Learning theorist at UNSW Sydney: sample compression, bandit PAC learning, exact recovery in regression
3
papers here
0
reviewed by Evolve
3
first-author
3
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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: OpenAlex records an h-index of 0 and an i10-index of 0, which likely reflects thin coverage of brand-new arXiv preprints rather than measured output. This axis would score publication volume, venue quality, and record completeness.
Impact—
Insufficient data: these 2026 arXiv preprints have no recorded citations, venue reception, or h-index evidence, so influence cannot be scored yet. This axis would measure citations, uptake by other researchers, and recognition within the learning-theory community.
Focus—
Insufficient data: OpenAlex topic labels concentrate on machine learning and algorithms (4 works) with related themes, but a single-year window is too short to score thematic concentration. This axis would measure consistency of research themes over time.
Collaboration0/100
Collaboration scores 0 in this snapshot: OpenAlex records zero distinct coauthors and lists no frequent coauthors, while all three Evolve-catalog papers place Zhang as first and corresponding author. This is consistent with independently led or solo-authored theory work, but a zero count under an OpenAlex-resolved identity can also reflect under-captured author metadata, so this collaboration footprint should be read as provisional.
Activity47/100
Activity scores 47. OpenAlex attributes four works to the recent three-year window (recent3yWorks: 4), and countsByYear places all four in 2026; three of them, each first-authored and corresponding, are already in the Evolve catalog as arXiv preprints. A single-year burst of three theory preprints points to an early-career researcher in a high-output phase, though the short observation window limits what a yearly pace can yet show.
Reproducibility—
Insufficient data: none of the three catalog papers carries a reproducibility assessment yet (reproReadiness unset), and artifact practices for these theory preprints are unrecorded. This axis would measure code and proof availability and result verifiability.
Research summary
Guangjian Zhang is an early-career learning theorist at UNSW Sydney whose profile is name-resolved from public OpenAlex bibliographic data. The snapshot records four works in the recent three-year window, all appearing in 2026, with an h-index and i10-index of zero — a pattern typical of a researcher whose arXiv preprints have not yet accumulated citations or venue metadata. Three of these works, each first-authored and corresponding, are in the Evolve catalog: an agnostic sample compression scheme for squared loss of near-linear size in the fat-shattering dimension; a study of bandit multiclass PAC learning that corrects prior lower bounds, identifies exact families, and exhibits a confidence direct-sum phenomenon; and exact recovery thresholds for weighted data selection in vector-valued linear regression. Together they sit squarely in statistical learning theory — sample compression, PAC-style guarantees under bandit feedback, and recovery analysis for regression — with OpenAlex topic labels clustering on machine learning and algorithms (4), stochastic gradient optimization techniques (3), and machine learning and data classification (3), plus single appearances of reinforcement learning in robotics and advanced bandit algorithms research. Collaboration records are sparse: zero distinct coauthors are captured and no frequent coauthors are listed, consistent with independently led theoretical work but also with under-captured metadata. The profile should therefore be read as an early-stage record of high velocity concentrated in a single productive year, with impact, output, focus, and reproducibility axes not yet measurable from the available data.
Collaboration pattern
The snapshot records zero distinct coauthors and no frequent coauthors, and all three Evolve-catalog papers list Zhang as first and corresponding author. This is consistent with independently led or solo-authored theory work, but a zero coauthor count from an OpenAlex-resolved identity may also reflect under-captured author metadata. As coverage of these arXiv preprints deepens, the collaboration picture may change materially.
Research keywords
Highlights
Near-linear sample compression for squared loss
First-authored, corresponding arXiv preprint (2026) constructing an agnostic sample compression scheme for squared loss whose size is near-linear in the fat-shattering dimension — a structural compression result for agnostic regression.
Corrected lower bounds for bandit multiclass PAC learning
First-authored, corresponding arXiv preprint (2026) revisiting multiclass PAC learning under bandit feedback: it corrects previously known lower bounds, identifies exact families, and exhibits a confidence direct-sum phenomenon.
Exact recovery thresholds in vector-valued regression
First-authored, corresponding arXiv preprint (2026) deriving exact recovery thresholds for weighted data selection in vector-valued linear regression.
Timeline
- 2026First-authored arXiv preprint: an agnostic sample compression scheme for squared loss of near-linear size in the fat-shattering dimension.
- 2026First-authored arXiv preprint on bandit multiclass PAC learning: corrected lower bounds, exact families, and a confidence direct-sum phenomenon.
- 2026First-authored arXiv preprint: exact recovery thresholds for weighted data selection in vector-valued linear regression.
- 2026OpenAlex counts 4 works in 2026 — the entire recent three-year output falls in this single year, with h-index still at 0.
Papers in the Evolve catalog
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 OpenAlex only; the identity is name-resolved from public bibliographic data without an ORCID, and 'Guangjian Zhang' may correspond to multiple people.
- h-index and i10-index of 0 reflect a snapshot dominated by brand-new 2026 arXiv preprints; citation impact is not yet measurable.
- All three catalog papers are arXiv preprints without peer-review verdicts, catalog keywords, or reproducibility assessments (evolveVerdict, keywords, reproReadiness all unset).
- Output, impact, focus, and reproducibility axes are unscored because underlying data is missing; career history before 2026 is not captured.