Early-career theory researcher at Peking University proving complexity guarantees for nonsmooth Langevin sampling
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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—
Insufficient data to score: this axis would measure total publication volume, venue quality, and productivity trend over time. The OpenAlex record here is thin — five works indexed in 2026 — so no output rating is assigned yet.
Impact—
Not yet scorable: impact would track citation counts, h-index and i10-index. Both reviewed papers are 2026 arXiv preprints and the recorded h- and i10-indexes are 0, so citation evidence has not yet accumulated.
Focus0/100
The field distribution is extremely narrow: it splits evenly (0.5/0.5) between two near-identical labels for machine-learning-theory work on sampling algorithms and computational statistics. Topic counts reinforce this — Markov chains and Monte Carlo methods (5), stochastic gradient optimization techniques (4), geometric analysis and curvature flows (4) — indicating a single agenda on Langevin-type sampling with analytic tools. The computed focus score is 0, which here signals that the indexed record (one year, 5 works) is too small for the topical-diversity measure to resolve, not evidence of scattered interests.
Collaboration28/100
The collaboration footprint is compact: 3 distinct coauthors, with the raw score at 28. Both reviewed papers are first-author arXiv preprints without corresponding-author roles, and the recurring partners are Jingxin Zhan and Zhihua Zhang. Topic overlap — Markov chains and Monte Carlo methods (5), stochastic gradient optimization (4) — suggests a small, stable theory group rather than a distributed network; more co-authorship and corresponding roles would move this axis.
Activity52/100
Recorded activity concentrates entirely in 2026: the counts-by-year data shows 5 works in 2026, recent3yWorks equals 5, and the raw activity score is 52 (mid-range). The two reviewed first-author preprints both date to 2026 and sit in sampling-algorithm theory, so output pace looks steady, but the observation window is a single year — any trend claim would be premature until more indexed years accumulate.
Reproducibility38/100
Evolve's reproducibility read averages 38 across 2 scored papers. The Rényi-tracking preprint is rated medium: coherent theory with conservative constants and no experiments. The MYULA Poisson-corrector preprint is rated low: rigorous and novel, but key proofs were absent from the reviewed excerpt. The pattern is theory-first work whose artifacts — code, experiments, complete proof chains — are not yet independently verifiable, which caps readiness despite strong mathematical coherence.
Research summary
Yuchen Xin (Peking University) is an early-career researcher in the theory of sampling algorithms, working at the interface of machine learning theory and computational statistics. This profile is name-resolved from OpenAlex bibliographic data, and the record is thin but coherent: five works indexed in 2026, two of which Evolve has reviewed in depth, both first-author arXiv preprints. The agenda targets nonsmooth sampling — Langevin Monte Carlo when the target density is nonsmooth or updated discretely. One preprint proves the first Rényi tracking bounds for Langevin Monte Carlo under discrete target updates and develops Moreau-envelope machinery for nonsmooth sampling; the other bounds the MYULA invariant bias by O(h)+Õ(h^{3/4}) via a Poisson-equation corrector, giving Õ(ε^{-4/3}) complexity. Topic counts corroborate the concentration: Markov chains and Monte Carlo methods (5), stochastic gradient optimization techniques (4), geometric analysis and curvature flows (4). Collaboration is compact — 3 distinct coauthors, with repeated work alongside Jingxin Zhan and Zhihua Zhang. Citation impact cannot yet be scored: h- and i10-indexes stand at 0. Evolve's reproducibility read is cautious (mean 38 over 2 papers), reflecting theory-first papers without experiments and incomplete proof coverage in one reviewed excerpt.
Collaboration pattern
The record shows a compact collaboration pattern: 3 distinct coauthors, with repeated work alongside Jingxin Zhan and Zhihua Zhang. Both cataloged papers are first-authored arXiv preprints without corresponding-author roles, consistent with a small theory group at Peking University rather than a large distributed network. The raw collaboration score is 28, and with only one indexed year (2026, 5 works) the picture may undercount informal mentorship and ongoing joint projects.
Research keywords
Highlights
First Rényi tracking bounds for Langevin dynamics with moving targets
Poisson-corrector analysis yields Õ(ε^{-4/3}) complexity for MYULA
A coherent program in nonsmooth Langevin sampling
Both 2026 first-author preprints attack the same problem — sampling when the target density is nonsmooth — via Moreau envelopes and Moreau–Yosida regularization, backed by log-Sobolev and Wasserstein tools. No single paperSlug applies: this is a cross-paper observation, and both works still lack experimental validation.
Timeline
- 2026First-author arXiv preprint: Rényi tracking bounds for Langevin dynamics with moving targets, introducing Moreau-envelope nonsmooth sampling.
- 2026First-author arXiv preprint: Poisson-corrector complexity bounds for Moreau–Yosida unadjusted Langevin sampling, giving Õ(ε^{-4/3}) complexity.
- 2026Five works indexed for 2026 in the OpenAlex snapshot; both cataloged papers are first-authored, with 3 distinct coauthors.
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.
- Bibliometrics come from an OpenAlex snapshot only; no ORCID is linked, so this profile is name-resolved from public bibliographic data and may merge or split same-name authors.
- All cataloged output is 2026 arXiv preprints; h-index (0) and i10-index (0) reflect recency rather than quality, and no citation-based impact is scored.
- The output and impact radar axes are unscored because the indexed record covers a single year with 5 works.
- The reproducibility mean (38) is based on 2 Evolve-reviewed papers; verdicts note conservative constants, no experiments, and proofs missing from the reviewed excerpts.