Statistical learning theorist at Purdue sharpening PAC sample complexity and SVM generalization bounds
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
18
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: the current extract lacks full productivity metrics such as total work counts, venue distribution, and long-run output trajectory. This axis would measure publication volume and its growth across the researcher's career, beyond the coarse yearly counts already on file.
Impact—
Not yet scored: only coarse h-index (18) and i10-index (26) snapshots are on file, with citation-level detail pending. This axis would measure citation influence — citations per work, highly cited papers, and field-normalized attention across the 2004–2026 record.
Focus100/100
Score 100, reflecting tight topical concentration. The field distribution lists only statistical learning theory (weight 1), and both reviewed 2026 papers sit squarely in that field. Topic counts are led by Machine Learning and Algorithms (139) with adjacent theory strands — bandit algorithms (61), domain adaptation and few-shot learning (52), stochastic gradient optimization (51) — all within learning theory rather than applied areas.
Collaboration28/100
Score 28. Despite 53 works in the last three years, only 3 distinct recent coauthors appear — a strikingly small network for that output volume. The frequent-coauthor list names Aryeh Kontorovich and Amirreza Shaeiri, and both reviewed 2026 papers are first-authored. The pattern points to small-team or solo theoretical writing, consistent with the low collaboration score despite maximal activity.
Activity100/100
Score 100, backed by 53 works in the last three years. Annual counts climb from 15 (2022) and 15 (2023) to a peak of 22 (2024), then 12 (2025) and 19 already in 2026 — well above earlier levels such as 9 in 2015 and 10 in 2018. This recent trajectory is the strongest in the 2004–2026 record and sustains a maximal activity rating.
Reproducibility55/100
Score 55, computed from 2 reproducibility-assessed papers (mean 55). Both reviewed works carry medium reproducibility readiness: the robust PAC paper proves optimal VC rates via binomial bagging but offers an impractical algorithm and no experiments; the SVM margin-bound paper yields a rate-1/m high-probability bound with huge constants (~648). The work is rigorous, self-contained theory, with empirical replication paths limited by design.
Research summary
Steve Hanneke is a researcher at Purdue University (West Lafayette, US) whose OpenAlex-resolved profile centers on statistical learning theory. The record spans 2004–2026 with an h-index of 18 and an i10-index of 26. Output has accelerated sharply: 15 works in 2022, 15 in 2023, a peak of 22 in 2024, 12 in 2025, and 19 already in 2026 — 53 works over the last three years, produced with only 3 distinct recent coauthors. Topic concentrations include Machine Learning and Algorithms (139 works), advanced bandit algorithms (61), domain adaptation and few-shot learning (52), stochastic gradient optimization (51), and ML-based data classification (40). Evolve has reviewed two 2026 first-authored arXiv papers, both in statistical learning theory. The first proves adversarially robust PAC learning needs no extra samples versus classical VC rates, via binomial bagging — elegant, though with an impractical algorithm and no experiments. The second establishes a rate-1/m high-probability risk bound for the homogeneous hard-margin SVM through a novel KKT-entropy deletion-counting argument, with reviewers noting huge constants (~648) and narrow scope. Both carry medium reproducibility readiness, reflecting purely theoretical contributions. Evolve's radar scores activity and focus at 100, reproducibility 55, and collaboration 28.
Collaboration pattern
The collaboration footprint is narrow: 53 works in the last three years with only 3 distinct recent coauthors, and a radar collaboration score of 28. Aryeh Kontorovich and Amirreza Shaeiri are the named frequent coauthors, and both reviewed 2026 papers are first-authored without corresponding-author flags. The pattern fits a theorist writing largely solo or with one or two stable partners rather than in large teams.
Research keywords
Highlights
Optimal VC rates for adversarially robust PAC learning
A rate-1/m high-probability risk bound for realizable hard-margin SVM
A surge of theoretical output since 2022
Annual recorded works jumped from single digits or low teens through most of 2004–2021 (e.g. 9 in 2015, 10 in 2018) to 15 in 2022, 15 in 2023, and a peak of 22 in 2024, with 53 works in the last three years — the steepest sustained rise in the OpenAlex record.
Timeline
- 2004Earliest recorded work in the OpenAlex profile (1 publication).
- 2015Nine works recorded — a then-high watermark in a modest early output curve.
- 2018Ten works recorded, another local peak before the recent acceleration.
- 2022Output accelerates: 15 works recorded, matched again in 2023.
- 2024Peak year with 22 works recorded.
- 2026Two first-authored arXiv papers reviewed by Evolve — optimal VC rates for adversarially robust PAC and sharp margin bounds for realizable SVM; 19 works already recorded this year.
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.
- Bibliometric figures come from the OpenAlex record only; citation-level output and impact dimensions remain unscored.
- Only two papers have been reviewed by Evolve, both 2026 arXiv preprints; peer-review status is not reflected in this profile.
- Reproducibility rests on just 2 assessed papers (mean 55) and may not generalize across the full record.
- The frequent-coauthor list includes a self-entry (Steve Hanneke), likely a name-resolution artifact; 3 distinct recent coauthors are reported.