Yang Li

Iowa State University · United States

Unclaimed · AI-structured profileOpenAlex

Emerging AI researcher at Iowa State: long-context LLM compression, LLM safety analysis, and agentic AI for science

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 cannot be scored: the record lists h-index and i10-index of 0 with no publication-volume metric, so this axis has insufficient data. It would measure total works, venue tier, and first- or corresponding-author share.

Impact—

Impact lacks sufficient data: h-index and i10-index sit at 0 and no citation counts are present. This axis would measure citations, citing works, and uptake of the reviewed preprints.

Focus—

No focus score is computed: the five assigned topics, from computer graphics and GAN synthesis to topic modeling and DC-DC converters, are too scattered for a verdict. This axis would measure coherence and depth of the research agenda.

Collaboration41/100

Collaboration scores 41 on the radar, derived from real bibliometric data. The indexed record lists 7 distinct coauthors across 7 works in the trailing three years, and Evolve's catalog surfaces 7 frequent collaborators, including Md Mokarram Chowdhury, Ernie Chang, Runyu Guan, Dehao Wu, Qiqi Xie, Haohan Wang, and Daniel Agyei Asante. On the reviewed papers Yang Li appears as last author twice and as a middle author once, consistent with a small, senior-role collaboration footprint.

Activity61/100

Activity scores 61, the stronger of the two computed axes. The OpenAlex record shows 7 works in the trailing three years, with counts by year of 1 in 2025 rising to 6 in 2026; all three Evolve-reviewed papers are 2026 preprints. This steep single-year jump indicates an intensive, current publishing phase rather than a long cumulative record, and the score should be read against that short history.

Reproducibility—

A reproducibility score is not available; the only signal is Evolve's low reproducibility-readiness rating for TopoCompress, citing missing code and hyperparameters. This axis would measure code release, data availability, and reporting completeness across works.

Research summary

Yang Li is a researcher at Iowa State University whose profile is name-resolved from public OpenAlex bibliographic data, since no ORCID is attached and the name is common. The indexed record is early-stage but accelerating: 7 works in the trailing three years, 1 appearing in 2025 and 6 in 2026, with h-index and i10-index both at 0. Assigned topics cluster around computer graphics and visualization techniques, 3D shape modeling and analysis, and generative adversarial networks and image synthesis, with single counts in topic modeling and advanced DC-DC converters — a spread that partly reflects thin topic assignment rather than a settled agenda. Evolve's catalog concentrates on large language model inference and safety: as last author of TopoCompress, a training-free graph-based long-context compression method whose reported gains lack significance tests, code, and hyperparameter values (reproducibility readiness: low); as last author of a component-resolved causal analysis of roleplay jailbreaks and refusal; and as a middle author on an agentic AI study of conserved cross-tissue protein co-abundance programs. Computed radar scores cover activity (61) and collaboration (41); output, impact, focus, and reproducibility lack sufficient data.

Collaboration pattern

Collaboration is small and stable: 7 distinct coauthors over 7 recent works (radar collaboration score 41). Seven frequent partners recur — Md Mokarram Chowdhury, Ernie Chang, Runyu Guan, Dehao Wu, Qiqi Xie, Haohan Wang, and Daniel Agyei Asante. On reviewed papers Yang Li sits last-author twice, suggesting a senior or supervisory role, and middle-author once, indicating selective contribution to cross-disciplinary AI-for-science work.

Research keywords

prompt compressionlong-context LLM inferencerelevance propagationgraph-based rankingtraining-free compressionLongBench evaluationroleplay jailbreaksLLM safety and refusalcausal analysisagentic AIprotein co-abundance

Highlights

TopoCompress: training-free graph-based long-context compression
As last author, Yang Li's group proposes compressing long contexts via graph-wired semantic trajectories with relevance propagation for long-context LLM inference. Evolve notes the reported gains over perplexity baselines lack significance tests, code, and hyperparameter values, and rates reproducibility readiness low.
Component-resolved causal analysis of roleplay jailbreak refusal
Yang Li is last author on this 2026 arXiv preprint, which examines the safety relay in roleplay jailbreaks through a component-resolved causal analysis of harm recognition and refusal. It sits in the Evolve catalog but has no review verdict, keywords, or reproducibility rating yet.
Agentic AI for cross-tissue protein co-abundance discovery
As a middle author, Yang Li contributes to this 2026 arXiv preprint reporting that agentic AI uncovers conserved cross-tissue protein co-abundance programs inaccessible to single-dataset analysis, extending the record beyond LLM inference into AI-for-science.

Timeline

2025The OpenAlex counts-by-year record shows the first indexed work (1 work in 2025), marking the start of the visible publication trail.
2026Indexed output jumps to 6 works; all three Evolve-reviewed papers are 2026 arXiv preprints.
2026TopoCompress, a training-free graph-based long-context compression method, is posted with Yang Li as last author; Evolve rates its reproducibility readiness low.
2026A component-resolved causal analysis of roleplay jailbreak refusal and an agentic AI protein co-abundance study enter the Evolve catalog, extending the range to LLM safety and AI-for-science.

Papers in the Evolve catalog

2026-08-31
TopoCompress: Long Context Compression via Graph-Wired Semantic Trajectories

Frequent co-authors

Md Mokarram Chowdhury×1 Ernie Chang×1 Runyu Guan×1 Dehao Wu×1 Qiqi Xie×1 Haohan Wang×1 Daniel Agyei Asante×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.

Bibliometric metrics come from OpenAlex only; no ORCID is available and 'Yang Li' is a common name, so this profile is name-resolved from public bibliographic data.
The record shows h-index and i10-index of 0 with no citation counts, so impact-related conclusions are not possible.
Only 3 recent works have Evolve reviews; two catalog entries lack verdicts, keywords, and reproducibility ratings.
All reviewed works are arXiv preprints; peer-reviewed venue status is unknown.