University of Pittsburgh ML researcher in efficient multimodal AI systems and quantization-robust LLM unlearning.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
1
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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—
Not yet scored: the snapshot provides no per-year publication series or venue-quality measures, so research output cannot be computed. This axis would measure annual paper volume, venue tier distribution, and the first/corresponding-author share of that output.
Impact—
Not yet scored: available citation indicators are too thin (h-index 1, i10-index 0) to measure influence. This axis would track citation counts, referencing works, and the downstream uptake of the unlearning and multimodal-efficiency preprints.
Focus0/100
Scores 0 on focus, reflecting a deliberately broad early-stage spread: the field distribution splits exactly 50/50 between multimodal AI systems and LLM unlearning/model compression, and the five topic clusters are flat, with 'Advanced Neural Network Applications' and 'Domain Adaptation and Few-Shot Learning' tied at six works each, ahead of multimodal (3), privacy (2) and GAN synthesis (2). No single dominant niche has yet emerged.
Collaboration64/100
Collaboration scores 64: the profile links to 24 distinct coauthors across just 12 recent works, indicating many small teams rather than one stable group. The frequent-coauthor list includes Huanrui Yang (listed twice), Zhijian Liu, Yingyan Celine Lin, Beidi Chen, Mohit Bansal and Xiaoming Liu — spanning compression, multimodal and NLP-affiliated researchers — plus Pan Wang, Siwei Song, Hui Ji, Siqi Cao and Heng Yu.
Activity75/100
Activity scores 75 on a sharply accelerating curve: countsByYear records 1 work in 2025 versus 11 in 2026, and the snapshot totals 12 works over the last three years — meaning essentially all visible output landed in 2026, largely as arXiv preprints. Both Evolve-reviewed papers carry 2026 stamps, consistent with a burst of survey-writing and experimental submissions rather than an established multi-year cadence.
Reproducibility55/100
Reproducibility scores 55, computed as a mean of 55 across 2 papers with reproducibility assessments. Both cataloged papers are rated 'medium' readiness; Evolve's verdict on the unlearning paper notes the absence of significance tests, and the efficiency survey is flagged for offering no quantitative synthesis — so evidence standards are plausible but not yet airtight on either preprint.
Research summary
Jingtong Hu is a machine-learning researcher at the University of Pittsburgh whose visible record, resolved from OpenAlex bibliographic data, splits evenly between two areas: multimodal AI systems, and LLM unlearning with model compression. The snapshot is early-stage: h-index 1 and i10-index 0, with 12 works in the last three years, of which countsByYear assigns 1 to 2025 and 11 to 2026 — a sharp recent ramp concentrated in arXiv preprints. Two preprints carry Evolve reviews. The 2026 survey 'From Models to Systems' proposes a Model-Algorithm-System taxonomy for efficient multimodal learning, covering model compression, KV cache management and mixture-of-experts; Evolve calls it coherent and well-sourced but lacking quantitative synthesis. The 2026 unlearning study attributes unlearning reversal under 4-bit quantization to sharp forget-basins in the retain-forget loss landscape; Evolve deems the diagnosis plausible but notes missing significance tests and general-LLM evidence. Topic clustering places 'Advanced Neural Network Applications' and 'Domain Adaptation and Few-Shot Learning' tied at six works each, ahead of multimodal (3), privacy (2) and GAN synthesis (2). Collaboration is broad for this stage: 24 distinct coauthors, including Yingyan Celine Lin, Beidi Chen, Zhijian Liu, Mohit Bansal and Xiaoming Liu. Radar scores stand at activity 75, collaboration 64 and reproducibility 55, while output and impact remain unscored.
Collaboration pattern
Collaboration runs through compact, shifting teams: 24 distinct coauthors across only 12 recent works. The recurring list mixes compression and multimodal-systems researchers — Huanrui Yang (appearing twice), Zhijian Liu, Yingyan Celine Lin, Beidi Chen — with NLP-affiliated Mohit Bansal and Xiaoming Liu, plus Pan Wang, Siwei Song, Hui Ji, Siqi Cao and Heng Yu. Neither cataloged paper lists Hu in the first or corresponding slot, so the visible role is contributor rather than lead.
Research keywords
Highlights
First Model-Algorithm-System taxonomy for efficient multimodal learning
Sharp forget-basins explain unlearning reversal under 4-bit quantization
Timeline
- 2025First visible activity in the snapshot: countsByYear records 1 work in 2025, the earliest year in this OpenAlex profile's recent window.
- 2026Output jumps to 11 works in 2026 per countsByYear, bringing the last-three-year total to 12 and driving the activity score of 75.
- 2026Evolve reviews the arXiv survey 'From Models to Systems: A Comprehensive Survey of Efficient Multimodal Learning' (last-author position), rating its reproducibility readiness medium.
- 2026Evolve reviews the arXiv study 'Quantization-Robust Unlearning through the Lens of Retain-Forget Loss Landscapes Interaction' (middle-author position), rating its reproducibility readiness medium.
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.
- Bibliometric indicators are computed from OpenAlex records only (identity: openalex); citation-based counts such as h-index 1 may undercount if same-name works are split or merged.
- The profile is name-resolved from public bibliographic data; the identity has not been manually verified against a full ORCID or institutional record.
- Evolve verdicts cover only the 2 arXiv preprints in the catalog; the remaining works in the snapshot are unreviewed.
- Output and impact axes are unscored (dimMissing): no per-year publication series and only thin citation indicators (h-index 1, i10-index 0) are available.