Virginia Tech machine learning researcher advancing LLM model merging, agent routing, and multimodal extraction
4
papers here
4
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
16
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 lacks the verified per-paper output aggregates needed for this axis, which would measure total publication volume and its growth. Per-year work counts exist (for example 32 recorded works in 2025), but no output score is computed here.
Impact—
Not yet scored: citation data is insufficient beyond the snapshot's h-index of 16 and i10-index of 28. This axis would measure citation influence and scholarly uptake; no per-paper citation totals are available, so no estimate is made.
Focus0/100
Scores 0, indicating no concentrated specialization: the four recorded fields each hold 0.25 of recent output, and topic counts are spread across topic modeling (29), advanced graph neural networks (19), multimodal machine learning applications (19), NLP techniques (15) and traffic prediction (14). This breadth matches a portfolio spanning model merging, LLM agents, bandits and multimodal extraction rather than one narrow niche.
Collaboration56/100
Scores 56, a mid-range collaborative pattern: 16 distinct coauthors across 67 works in the last three years, anchored by a recurring core including Ming Cheng, Jiaying Gong, Franck Dernoncourt, Ryan A. Rossi, Branislav Kveton and Samyadeep Basu. All four catalog papers place Eldardiry in the last author position, consistent with a senior supervisory role rather than a very large network hub.
Activity100/100
Scores 100, the maximum in this snapshot: annual recorded works rise from about 4 per year in 2016–2019 to 10 in 2020 and 11 in 2021, then 15 in 2024 and 32 in 2025, with 20 already recorded in 2026 and 67 works in the recent three-year window. This steep late-stage ramp is the strongest quantitative signal in the profile.
Reproducibility55/100
Scores 55, derived from Evolve's reproducibility assessment of 4 papers (reproMean 55). All four 2026 catalog papers carry medium reproducibility readiness, and the review verdicts flag concrete gaps: no significance tests and no optimal-transport baselines in the merging work, only 75 evaluation queries in the routing work, and proof gaps in the bandit paper.
Research summary
Hoda Eldardiry is a researcher at Virginia Tech whose recent program centers on making large language model systems more efficient and more capable — through model merging, agent skill routing, bandit-based routing under limited feedback, and multimodal information extraction. The OpenAlex snapshot records an h-index of 16 and an i10-index of 28, with clear late-stage acceleration: annual recorded works climb from about four per year in 2016–2019 to 10–11 in 2020–2021, 15 in 2024, and 32 in 2025, with 67 works in the most recent three-year window. Topic modeling (29), advanced graph neural networks (19) and multimodal machine learning applications (19) lead a broad topic profile that also spans NLP techniques (15) and traffic prediction (14). Four 2026 papers reviewed in the Evolve catalog — all with Eldardiry as last author — cover visual-search-augmented chain-of-thought for product video attribute extraction, hierarchical Wasserstein merging of LLM specialists, diversity-aware skill routing for LLM agents, and online learning with LLM experts under limited feedback; each carries medium reproducibility readiness, with a reproducibility mean of 55 across 4 assessed papers. Collaboration is mid-range (score 56), with 16 distinct recent coauthors including Ming Cheng, Jiaying Gong, Franck Dernoncourt, Ryan A. Rossi and Branislav Kveton. Output and impact dimensions remain unscored pending fuller citation data.
Collaboration pattern
Eldardiry's recent collaboration pattern is a moderate-size senior network: 16 distinct coauthors across 67 works in the last three years, with a recurring core including Ming Cheng, Jiaying Gong, Wang Wei, Franck Dernoncourt, Ryan A. Rossi, Soumyabrata Pal, Branislav Kveton and Samyadeep Basu. In all four Evolve-catalog papers she appears as last author (none flagged corresponding), consistent with supervising junior first authors drawn from this pool.
Research keywords
Highlights
Training-free visual search plus chain-of-thought for product video attribute extraction
Hierarchical Wasserstein merging turns LLM specialists into a generalist
Diversity-aware skill routing for LLM agents via determinantal point processes
Online learning routes prompts to LLM experts under limited feedback
Timeline
- 2008Earliest work in the OpenAlex snapshot, with 1 recorded publication this year.
- 2016Entering a steady phase of about 4 recorded works per year, sustained through 2019.
- 2020Output rises to 10 recorded works, followed by 11 in 2021 — roughly double the prior steady level.
- 202415 recorded works, continuing the upward climb.
- 2025Peak year in the snapshot with 32 recorded works, the highest annual count.
- 2026Four last-authored papers reviewed in the Evolve catalog so far, with 20 works recorded year-to-date.
Papers in the Evolve catalog
Visual Search Augmented Chain-of-Thought Reasoning for Attribute Value Extraction from Product Videos
Hierarchical Wasserstein Merging for Multi-Domain Multi-Task Learning: From Specialists to a Generalist
Beyond Top-$k$ Skill Retrieval: Diversity-Aware Skill Routing for LLM Agents
Online Learning with LLM Experts from Limited Feedback
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 figures come from an OpenAlex snapshot; output and impact dimensions are unscored because the underlying citation-volume data is insufficient.
- The four catalog papers are arXiv preprints whose Evolve verdicts flag partial leakage audits, missing significance tests, small evaluation sets and proof gaps.
- 2026 publication counts cover only part of the year and may change.
- The profile is name-resolved from public bibliographic data (ORCID 0000-0002-9712-6667); no biographical details beyond the snapshot were used.