Harvard researcher at the intersection of computational cognitive science and reinforcement learning systems
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Not yet scored. This axis measures total publication volume and recent productivity, but a full OpenAlex output record is unavailable in this snapshot; only the two reviewed 2026 arXiv preprints are visible, so output cannot be characterized honestly.
Impact—
Not yet scored. This axis measures citation accumulation and downstream influence across the literature; the snapshot contains no citation counts or impact record, so influence cannot be estimated without inventing figures.
Focus0/100
Focus scores 0, and the raw field distribution explains why: the reviewed catalog splits exactly evenly — 50% computational cognitive science / AI evaluation and 50% reinforcement learning (environments and ML systems). With one 2026 paper in each field, the evidence shows a deliberately dual-track portfolio rather than a concentrated specialization; the score reflects breadth across the two fields, not an absence of research direction.
Collaboration82/100
Collaboration scores 82, backed by 59 distinct coauthors and a stable core of twelve frequent collaborators, including Lance Ying, Katherine M. Collins, and Akshay K. Jagadish. On both cataloged 2026 papers Gershman appears as a middle author and not the corresponding author, consistent with contributing inside larger team projects rather than leading small groups — a broad, networked collaboration pattern.
Activity—
Not yet scored. This axis measures publication cadence and research tempo over time; the snapshot covers only two 2026 papers with no multi-year record, so activity level cannot be assessed yet.
Reproducibility78/100
Reproducibility averages 78 across the two scored papers. PlayTrain rates high readiness — the Evolve verdict highlights strong engineering and benchmarks, with over 1M agent-steps per second on one GPU node — while CogGym rates medium despite being judged a rigorous, well-validated benchmark. With only two papers scored, this is a promising but coarse early signal of artifact quality and openness.
Research summary
Samuel J. Gershman is a researcher at Harvard University whose reviewed work, as indexed by Evolve, splits roughly evenly between computational cognitive science / AI evaluation and reinforcement learning environments and ML systems. The catalog currently holds two 2026 arXiv preprints. The first, CogGym, unifies 258 cognitive experiments into a large-scale benchmark for comparative evaluation of human and machine cognition; Evolve's verdict finds LLM-human alignment scales with model size but remains far below human reliability, calling it a rigorous, well-validated benchmark. The second, PlayTrain, trains agents in LLM-generated JavaScript game environments at more than one million agent-steps per second on a single GPU node; Evolve rates its engineering and benchmarks strong while noting that clone fidelity remains unquantified. Collaboration metrics are strong: an Evolve collaboration score of 82, 59 distinct coauthors, and a recurring circle of twelve frequent collaborators including Lance Ying, Katherine M. Collins, and Akshay K. Jagadish. On both cataloged papers Gershman appears as a middle author rather than corresponding author, consistent with a contributor role on larger team projects. Reproducibility readiness averages 78 across the two scored papers — high for PlayTrain, medium for CogGym. Output volume, citation impact, and publication cadence are not yet scored, so this remains an early-stage profile built on 2026 preprints.
Collaboration pattern
An Evolve collaboration score of 82 reflects a wide network: 59 distinct coauthors and a recurring core of twelve frequent collaborators, including Lance Ying, Jinzhou Wu, Katherine M. Collins, Akshay K. Jagadish, and Guangyuan Jiang. On both cataloged papers Gershman appears as a middle author and not the corresponding author, a pattern consistent with contributing within larger team efforts rather than leading small single-group projects.
Research keywords
Highlights
CogGym unifies 258 cognitive experiments into one benchmark
Scaling narrows — but does not close — the human-machine gap
PlayTrain trains LLM-generated game environments at >1M agent-steps/s
Timeline
- 2026Co-authors CogGym, a large-scale comparative evaluation of human and machine cognition (middle author, arXiv preprint).
- 2026Co-authors PlayTrain, a high-throughput reinforcement learning framework for LLM-generated JavaScript games (middle author, arXiv preprint).
- 2026Both preprints enter the Evolve review catalog, one in each of his two tracked fields, with reproducibility readiness rated medium and high respectively.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometric dimensions are derived from OpenAlex-based data only; the output, impact, and activity axes are unscored because underlying records are insufficient.
- Coverage is limited to the two 2026 arXiv preprints Evolve has reviewed; the catalog is not a complete publication record for this researcher.
- The reproducibility mean of 78 rests on only two scored papers, so the estimate is coarse.
- Identity is name-resolved via ORCID 0000-0002-6546-3298 from public bibliographic data; country-level affiliation is not specified in this snapshot.