Yi Chang

University of Waterloo · Canada

Unclaimed · AI-structured profileOpenAlex

University of Waterloo researcher building benchmarks that stress-test LLM recommenders and multimodal models

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

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—

Output scoring is unavailable: this snapshot lacks the complete, venue-resolved publication record the axis requires. It would measure total indexed outputs, their spread across venues and years, and the balance between arXiv preprints and peer-reviewed papers.

Impact—

Impact cannot yet be scored: the snapshot's citation footprint is minimal (h-index 1, i10-index 0) on a very thin record. This axis would measure citation uptake, h- and i10-index trajectories over time, and uptake beyond the author's own circle.

Focus—

No focus score is available yet; insufficient topical metadata prevents computation. The axis would measure how concentrated the research program is — the balance between core benchmark-evaluation work and adjacent areas such as topic modeling and vision-imaging applications.

Collaboration36/100

Collaboration scores 36/100, indicating a compact network: only 5 distinct coauthors appear in the snapshot. All three 2026 catalog papers place Chang in the last author position, working repeatedly with Yuan Wu, Gengxu Li, Zhongru Chen, Jinzhe Li, and Jinnan Li — a small, stable benchmark-building team rather than a large distributed collaboration.

Activity87/100

Activity scores 87/100, driven by a sharp recent burst: the countsByYear record shows 1 work each in 2018 and 2019, 3 in 2025, then 16 in 2026, with 19 works across the last three years. The trajectory points to a current, sustained push of benchmark papers onto arXiv.

Reproducibility—

No axis-level reproducibility score exists yet; aggregated signals on code release, dataset hosting, and artifact availability are missing. The only hint is Evolve's per-paper 'medium' readiness rating for both reviewed benchmarks. This axis would measure replication support across outputs.

Research summary

Yi Chang is a researcher at the University of Waterloo whose recent work centers on building evaluation benchmarks for large language models and multimodal systems. The OpenAlex-derived snapshot records 19 works in the last three years, concentrated in 2026 (16 works) after single indexed works in 2018 and 2019 and three in 2025 — an activity profile scoring 87/100. Topical weight places Topic Modeling and Multimodal Machine Learning Applications jointly first (7 works each), followed by Advanced Neural Network Applications and Advanced Vision and Imaging (4 each) and Natural Language Processing Techniques (3). In the Evolve catalog, Chang appears as last author — though not corresponding author — on three 2026 arXiv benchmark papers. RPCBench (4,623 instances) tests whether LLM-based recommenders proactively flag faulty user premises; Evolve rated its taxonomy and metrics valuable while flagging LLM generator-judge circularity. The OCR-MetaReasoning Benchmark (1,500 samples) probes meta-reasoning in text-rich image understanding, finding deduction and layout hardest for multimodal LLMs, with contamination and abductive-only edits as limits. MMPCBench extends the proactive-critique program to flawed multimodal inputs and awaits review. Collaboration is small-team — 5 distinct coauthors, including recurring partners Yuan Wu, Gengxu Li, Zhongru Chen, Jinzhe Li, and Jinnan Li (collaboration score 36). The citation footprint is minimal (h-index 1, i10-index 0), so this reads as an early-stage, output-driven profile name-resolved from public bibliographic data.

Collaboration pattern

Collaboration is tight and stable: 5 distinct coauthors in the snapshot and a collaboration score of 36/100. Chang sits in the last author position on all three 2026 catalog papers, alongside recurring collaborators Yuan Wu, Gengxu Li, Zhongru Chen, Jinzhe Li, and Jinnan Li — a pattern consistent with a small benchmark-building group whose senior author shapes the evaluation agenda.

Research keywords

LLM evaluationbenchmark constructionLLM-based recommendationpremise critiquemultimodal large language modelsOCR reasoningmeta-reasoningproactive error detectionLLM-as-a-judgetext-rich image understandingtopic modelingevidence faithfulness

Highlights

RPCBench: stress-testing whether LLM recommenders catch faulty premises
A 4,623-instance benchmark asks LLM-based recommenders to proactively flag flawed user premises. Evolve's verdict: the taxonomy and metrics are valuable, but generator-judge circularity — the same LLM family producing and judging errors — weakens the results. Reproducibility readiness is rated medium.
OCR-MetaReasoning: where multimodal LLMs stumble on text-rich images
A rigorous 1,500-sample benchmark measures meta-reasoning — knowing when and how to reason — in text-rich image understanding. Deduction and layout emerge as the hardest skills for multimodal LLMs; data contamination and abductive-only edits limit comparisons across reasoning types. Reproducibility readiness is rated medium.
MMPCBench extends the proactive-critique program to flawed multimodal inputs
The third 2026 benchmark paper evaluates multimodal large language models on proactively critiquing flawed inputs, rounding out a trilogy spanning recommendation, OCR reasoning, and multimodal critique. It sits in the Evolve catalog without a review, keywords, or a reproducibility rating yet.

Timeline

2018First recorded output in the OpenAlex snapshot: 1 indexed work.
2019One further indexed work; the record then goes quiet until 2025.
2025Output resumes with 3 indexed works, marking renewed activity.
2026Peak year: 16 works, including three last-author arXiv benchmark papers — RPCBench, OCR-MetaReasoning, and MMPCBench.

Papers in the Evolve catalog

Frequent co-authors

Yuan Wu×3 Gengxu Li×2 Zhongru Chen×1 Jinzhe Li×1 Jinnan Li×1

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 an OpenAlex snapshot only; no ORCID was available, and 'Yi Chang' is a relatively common name, so this profile is name-resolved from public bibliographic data with moderate identity confidence.
The citation footprint is minimal in the snapshot (h-index 1, i10-index 0), so the output, impact, focus, and reproducibility dimensions are left unscored rather than estimated.
The gap in the countsByYear record between 2019 and 2025 may reflect either a pause in indexed output or identity fragmentation in OpenAlex; it is not interpreted as a career fact.
Only two of the three catalog papers carry Evolve review verdicts; MMPCBench is listed without a review, keywords, or a reproducibility rating.