Northwestern researcher bridging LLM agents, statistical evaluation limits, and calibrated travel simulation
3
papers here
3
reviewed by Evolve
3
first-author
1
corresponding
—
works · OpenAlex
0
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—
Insufficient data: this axis would measure publication volume, venues, and completeness of the bibliographic record over time, but citation-indexed output totals are not yet available for this profile — h- and i10-indices are recorded as 0 and flagged unreliable.
Impact—
Insufficient data: this axis would measure citations, downstream uptake, and influence of the work, but citation counts are not yet reliably captured for this profile, so impact is left unscored rather than estimated from partial signals.
Focus0/100
Scores 0 because recorded work splits evenly across three fields — machine learning (LLM-based agents), statistics for machine learning evaluation, and transportation systems modeling/ML, each at 33.3% — while OpenAlex topics disperse further (AI-based problem solving and planning: 2; speech and dialogue systems: 2; topic modeling, software engineering research, transportation planning and optimization: 1 each). The zero reflects breadth, not absence of direction.
Collaboration44/100
Scores 44 on collaboration. The profile records 8 distinct co-authors, with Runjia Du, Zeming Liu, Qibai Chen, Hang Lyu, Yankai Zeng, Bojun Lin, and Yilan Wei appearing most frequently. All three catalog papers are first-authored (CALM additionally corresponding), indicating a lead-author working pattern supported by a moderate-sized, stable co-author network rather than large consortium collaboration.
Activity52/100
Scores 52 on activity. OpenAlex records 5 works in the last three years, and the by-year counts place all 5 in 2026 — matching the three 2026 arXiv preprints in the Evolve catalog. Output is recent and clustered rather than accumulated over a long track record, consistent with an early-stage publishing rhythm.
Reproducibility70/100
Scores 70, computed as the mean readiness across 3 assessed papers. Evolve rates the hard-budget evaluation paper as high readiness, while the agent-memory and CALM papers are medium — the former resting on one templated benchmark and a single model, the latter showing strong holdout calibration (JSD 0.156 to 0.004) but with live-LLM tests that are small and no code stated.
Research summary
Yezhou Cheng is a researcher at Northwestern University whose profile, name-resolved from public OpenAlex bibliographic data, spans three evenly weighted areas: LLM-based agents, statistics for machine learning evaluation, and transportation systems modeling. The Evolve catalog holds three 2026 arXiv preprints, each with Cheng as first author — and corresponding author on the CALM traveler-simulation framework. The first proposes scoped retrieval to prevent cross-family interference in persistent agent memory; the second proves sharp replication limits for hard-budget repeated LLM evaluation via minimax lower bounds and stratified designs; the third, CALM, couples LLM choice networks with calibrated discrete choice models, improving holdout calibration from JSD 0.156 to 0.004. OpenAlex records 5 works in the last three years, 8 distinct co-authors, and a reproducibility mean of 70 across 3 assessed papers. Citation-side metrics (h-index, i10-index) are recorded as 0 and flagged insufficient, so output and impact are not scored. The profile reads as early-stage and method-focused: small in volume, concentrated in 2026, and consistently oriented toward honest evaluation and reproducible methodology.
Collaboration pattern
Cheng publishes as lead author: first author on all three catalog papers and corresponding author on CALM. The recorded network is moderate — 8 distinct co-authors — with Runjia Du, Zeming Liu, Qibai Chen, Hang Lyu, Yankai Zeng, Bojun Lin, and Yilan Wei as the most frequent collaborators. Co-authorship spans all three of the profile's fields, suggesting small focused teams rather than consortium-style collaboration, and the collaboration dimension scores 44.
Research keywords
Highlights
Sharp replication limits for hard-budget LLM evaluation
First-author arXiv preprint proving sharp limits for honest uncertainty in repeated evaluation under hard budget constraints, with matching designs, equal-budget evidence, and minimax lower bounds; Evolve rates its reproducibility readiness high while noting practical gains are regime-dependent.
Scoped retrieval against cross-family interference in agent memory
First-author work arguing that scoped retrieval prevents cross-family interference in persistent memory for self-improving LLM agents; Evolve finds the paired evidence solid but drawn from a single templated benchmark and one model (reproducibility readiness: medium).
CALM: calibrated LLM choice network for traveler simulation
First- and corresponding-author framework coupling LLM agents with calibrated discrete choice models for activity-based traveler simulation; holdout calibration improves sharply from JSD 0.156 to 0.004, though live-LLM tests remain small and no code is stated.
Timeline
- 2026OpenAlex records 5 works in 2026 — the entire recent-3-year output for this profile — marking 2026 as the entry year into the indexed record.
- 2026Publishes a first-author arXiv preprint proposing scoped retrieval to prevent cross-family interference in persistent agent memory.
- 2026Publishes a first-author arXiv preprint proving sharp limits for honest uncertainty in hard-budget repeated evaluation, rated high for reproducibility readiness.
- 2026Publishes CALM as first and corresponding author — a calibrated LLM choice network framework for activity-based traveler simulation.
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.
- Bibliometrics are drawn from OpenAlex only; no ORCID is linked, so this profile is name-resolved from public bibliographic data.
- h-index and i10-index are recorded as 0 with output and impact flagged as insufficient — likely reflecting thin citation coverage for a 2026-heavy record rather than a measured absence of citations.
- All three catalog papers are 2026 arXiv preprints; no peer-reviewed venue information is present in this snapshot.
- Frequent-coauthor data comes as provided by the platform; the duplicate "Yilan Wei" entry was collapsed, and no further co-author metadata was available.