Optimization theorist: lower bounds for asynchronous stochastic optimization and variance reduction for flow matching
2
papers here
2
reviewed by Evolve
2
first-author
2
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-20) 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—
Output metrics are not yet available: no publication counts, venue record, or productivity statistics underlie this axis. It would measure total scholarly output, venue placement, and output volume across the researcher's career once fuller bibliographic data is linked.
Impact—
Impact data is insufficient to score this axis: no citation counts or uptake statistics are on file. It would measure how widely the asynchronous-optimization lower bounds and flow-matching variance-reduction results are cited, reused, and built upon by other research groups.
Focus0/100
The radar records a focus score of 0, with the field distribution split exactly 0.5/0.5 between optimization for generative models and distributed/asynchronous stochastic optimization. Both branches share an optimization-theory core of bounds and convergence analysis, but a two-paper sample makes any concentration reading provisional.
Collaboration0/100
Collaboration scores 0 on the radar, with zero distinct coauthors recorded and no frequent collaborators listed. Both reviewed papers name Tyurin as first and corresponding author; given only two 2026 papers, this reflects catalog coverage limits rather than a measured solo working pattern.
Activity—
Activity data is missing: the underlying longitudinal record of papers per year is not available, so no cadence score is computed. The axis would measure publication rate, recency, and consistency of output across recent years, based on a complete yearly timeline.
Reproducibility55/100
Reproducibility scores 55, the mean readiness rating across the two reviewed papers, both rated medium. Both works are theory-first — lower bounds and one-dimensional variance bounds — and the flow-matching paper reports only modest real-data gains, consistent with a mid-level readiness profile.
Research summary
Alexander Tyurin is an optimization theorist whose reviewed work in Evolve's catalog spans two adjacent areas: distributed and asynchronous stochastic optimization, and optimization for generative models — the recorded field distribution splits exactly 0.5/0.5 between them. Two 2026 arXiv papers, both first-authored and corresponding, anchor the profile. The first, 'Bridging the Gap Between Homogeneous and Heterogeneous Asynchronous Optimization Is Surprisingly Difficult,' proves that standard similarity and weak-interpolation assumptions cannot close the time-complexity gap between homogeneous and heterogeneous asynchronous settings, while strong interpolation combined with a local PL condition suffices — work Evolve characterizes as rigorous and honest theory. The second, 'Same Flow, Different Paths: Variance Reduction in Flow Matching,' shows that paths inducing identical flow-matching objectives can produce very different SGD variance, establishing tight one-dimensional bounds and feasible path constraints, though real-data gains remain modest. Both papers carry a medium reproducibility readiness rating, averaging 55 across the two. Radar dimensions for output, impact, and activity are missing for lack of underlying data, and the collaboration record lists zero distinct coauthors. The profile is name-resolved from public bibliographic data.
Collaboration pattern
The current record lists zero distinct coauthors and no frequent collaborators; both 2026 papers name Tyurin as first and corresponding author, a pattern consistent with independently led theory projects. Because the sample is only two arXiv papers, the empty collaboration graph likely reflects catalog coverage limits rather than a confirmed solitary style, and the radar collaboration score of 0 should be read with that caveat.
Research keywords
Highlights
Proving the async homogeneous–heterogeneous gap is hard to close
Identical flow-matching objectives, very different SGD variance
Timeline
- 2026Publishes a first-authored, corresponding arXiv paper proving that standard similarity and weak-interpolation assumptions cannot close the asynchronous homogeneous–heterogeneous time-complexity gap.
- 2026Publishes a first-authored, corresponding arXiv paper on variance reduction in flow matching, establishing tight one-dimensional bounds on SGD variance across paths.
- 2026Both reviewed papers appear in the same year, split evenly between asynchronous optimization theory and optimization for generative models, each rated medium for reproducibility readiness.
Papers in the Evolve catalog
- first authorcorresponding
Bridging the Gap Between Homogeneous and Heterogeneous Asynchronous Optimization Is Surprisingly Difficult
- first authorcorresponding
Same Flow, Different Paths: Variance Reduction in Flow Matching
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 context is thin: no citation counts, venue record, affiliation, country, or ORCID is available; the profile is name-resolved from public bibliographic data plus the two reviewed arXiv papers.
- Output, impact, and activity radar dimensions are missing because the underlying productivity and citation data is insufficient; they are reported as unscored, not estimated.
- Collaboration and focus scores of 0 rest on a two-paper snapshot and should not be read as established career patterns.
- The reproducibility mean of 55 covers only the two Evolve-reviewed papers, both rated medium readiness.