UCLA machine learning researcher bridging PAC-Bayes system identification and causal Bayesian optimization
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
—
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—
Insufficient data to score: the catalog contains only two 2026 arXiv preprints, with no publication counts, venue history, or authorship distribution over time. This axis would measure total output volume and recency.
Impact—
Not yet measurable: no citation counts, h-index, or usage signals accompany these recently posted preprints. This axis would measure citations and uptake of the PAC-Bayes identification work and the causal Bayesian optimization survey.
Focus0/100
The focus score sits at 0 because the field distribution is an even 50/50 split between "machine learning — learning theory and system identification" and "machine learning — causal inference and Bayesian optimization", one paper in each. The record therefore spans two subfields rather than converging on a single specialization, though they share methodological ground: concentration bounds, Gaussian processes, and structural causal models connect both threads.
Collaboration32/100
The collaboration score is 32, based on 4 distinct coauthors recorded across the reviewed papers. The recurring names are George Michailidis, Thuy T. Le, Zixuan Ma, and Hien Tran, suggesting a small, stable working group rather than a broad international network. Both 2026 catalog papers are first-authored without corresponding-author status. This compact collaboration footprint is consistent with an early-stage research record.
Activity—
Cannot be scored yet: the only dated evidence is two arXiv postings in 2026, with no multi-year output series or submission cadence. This axis would measure recent publication rate and sustained research activity.
Reproducibility100/100
Reproducibility scores 100, the maximum, computed as a mean of 100 across 2 assessed papers. Both the PAC-Bayesian meta-learning paper and the causal Bayesian optimization survey carry high reproducibility readiness in Evolve's review. The survey's honest benchmarking — reporting that no method dominates — and the identification paper's clearly scoped model-matched synthetic experiments support this strong openness signal.
Research summary
Chenfeng Huang is a machine learning researcher at the University of California, Los Angeles, whose Evolve profile currently rests on two 2026 arXiv preprints, both first-authored. The first applies PAC-Bayesian meta-learning to few-shot identification of linear dynamical systems, building guarantees via martingale concentration; Evolve's review reads the theory as a diagnostic, with empirical gains resting on model-matched synthetic tests. The second is a survey of causal Bayesian optimization that contributes a unified taxonomy across foundations, methods, and applications, plus an honest benchmark in which no method dominates and standard Bayesian optimization stays competitive, though benchmark coverage is partial. Both papers carry high reproducibility readiness, yielding a reproducibility mean of 100 across two assessed papers. The record splits evenly — 0.5 and 0.5 — between learning theory and system identification on one side and causal inference and Bayesian optimization on the other, linked by shared statistical machinery: concentration bounds, Gaussian processes, and structural causal models. Collaboration is compact: four distinct coauthors are recorded — George Michailidis, Thuy T. Le, Zixuan Ma, and Hien Tran — and both papers are first-authored without corresponding status. Output, citation impact, and activity-over-time metrics are not yet available for this profile.
Collaboration pattern
The record shows a compact collaboration pattern: 4 distinct coauthors across the reviewed first-author papers, with George Michailidis, Thuy T. Le, Zixuan Ma, and Hien Tran recurring. Both 2026 papers are first-authored without corresponding status, consistent with work led within a small, stable methodological group spanning system identification and causal optimization rather than a large distributed network.
Research keywords
Highlights
PAC-Bayes guarantees for few-shot system identification
Unified taxonomy for causal Bayesian optimization
Honest benchmark: no CBO method dominates
Timeline
- 2026First-authored paper on PAC-Bayesian meta-learning for few-shot identification of linear dynamical systems is posted to arXiv.
- 2026First-authored survey "Causal Bayesian Optimization: Foundations, Methods, and Applications" is posted to arXiv.
- 2026Both papers enter the Evolve catalog with high reproducibility readiness, giving a reproducibility mean of 100 across 2 assessed papers.
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.
- Radar scores are computed from public bibliometric data; output, impact, and activity axes lack sufficient underlying data and are not scored.
- The profile rests on two 2026 arXiv preprints in the Evolve catalog; citation counts and peer-reviewed venue outcomes are not yet reflected.
- Name is resolved from public bibliographic data (ORCID 0000-0003-3779-525X); the Chinese character form of the name is unconfirmed.
- Both catalog entries list arXiv as source; peer-review status is not recorded in the reviewed data.