Harvard systems researcher advancing distributed RL training infrastructure and efficient LLM inference
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) 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: no aggregate publication-count or venue data is available in this snapshot. This axis would measure total research output across venues and years, beyond the two arXiv papers currently reviewed in the Evolve catalog.
Impact—
Not yet scored: citation counts, h-index, and downstream adoption are absent from this snapshot. This axis would measure how widely the work is cited and reused, which cannot be inferred from two recent preprints.
Focus0/100
Scores 0 on focus, the lowest reading on this axis. The snapshot's field distribution splits exactly 50/50 between two distinct areas — machine learning systems (distributed RL infrastructure for LLM training) and efficient LLM inference — and the two 2026 catalog papers, WeightBridge and H-Spec, sit one in each. That even split reads as breadth across the LLM systems stack rather than concentration in a single subfield.
Collaboration54/100
Scores 54. The snapshot records 14 distinct coauthors, with a stable recurring circle including Xuanlin Jiang, Samuel Hsia, Michael Kuchnik, Zachary DeVito, Carole-Jean Wu, and Krishna Teja Chitty-Venkata among twelve frequent collaborators. On both catalog papers Minlan Yu appears as a middle author without corresponding role, consistent with contributor-style work inside engineering-heavy team projects rather than solo authorship.
Activity—
Not yet scored: the snapshot carries no per-year publication volume, so recent-productivity trends cannot be computed. This axis would measure publication cadence over time; the only dated evidence available is two arXiv papers, both from 2026.
Reproducibility38/100
Scores 38, computed from reproducibility assessments of 2 papers (mean 38). Readiness differs by project: H-Spec rates medium — the verdict notes controlled experiments back its claims — while WeightBridge rates low, with MoE-only evaluation and absent competing-system baselines tempering its 5.4-42x gains over Miles. Overall, artifacts look promising but are not yet uniformly easy to reproduce.
Research summary
Minlan Yu is a researcher at Harvard University whose cataloged work sits on the systems side of large language models, split nearly evenly between two complementary areas: distributed reinforcement-learning infrastructure for LLM training, and efficient LLM inference. The Evolve catalog holds two 2026 arXiv papers, on each of which Minlan Yu appears as a middle author without a corresponding role. WeightBridge is an efficient weight transfer library for reinforcement learning, reporting near-ideal transfer across heterogeneous parallelism layouts at 16B-1T scale and 5.4-42x gains over Miles, though its evaluation is MoE-only and lacks competing-system baselines. H-Spec targets parallel speculative decoding, removing the drafter-side KV cache through hybrid target-context injection, with controlled experiments backing its claims. The snapshot records 14 distinct coauthors and a recurring circle of twelve frequent collaborators, including Xuanlin Jiang, Samuel Hsia, Michael Kuchnik, Zachary DeVito, and Carole-Jean Wu, consistent with team-based systems engineering. Bibliometric coverage remains thin: output, impact, and activity axes lack sufficient data to score, and reproducibility averages 38 across the two assessed papers (medium for H-Spec, low for WeightBridge). This profile is name-resolved from public bibliographic data tied to ORCID 0000-0002-2381-0212, and reflects only that snapshot.
Collaboration pattern
Minlan Yu's cataloged work is team-based: both 2026 papers list Minlan Yu as a middle author without corresponding role, within a circle of 14 distinct coauthors and twelve frequent collaborators, including Xuanlin Jiang, Samuel Hsia, Michael Kuchnik, Zachary DeVito, Carole-Jean Wu, Weifan Jiang, and Krishna Teja Chitty-Venkata. The collaboration score is 54. Recurring co-authorship spanning two subfields — RL training infrastructure and inference acceleration — points to a closely cooperating group rather than one-off partnerships.
Research keywords
Highlights
WeightBridge: near-ideal weight transfer across heterogeneous RL layouts
H-Spec: parallel speculative decoding without a drafter-side KV cache
Timeline
- 2026WeightBridge, an efficient weight transfer library for reinforcement learning, appears on arXiv (middle author, non-corresponding).
- 2026H-Spec, parallel speculative decoding without a drafter-side KV cache, appears on arXiv (middle author, non-corresponding).
- 2026Both cataloged papers land in 2026, splitting the field distribution evenly between distributed RL training infrastructure and efficient LLM inference.
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.
- All bibliometric scores and raw metrics derive from a single OpenAlex-based snapshot; coverage may lag recent publications.
- Only the two arXiv papers listed here have been reviewed by Evolve; other work by this author is not reflected in highlights or verdicts.
- Output, impact, and activity axes are unscored because citation and per-year publication data are missing from the snapshot.
- Profile is name-resolved from public bibliographic data matched to ORCID 0000-0002-2381-0212; country is not recorded in the input data.