientist just sends an updated CV, and the funder can decide whether to give more.
October 20, 2025 at 4:41 PM
Everything would be much simpler/cheaper/efficient if scientists could apply for funding by just sending their CV to the funder. If the funder likes what's on the CV, they send money. Instead of progress reports, the scientist just sends an updated CV, and the funder can decide whether to give more.
When I was an undergrad, I saw a talk by Oliver Sacks about music. During the Q&A, someone asked Sacks if he thought science would reveal the deepest mysteries of art. Sacks (a rare humanist among scientists) said that he doubted it, at which Eric Kandel (the host) leaped up and grabbed the mic.
October 15, 2025 at 9:27 AM
When I was an undergrad, I saw a talk by Oliver Sacks about music. During the Q&A, someone asked Sacks if he thought science would reveal the deepest mysteries of art. Sacks (a rare humanist among scientists) said that he doubted it, at which Eric Kandel (the host) leaped up and grabbed the mic.
Reposted by Sam Gershman
New preprint!
"Non-commitment in mental imagery is distinct from perceptual inattention, and supports hierarchical scene construction"
(by Li, Hammond, & me)
link: doi.org/10.31234/osf...
-- the title's a bit of a mouthful, but the nice thing is that it's a pretty decent summary
"Non-commitment in mental imagery is distinct from perceptual inattention, and supports hierarchical scene construction"
(by Li, Hammond, & me)
link: doi.org/10.31234/osf...
-- the title's a bit of a mouthful, but the nice thing is that it's a pretty decent summary
October 14, 2025 at 1:22 PM
New preprint!
"Non-commitment in mental imagery is distinct from perceptual inattention, and supports hierarchical scene construction"
(by Li, Hammond, & me)
link: doi.org/10.31234/osf...
-- the title's a bit of a mouthful, but the nice thing is that it's a pretty decent summary
"Non-commitment in mental imagery is distinct from perceptual inattention, and supports hierarchical scene construction"
(by Li, Hammond, & me)
link: doi.org/10.31234/osf...
-- the title's a bit of a mouthful, but the nice thing is that it's a pretty decent summary
Really interesting work by Bakhurin and colleagues challenging the reward prediction error hypothesis of dopamine:
www.nature.com/articles/s41...
I love this figure which both echoes and undermines the famous figure from Schultz et al. (1997).
www.nature.com/articles/s41...
I love this figure which both echoes and undermines the famous figure from Schultz et al. (1997).
October 14, 2025 at 11:05 AM
Really interesting work by Bakhurin and colleagues challenging the reward prediction error hypothesis of dopamine:
www.nature.com/articles/s41...
I love this figure which both echoes and undermines the famous figure from Schultz et al. (1997).
www.nature.com/articles/s41...
I love this figure which both echoes and undermines the famous figure from Schultz et al. (1997).
Everyone knows that temporal contiguity is important for associative learning. As the interval between a cue (e.g., a light) and an outcome (e.g., shock) gets longer, the conditioned response (e.g., freezing to the tone) is acquired less quickly.
October 10, 2025 at 11:17 AM
Everyone knows that temporal contiguity is important for associative learning. As the interval between a cue (e.g., a light) and an outcome (e.g., shock) gets longer, the conditioned response (e.g., freezing to the tone) is acquired less quickly.
Reposted by Sam Gershman
Interested in understanding how young humans think about social relationships? I am reading PhD applications this year! **Please note**, that Harvard now requires the GRE. More information here: www.ashleyjthomas.com/workwithme
WANT TO WORK WITH ME? | Mysite
www.ashleyjthomas.com
October 6, 2025 at 8:55 PM
Interested in understanding how young humans think about social relationships? I am reading PhD applications this year! **Please note**, that Harvard now requires the GRE. More information here: www.ashleyjthomas.com/workwithme
Reposted by Sam Gershman
We are thrilled to announce that our NEW Large Language Model will be released on 11.18.25.
October 1, 2025 at 2:38 PM
We are thrilled to announce that our NEW Large Language Model will be released on 11.18.25.
Reposted by Sam Gershman
Ever wonder how habituation works? Here's our attempt to understand:
A stimulus-computable rational model of visual habituation in infants and adults doi.org/10.7554/eLif...
This is the thesis of two wonderful students: @anjiecao.bsky.social @galraz.bsky.social, w/ @rebeccasaxe.bsky.social
A stimulus-computable rational model of visual habituation in infants and adults doi.org/10.7554/eLif...
This is the thesis of two wonderful students: @anjiecao.bsky.social @galraz.bsky.social, w/ @rebeccasaxe.bsky.social
September 29, 2025 at 11:38 PM
Ever wonder how habituation works? Here's our attempt to understand:
A stimulus-computable rational model of visual habituation in infants and adults doi.org/10.7554/eLif...
This is the thesis of two wonderful students: @anjiecao.bsky.social @galraz.bsky.social, w/ @rebeccasaxe.bsky.social
A stimulus-computable rational model of visual habituation in infants and adults doi.org/10.7554/eLif...
This is the thesis of two wonderful students: @anjiecao.bsky.social @galraz.bsky.social, w/ @rebeccasaxe.bsky.social
I was part of an interesting panel discussion yesterday at an ARC event. Maybe everybody knows this already, but I was quite surprised by how "general" intelligence was conceptualized in relation to human intelligence and the ARC benchmarks.
September 28, 2025 at 10:06 AM
I was part of an interesting panel discussion yesterday at an ARC event. Maybe everybody knows this already, but I was quite surprised by how "general" intelligence was conceptualized in relation to human intelligence and the ARC benchmarks.