Theory-first ML researcher at Shihezi University: minimax lower bounds and provably debiased LLM distillation.
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
6
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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—
No output score is available yet: the record lacks a validated count of peer-reviewed publications for this identity. This axis would measure cumulative verified output volume; present evidence covers only 18 recent works and 2 reviewed papers.
Impact—
No impact score is available yet: per-paper citation data is missing, so influence cannot be graded despite raw h-index 6 and i10-index 4. This axis would measure citation uptake and the scholarly influence of the work.
Focus0/100
Focus scores 0 — the lowest concentration signal — and the topic list explains why: five topics carry near-even weight, with Machine Learning and Algorithms (7) and Advanced Bandit Algorithms Research (7) tied on top, followed by Reinforcement Learning in Robotics (5), Advanced Optical Imaging Technologies (5), and Stochastic Gradient Optimization Techniques (4). The 50/50 field split between statistics/ML theory and LLM-distillation theory reinforces the breadth.
Collaboration28/100
The collaboration score of 28 reflects a compact network: 3 distinct coauthors across 18 works in the last three years. The recurring names in the record are David Simchi-Levi and Yuheng Zhang, and both Evolve-reviewed papers are first-authored (non-corresponding). This is the footprint of a small working circle rather than a large experimental lab.
Activity86/100
Activity is the strongest axis at 86, backed by raw velocity: 18 works in the last three years, with OpenAlex yearly counts climbing from 3 (2024) to a peak of 8 (2025) and 7 already recorded in 2026, on top of h-index 6 and i10-index 4 in a record dating back to 2016.
Reproducibility20/100
Reproducibility scores 20 (reproMean 20 across 2 repro-scored papers), matching Evolve's low repro-readiness ratings on both catalog entries: the distillation paper is explicitly theory-only with no experiments, and the minimax paper contributes a four-experiment lower-bound construction rather than a runnable method. Empirical release is the clear gap.
Research summary
Haichen Hu is a theory-oriented machine learning researcher at Shihezi University (CN) whose profile, resolved from OpenAlex bibliographic data, splits evenly between theoretical statistics / machine learning theory and the theory of LLM distillation and post-training. The Evolve catalog contains two 2026 arXiv papers, both first-authored. The first proves a sharp, structure-agnostic minimax lower bound for partial linear models via a clean four-experiment construction matching Gu (2026)'s upper bound, resolving Gu (2025)'s open problem; Evolve judges the proof sound but assumption-heavy. The second proposes Coupled Calibration and Learning (CCL), which provably distills an unbiased oracle student from a biased teacher using only source-domain reward feedback, with covariate-shift analysis and KL regularization; Evolve flags rigorous but assumption-heavy theory with no experiments. Bibliometric signals show high velocity with an early-stage citation footprint: h-index 6, i10-index 4, 18 works in the last three years, and a yearly curve rising from 3 works (2024) to 8 (2025) with 7 already indexed in 2026. Topic spread is wide — bandit algorithms, RL in robotics, optical imaging, and stochastic gradient optimization — producing a focus score of 0 against an activity score of 86. The coauthor network is small (3 distinct coauthors over three years, with David Simchi-Levi and Yuheng Zhang recurring), and reproducibility is rated low (mean 20), consistent with theory-only output.
Collaboration pattern
Collaboration is compact and theory-centric: only 3 distinct coauthors across 18 works in the last three years, yielding a collaboration score of 28. The recurring names in the record are David Simchi-Levi and Yuheng Zhang. Both Evolve-reviewed papers are first-authored but not corresponding, and the topics involved — minimax theory and distillation — are single-thread proof projects rather than lab-scale efforts, consistent with a small working circle.
Research keywords
Highlights
Sharp structure-agnostic minimax lower bound for partial linear models
Provably unbiased LLM distillation from biased teachers (CCL)
Timeline
- 2016Earliest activity in the OpenAlex record: 4 works indexed in 2016, the highest single-year count until 2025.
- 2024Output stabilizes at 3 indexed works after sparse years (1-2 per year in 2017, 2018, 2022, 2023), beginning the current ramp-up.
- 2025Output peaks at 8 indexed works, the highest annual count in the record and the base of the 86 activity score.
- 2026Two first-author arXiv preprints enter the Evolve catalog — the partial-linear minimax bound and the CCL distillation framework — with 7 works already indexed 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.
- All bibliometrics (h-index 6, i10-index 4, yearly counts, topics) come from an OpenAlex-resolved author identity; name disambiguation may merge or split records, so the profile is name-resolved from public bibliographic data.
- Only the two 2026 arXiv papers have Evolve-reviewed verdicts; other indexed works are counted but not content-assessed.
- Output and impact dimension scores are missing for insufficient data; per-paper citation counts are not available in the input.
- The record lists Shihezi University (CN) as the affiliation and offers no further biographical detail, so no claims about career stage or awards are made.