LLM alignment and retrieval researcher: zeroth-order preference optimization and attention-based reranking calibration
2
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-20) 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: total publication counts, venue breakdowns and yearly productivity could not be computed from the current snapshot, so this axis — which would measure the sheer volume of research output — is left honestly unscored.
Impact—
No impact score is available: citation counts, per-paper influence and citation-based aggregates are absent from the current data, so this axis — which would measure community uptake of the work — remains unscored.
Focus0/100
The focus score is 0/100, computed directly from the field distribution of the record: exactly half (0.5) of the reviewed output sits in natural language processing / information retrieval and the other half in machine learning (LLM alignment). With no dominant field, concentration is minimal — the profile reads as a two-track researcher spanning reranking behavior analysis and preference-alignment optimization rather than a single-area specialist.
Collaboration50/100
Collaboration scores 50/100, derived from 11 distinct coauthors across the two catalog papers. That network size suggests steady but not yet large-scale teamwork: the alignment paper and the reranking paper draw on overlapping collaborator pools, including Petros Karypis, Noveen Sachdeva, Julian McAuley, Hossein Rajaby Faghihi, Wotao Yin and Tianyi Lin. The list mixes alignment- and optimization-oriented names, consistent with the even 50/50 split between the two fields.
Activity—
No activity score is available: recent publication cadence and year-over-year output trends cannot be computed from the current snapshot, so this axis — which would measure ongoing research pace over time — is left honestly unscored.
Reproducibility55/100
Reproducibility is scored 55/100, the mean across exactly two scored papers, both rated medium on reproducibility readiness. Neither the zeroth-order alignment work nor the reranking-calibration study is yet rated high readiness: ComPO's dependence on comparison oracles and judge models, and the reranking study's reliance on instruction-following LLM rerankers, leave replication friction that the medium ratings reflect. The score rests on only two papers, so it should be read as provisional.
Research summary
Peter Chen is an early-stage profile built from public bibliographic data, currently anchored by two 2026 arXiv papers reviewed in the Evolve catalog. The record splits evenly — 50/50 — between natural language processing / information retrieval and machine learning (LLM alignment), which explains a focus score of 0/100: the work spans two adjacent areas rather than concentrating in one. The first-authored paper, 'A Zeroth-Order Paradigm for LLM Preference Alignment,' introduces ComPO, which uses comparison oracles on low-margin preference pairs; Evolve's review notes promising gains across LLM benchmarks while flagging an unquantified theory-practice gap and judge variance. The middle-authored paper, 'How Calibration Content Shapes Attention-Based Reranking,' finds that null-query calibration breaks on per-query instructions and that training-free interpolated nulls fix it, though RankGPT still wins on FollowIR. Both papers carry a medium reproducibility rating, averaging 55/100 across the two, which sets the reproducibility score. Collaboration scores 50/100 on a network of 11 distinct coauthors, including Petros Karypis, Noveen Sachdeva, Julian McAuley, Hossein Rajaby Faghihi, Wotao Yin and Tianyi Lin. Output, impact and activity axes lack the underlying bibliometric data and are left unscored. Because 'Peter Chen' is a common name, this profile is name-resolved from public bibliographic records; institution, country and ORCID are unverified.
Collaboration pattern
With 11 distinct coauthors behind the two catalog papers, Peter Chen's collaboration score is 50/100 — a mid-sized, plausibly expanding network. Recurring names include Petros Karypis, Noveen Sachdeva, Julian McAuley, Hossein Rajaby Faghihi, Rui Zhu, Yan Zhu, Xi Chen, Wotao Yin and Tianyi Lin; one listed entry duplicates the researcher's own name, a likely name-resolution artifact. The mix of alignment- and optimization-oriented collaborators mirrors the 50/50 field split.
Research keywords
Highlights
ComPO: comparison oracles for low-margin preference pairs
Training-free interpolated nulls repair null-query calibration
An even split between alignment and retrieval
Across the two reviewed papers the field distribution is exactly 50/50 between machine learning (LLM alignment) and natural language processing / information retrieval — optimization-driven alignment methods on one side, empirical reranking behavior analysis on the other, yielding the 0/100 focus score.
Timeline
- 2026Publishes first-authored 'A Zeroth-Order Paradigm for LLM Preference Alignment' on arXiv, introducing ComPO with comparison oracles for low-margin preference pairs.
- 2026Co-authors 'How Calibration Content Shapes Attention-Based Reranking', showing that interpolated null queries repair null-query calibration failures under per-query instructions.
- 2026The catalog record splits evenly between LLM alignment and NLP/IR, with both papers rated medium for reproducibility readiness (mean 55/100).
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 scores are derived from OpenAlex-sourced public bibliometric records only; coverage beyond the two cataloged papers is not represented in this profile.
- 'Peter Chen' is a common name: this profile is name-resolved from public bibliographic data, and the frequent-coauthor list even contains a duplicate of the researcher's own name; institution, country and ORCID are unverified.
- Output, impact and activity dimensions are unscored because the underlying publication-volume, citation and cadence data are missing from the snapshot.
- Both catalog papers are 2026 arXiv preprints with medium reproducibility ratings; peer-reviewed venue status is not reflected in this data.