Online learning theorist at Google: minimax regret, expert advice, and horizon-uniform optimization guarantees
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
24
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—
Not scored: no reliable total-publication count is available. This axis would measure cumulative works and yearly productivity, but the OpenAlex record may conflate namesakes and no per-paper counts or venue data were supplied, so any output figure would be an estimate rather than a measurement.
Impact—
Not scored: citation-based influence could not be derived for this identity. The snapshot reports h-index 24 and i10-index 33, but impact normally aggregates citations across all works; without per-paper citation data the axis is left unscored rather than estimated.
Focus0/100
Focus scores 0 because attention is not concentrated: the field distribution splits evenly, 0.5 online learning / learning theory and 0.5 online learning and optimization theory. The OpenAlex topic profile is dominated by logic, programming and type systems (37), formal methods in verification (34) and logic, reasoning and knowledge (19), with stochastic gradient optimization at 6 — a mismatch with the catalog fields that suggests disambiguation noise.
Collaboration50/100
Collaboration scores 50 on 11 distinct coauthors in the snapshot. The frequent-coauthor list shows repeat partners — Yang Cai and Di Wang each appear twice — alongside Yanchen Jiang, Christopher Liaw, Aranyak Mehta, Grigoris Velegkas, Anupam Gupta, Guru Guruganesh and Renato Paes Leme. Both reviewed 2026 papers are middle-authored team papers rather than solo works.
Activity47/100
Activity scores 47. The counts-by-year series is sparse in recent years — recent3yWorks is 4 — but 2026 shows a burst of 4 works, including both reviewed arXiv papers. The record reaches back to 1994 (1 work), peaked at 11 in 1999 and 7 in 2002, then thinned to 1–3 works per year across 2013–2020.
Reproducibility100/100
Reproducibility scores 100, the maximum, computed from 2 scored papers with a reproMean of 100. Both Evolve-reviewed papers carry high reproducibility readiness and human-verified proofs: the anytime-regret argument checks out despite asymptotic overhead that vanishes very slowly, and the O(d)-regret proof is short, self-contained and correct on extraction.
Research summary
Vineet Gupta is a researcher at Google (United States) whose Evolve-visible work sits in online learning and optimization theory: the profile's field distribution splits evenly (0.5 / 0.5) between online learning / learning theory and online learning and optimization theory. The catalog contains two 2026 arXiv papers, both human-verified theory with high reproducibility readiness. The first, on prediction with expert advice, gives an anytime algorithm whose regret with many experts matches the fixed-time constant, removing a √2 factor. The second establishes deterministic O(d) regret for proper online inverse optimization, uniform in the horizon, via a trace-power potential. The bibliometric snapshot (OpenAlex identity) reports an h-index of 24 and an i10-index of 33, with publication counts stretching from 1994 (1 work) to a 1999 peak of 11 and 2026 (4 works, matching the recent-3-year total). Radar scores place reproducibility at 100 (reproMean 100 over 2 papers), collaboration at 50 (11 distinct coauthors), and activity at 47, while focus is scored 0 because the topic profile — dominated by logic, programming and type systems (37) and formal methods in verification (34) — does not concentrate on the catalog fields, a mismatch consistent with name-disambiguation noise. Output and impact dimension scores are not yet computable from the supplied data.
Collaboration pattern
The snapshot lists 11 distinct coauthors. Repeat collaborations stand out: Yang Cai and Di Wang each appear twice in the frequent-coauthor list, together with Yanchen Jiang, Christopher Liaw, Aranyak Mehta, Grigoris Velegkas, Anupam Gupta, Guru Guruganesh and Renato Paes Leme. Both reviewed 2026 papers are multi-author works with Gupta in a middle authorship position, consistent with a coequal-team theory collaboration style; the collaboration dimension scores 50.
Research keywords
Highlights
Anytime expert-advice regret matches the fixed-time constant
Deterministic O(d) regret for online inverse optimization
A verified-theory 2026 output burst
Both catalog papers appeared in 2026, matching the snapshot's 4 works that year (also the recent-3yWorks figure). Each was human-verified with high reproducibility readiness, driving the reproducibility dimension to 100 (reproMean 100 over 2 scored papers).
Timeline
- 1994Earliest year in the counts-by-year record, with 1 publication; the OpenAlex activity trail spans three decades.
- 1999Most prolific recorded year, with 11 publications.
- 20027 publications — the second-strongest year in the record.
- 2013Output thins to 1–3 works per year through 2020 (2 in 2013, 3 in 2015), a prolonged low-activity stretch.
- 20264 works recorded (matching recent3yWorks), including both Evolve-reviewed arXiv papers on anytime expert-advice regret and online inverse optimization.
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 are resolved to an OpenAlex identity; 'Vineet Gupta' is a common name, and the topic profile (logic/type systems, formal methods, Petri nets) diverges from the catalog's online-learning papers, so some records may belong to namesakes.
- Output and impact dimension scores are unavailable (dimMissing); only h-index 24 and i10-index 33 are reported in the raw snapshot.
- Evolve has reviewed only 2 papers, both 2026 arXiv preprints with middle authorship; venue, per-paper citation, and career-history data were not provided.
- Counts by year span 1994–2026 with gaps (e.g., 2005, 2007–2008, 2010–2012, 2016, 2021–2025 absent), so the activity record may be incomplete.