Samuel Larson — early-career ML researcher probing rank recruitment in trained matrix memories and latent reasoning
3
papers here
3
reviewed by Evolve
3
first-author
3
corresponding
—
works · OpenAlex
0
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—
The output axis measures total publication volume and its distribution over time; it is not yet scored because the snapshot records only four works in the last three years, all in 2026, too little for a stable output profile.
Impact—
Impact would measure citations and community uptake; the snapshot reports h-index 0 and i10-index 0 with no citation counts, so no impact score can yet be computed for this 2026-starting record.
Focus0/100
Scores 0 on focus because the field distribution splits evenly across three labels — machine learning, machine learning with representation learning and mechanistic interpretability, and mechanistic interpretability of latent reasoning — at 0.33 each. Topically, however, the three catalog papers converge on a single question: whether trained matrix memories recruit the effective rank a task demands, spanning associative memory, group composition, and latent reasoning.
Collaboration0/100
The collaboration score is 0, consistent with zero distinct co-authors in the snapshot and an empty co-author list. All three catalog papers list Larson as first and corresponding author with no recorded co-authors, indicating a fully solo workflow on arXiv preprints. This is typical of an independent early-stage researcher; future collaborations would be needed to shift this profile.
Activity47/100
Activity scores 47, reflecting a very recent start: the counts-by-year record shows all 4 works appearing in 2026, and the last-three-year count is also 4. With output beginning only in 2026, the mid-range score captures genuine current productivity — three arXiv preprints in the Evolve catalog — against a short publication window that limits year-over-year evidence.
Reproducibility85/100
Reproducibility is the strongest axis: a mean readiness score of 85 across 3 reviewed papers. Two catalog papers are rated high readiness (the matrix-memory rank paper and the matrix-CODI negative result) and one medium (the group-composition rank law, noted for rigorous preregistration but analytic-only necessity). Evolve verdicts flag single-seed analyses and small synthetic setups as the main caveats.
Research summary
Samuel Larson is an early-stage machine-learning researcher whose 2026 arXiv output concentrates on one sharply defined question: how trained matrix memories recruit effective rank, and whether that rank is causally required by the task. The Evolve catalog holds three papers, all with Larson as first and corresponding author and none with recorded co-authors. 'When the Gradient Sees Rank' (2026) reports provable necessity, causal recruitment, and composition in trained matrix memories, with a sharp transition at k=K — evidence Evolve calls convincing at small scale, though key analyses rest on single seeds. 'The Rank the Task Demands' (2026) frames a causal rank law for matrix memories trained on group composition, grounded in representation theory; its preregistration is rigorous, but necessity is established analytically rather than empirically. 'The Gradient Does Not See Rank' (2026) is an honest negative result on matrix-CODI latents over ProsQA, with good controls, though limited to one task and weak accuracy. Bibliometrics are thin: the OpenAlex snapshot shows 4 works, all in 2026, h-index and i10-index of 0, and zero distinct co-authors. Reproducibility is the standout signal — a mean readiness of 85 across 3 reviewed papers. The profile is name-resolved from public bibliographic data; no institution or country is on record.
Collaboration pattern
The snapshot records 0 distinct co-authors and an empty frequent-co-author list, and all three Evolve-catalog papers carry Larson as first and corresponding author with no co-authors, matching the collaboration score of 0. The work is entirely solo across the 2026 arXiv series. Given the single-name identity resolution from OpenAlex, co-author coverage could also be incomplete; no collaboration pattern beyond independent work can be established from current data.
Research keywords
Highlights
Sharp rank transition in trained matrix memories
A causal rank law grounded in group composition
An honest negative result: rank-indifference in matrix-CODI
Timeline
- 2026Reports provable necessity, causal recruitment, and composition in trained matrix memories, with a sharp transition at k=K — the flagship positive result of the 2026 arXiv series.
- 2026Formulates a causal rank law for matrix memories trained on group composition, with rigorous preregistration and representation-theoretic grounding.
- 2026Documents rank-indifference in matrix-CODI latents on ProsQA as an honest negative result, completing a 2026 output of 4 recorded works.
Papers in the Evolve catalog
- first authorcorresponding
When the Gradient Sees Rank: Provable Necessity, Causal Recruitment, and Composition in Trained Matrix Memories
- first authorcorresponding
The Rank the Task Demands: A Causal Rank Law for Matrix Memories Trained on Group Composition
- first authorcorresponding
The Gradient Does Not See Rank: Rank-Indifference in Matrix-CODI on ProsQA
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 data comes from OpenAlex only (identity: openalex); h-index and i10-index are recorded as 0 with no citation counts yet.
- No institution, country, or ORCID is available; the profile is name-resolved from public bibliographic data, and 'Samuel Larson' is a common name, so identity confidence is limited.
- All three catalog papers are 2026 arXiv preprints; no peer-reviewed venue information is in scope.
- Evolve verdicts flag single-seed analyses, small-scale synthetic setups, and single-task evaluation as scope limits of the underlying work.