Hanghang Tong

University of Illinois Urbana-Champaign · United States

Unclaimed · AI-structured profileOpenAlex

UIUC researcher working on LLM governance, vision-language test-time RL, and efficient diffusion language 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—

This axis would measure total publication volume and per-year output, but the underlying counts are unavailable: the snapshot returns zeroed h-index and i10-index values and no reliable cumulative work totals, so output cannot be scored yet.

Impact—

This axis would measure citation-based influence such as citation counts and index trajectories, but the snapshot contains no citation data and zeroed index values, so impact is not scored here and should not be inferred from other axes.

Focus0/100

The focus score of 0 reflects topical dispersion rather than absence of work: the field distribution splits almost exactly into thirds — NLP (0.333), machine learning with vision-language models (0.333), and NLP/LLM safety and governance (0.333) — while OpenAlex topic counts span five areas (Topic Modeling 19, NLP Techniques 10, Advanced GNNs 8, Multimodal ML 7, RL in Robotics 6). The reviewed 2026 papers likewise cross governance, VLMs, and diffusion LMs.

Collaboration56/100

Collaboration scores 56: the 28 works from the last three years involve 16 distinct co-authors, and twelve recurring collaborators appear in the snapshot, including Yuanchen Bei, Jingrui He, Haobo Xu, and Yuchen Yan. The three Evolve-reviewed papers place this researcher in middle or last author positions with no corresponding-author role, consistent with a contributor pattern inside larger teams rather than a small fixed duo.

Activity98/100

Activity scores 98, the strongest axis in this profile. The snapshot records 28 works in the last three years, with countsByYear showing 27 works dated 2026 — a concentrated burst compared with 3 works in 2021 and 1 in 2024. All three Evolve-reviewed papers (PolicyMem, TTIQ, and PILL) appeared on arXiv in 2026, reinforcing that the current publication tempo is unusually high.

Reproducibility70/100

Reproducibility scores 70, averaged over the 3 papers Evolve has reviewed (reproMean 70, reproPapers 3). Readiness varies by paper: PILL (adaptive-length infilling for diffusion LMs) is rated high, while PolicyMem and TTIQ are rated medium — the verdicts cite self-verification without significance tests for PolicyMem, and for TTIQ and PILL single-run results without variance statistics, with λ tuned on the test set in TTIQ.

Research summary

Hanghang Tong is a researcher at the University of Illinois Urbana-Champaign whose recent output concentrates on large language model (LLM) safety and governance, vision-language modeling, and natural language processing. The bibliometric snapshot (OpenAlex) records 28 works in the last three years, 27 of them dated 2026, produced with 16 distinct co-authors — a concentrated burst that yields an activity score of 98. Topic assignments span Topic Modeling (19), Natural Language Processing Techniques (10), Advanced Graph Neural Networks (8), Multimodal Machine Learning (7), and Reinforcement Learning in Robotics (6), while the field distribution splits almost evenly (about one-third each) across NLP, vision-language machine learning, and LLM safety and governance — a breadth consistent with the focus score of 0. Three 2026 arXiv papers have been reviewed by Evolve: PolicyMem casts LLM policies as reusable low-rank subspaces for governance, with strong detection on 5 benchmarks; TTIQ uses image-question dependence as a test-time reinforcement-learning reward for vision-language models; and PILL predicts infill length for diffusion language models, beating CAL by +4.8 code pass rate at 1.82x speed. Reproducibility averages 70 across these 3 papers. Citation-based output and impact metrics are missing from the snapshot and are left unscored.

Collaboration pattern

Work in the last three years spans 16 distinct co-authors, and the snapshot lists twelve recurring collaborators — including Yuanchen Bei, Jingrui He, Haobo Xu, Yuchen Yan, and Dongqi Fu — indicating a broad, team-based pattern. On the three Evolve-reviewed 2026 papers this researcher appears in middle or last author position and never as corresponding author, consistent with contributing specialist expertise within larger groups (collaboration score: 56).

Research keywords

LLM governancesafety guardrailstest-time reinforcement learningvision-language modelsvisual question answeringdiffusion language modelstext infillinginference efficiencytopic modelinggraph neural networksreward designlow-rank subspaces

Highlights

PolicyMem: casting LLM policies as reusable low-rank subspaces
The Evolve review notes strong unsafe-content detection across 5 benchmarks by casting LLM policies as reusable low-rank subspaces with projection energy, while cautioning that results are self-verified with no significance tests (reproReadiness: medium).
TTIQ: image-question dependence as a test-time RL reward for VLMs
Reviewed as harnessing the dependence between image and question as a reinforcement-learning reward at test time for vision-language models, built on GRPO; results are strong and well ablated, though λ is tuned on the test set and significance tests are absent.
PILL: one-shot length prediction speeds up diffusion language models
PILL predicts infill length with a probe and decodes in parallel, beating CAL by +4.8 code pass rate at 1.82x speed; Evolve rates its reproducibility high while noting the results are single-run without variance statistics.

Timeline

2021Earliest year in the bibliometric snapshot, with 3 works recorded in countsByYear.
2024Only 1 work is recorded for this year in the snapshot; coverage of intervening years appears sparse.
2026Peak output: 27 works recorded, and all three Evolve-reviewed arXiv papers — PolicyMem (LLM governance), TTIQ (VLM test-time RL), and PILL (diffusion LM infilling) — appeared this year.

Papers in the Evolve catalog

Frequent co-authors

Yuanchen Bei×2 Jingrui He×2 Haobo Xu×1 Sirui Chen×1 Lingjie Chen×1 Yuchen Yan×1 Dongqi Fu×1 Xinrui He×1 Ting-Wei Li×1 Junting Wang×1 Mengting Ai×1 Xinyu He×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.

All bibliometrics come from a single OpenAlex snapshot; no ORCID is linked, so this identity is name-resolved from public bibliographic data and may merge or split records.
h-index and i10-index appear as zero and citation counts are absent, so the output and impact dimensions are unscored rather than estimated.
countsByYear covers only 2021 (3 works), 2024 (1), and 2026 (27); intervening and earlier years are likely undercounted.
The reproducibility score reflects only the 3 Evolve-reviewed arXiv papers, all from 2026.