Tsinghua machine-learning researcher on efficient long-context LLM inference and multimodal long-video understanding
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-09) 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 metrics are not yet available in the underlying data, so this axis is unscored. It would measure publication volume, venue quality, and first/corresponding-author share; the listed h-index of 0 suggests the record is too new to aggregate.
Impact—
Impact data is insufficient for scoring; citation-based influence has not yet accrued, with h-index and i10-index both listed as 0. This axis would measure citations, uptake, and downstream use of the two 2026 arXiv papers once they accumulate citations.
Focus0/100
The focus score is 0, and the raw topics explain why: works spread across five OpenAlex topics — multimodal ML applications (7), domain adaptation and few-shot learning (6), remote-sensing image classification (5), advanced neural-network applications (4), and image/video retrieval (4) — while the field distribution splits evenly at 0.5 between long-context LLM inference and multimodal video understanding. The profile is broad rather than concentrated.
Collaboration58/100
Collaboration scores 58, built on 17 distinct co-authors in the recent window. The most frequent collaborators include Zhen Huang, Ruizhe Yao, Danyi Liu, Xinrui Chen, Shuwei Li, and Siru Zhong. On the two Evolve-reviewed papers the recorded positions are last author (CompKV) and middle author (PREM), with no corresponding-author role — a senior-collaborator pattern within project teams rather than a solo-line output.
Activity79/100
Activity is the strongest axis at 79. The OpenAlex counts-by-year series shows 1 work in 2025 rising to 13 in 2026, and recent3yWorks totals 14 — meaning nearly all recorded output lands in the last two years. This is a steep, very recent acceleration rather than a long cumulative record, consistent with the still-zero citation indices.
Reproducibility55/100
Reproducibility scores 55, computed over 2 papers, both rated medium readiness. CompKV claims first compensation-aware KV selection with best sparse accuracy on RULER and up to 6.85x self-attention speedup, but its evidence is per-layer latency only, without significance tests; PREM shows modest but well-controlled gains with minor per-benchmark losses. Method detail is solid, yet statistical rigor is uneven across the two works.
Research summary
Haohuan Fu is a machine-learning researcher at Tsinghua University whose record, name-resolved from OpenAlex, centers on two evenly weighted directions: efficient long-context LLM inference and multimodal video understanding (field distribution 0.5/0.5). The bibliometric snapshot is early-stage: h-index and i10-index are both listed as 0, with 14 works in the last three years and a steep ramp from 1 work in 2025 to 13 in 2026, yielding an activity score of 79. OpenAlex topics span multimodal machine-learning applications (7 works), domain adaptation and few-shot learning (6), remote-sensing image classification (5), advanced neural-network applications (4), and image/video retrieval (4) — a spread that underlies a focus score of 0. Collaboration is broad but thinly repeated: 17 distinct co-authors, with Zhen Huang, Ruizhe Yao, Danyi Liu, Xinrui Chen, Shuwei Li, and Siru Zhong among the most frequent; the collaboration score is 58. Evolve has reviewed two 2026 arXiv papers: CompKV, a compensation-aware KV selection method reporting best sparse accuracy on RULER and up to 6.85x self-attention speedup (last author), and PREM, a prefix-steered recurrent memory for long-video understanding (middle author). Both are rated medium reproducibility readiness, with a mean score of 55 across the 2 repro-scored papers.
Collaboration pattern
The recent network spans 17 distinct co-authors, suggesting many short-lived collaborations rather than one deep cluster. Frequently recurring names include Zhen Huang, Ruizhe Yao, Danyi Liu, Xinrui Chen, Shuwei Li, Siru Zhong, Zijian Cao, Yushan Lai, Mingming Guo, Weijie Zheng, and Qiongyan Wang. On the two reviewed papers the roles are last author (CompKV) and middle author (PREM), neither corresponding — consistent with a senior collaborator or supervisory position within project teams.
Research keywords
Highlights
CompKV: compensation-aware KV selection with up to 6.85x self-attention speedup
PREM: prefix-steered recurrent memory for long-video understanding
Output ramped from 1 work in 2025 to 13 in 2026
Counts-by-year data shows the profile moving from 1 work in 2025 to 13 in 2026 (14 works over the last three years), a sharp acceleration that drives the activity score of 79 and coincides with the two Evolve-reviewed arXiv papers.
Timeline
- 2025OpenAlex counts-by-year records the first visible work in the profile (1 work in 2025), marking the start of the trackable record.
- 2026CompKV, a compensation-aware KV selection method for long-context LLM inference, appears on arXiv with the researcher as last author.
- 2026PREM, prefix-steered recurrent memory for long-video understanding, appears on arXiv with the researcher as a middle author.
- 2026Counts-by-year output reaches 13 works in 2026 (14 over the last three years), with 17 distinct co-authors recorded in the recent window.
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.
- All bibliometrics come from OpenAlex (identity: openalex); no ORCID was provided, so name disambiguation is not independently confirmed.
- The listed h-index and i10-index are both 0, and the output and impact dimensions lack sufficient data, so career-level influence cannot be assessed from this snapshot.
- Only 2 papers carry Evolve reviews, both 2026 arXiv preprints; venue acceptance and citation uptake are unknown.
- Topic and field assignments (five topics, a 0.5/0.5 field split) are OpenAlex-generated and may overstate or misplace the topical spread behind the focus score of 0.