Ryan A. Rossi

Adobe Systems (United States) · United States

Unclaimed · AI-structured profileORCID OpenAlex

Adobe researcher bridging graph ML and LLM systems: bandit routing, skill retrieval, sparse long-context inference

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—

Not scored: this snapshot provides no publication-volume or venue-normalized output metric for the missing output axis. It would measure total scholarly production and its spread across venues and years; only raw yearly indexed counts exist here.

Impact—

Not scored: citation counts, citation-quality, or field-normalized impact data are absent from this snapshot, so impact cannot be estimated. This axis would measure how widely the work is cited and adopted relative to field baselines.

Focus0/100

The focus score is 0, the lowest possible, reflecting exceptionally broad topical dispersion: five substantial clusters — Topic Modeling (131), Complex Network Analysis Techniques (101), Advanced Graph Neural Networks (90), Graph Theory and Algorithms (89), and Multimodal ML (74) — plus an even 0.5/0.5 field split between bandits-for-LLM-systems and LLM-agents/IR work. Breadth, not concentration, characterizes this record.

Collaboration58/100

Scoring 58, collaboration sits mid-range: 17 distinct coauthors appear in the recent three-year window, and a recurring core — Franck Dernoncourt, Branislav Kveton, Hoda Eldardiry, Soumyabrata Pal, Koyel Mukherjee and seven others — repeats across the record. All three reviewed 2026 papers list Rossi in a middle author position, none as corresponding author, consistent with contributing within larger teams rather than leading small ones.

Activity100/100

Activity earns the maximum score of 100: recent three-year output totals 211 works, and yearly indexed counts accelerate from 39 (2023) to 63 (2024) and 88 (2025), with 60 already recorded in 2026 — far above 22 in 2018 and 29 in 2021. Three 2026 arXiv papers sit in Evolve's reviewed catalog, marking a steep, sustained productivity ramp in the LLM-systems period.

Reproducibility55/100

Reproducibility scores 55 (reproMean averaged over 2 repro-scored papers). All three reviewed papers carry medium reproReadiness: the DPP skill-routing study reports ablations but small gains on only 75 queries; the bandit-routing paper's proofs have gaps with gains within noise; CRISP's evidence is single-run with no code released. Adequate for method papers, not yet independently verifiable.

Research summary

Ryan A. Rossi is a researcher at Adobe Systems (United States) whose bibliometric record, resolved from OpenAlex, spans graph-based machine learning and, more recently, LLM systems. The snapshot lists h-index 36 and i10-index 110, with 211 works in the last three years and indexed output climbing from 39 (2023) to 63 (2024) and 88 (2025). Topic mass concentrates in Topic Modeling (131), Complex Network Analysis Techniques (101), Advanced Graph Neural Networks (90), Graph Theory and Algorithms (89), and Multimodal Machine Learning Applications (74), while the reviewed recent work splits evenly (0.5/0.5) between online learning/bandits for LLM systems and LLM agents with information retrieval. Evolve's catalog includes three 2026 arXiv papers, each with Rossi in a middle author position: a determinantal-point-process diversity reranker for LLM skill routing, a contextual-bandit scheme for routing prompts to LLM experts under limited feedback, and CRISP, a sparse-prefilling method for long-context inference. Evolve's verdicts are consistent: methods are reasonable and ablation-supported, but evidence is thin — small or noise-level gains, single-run evaluations, and no released code. Collaboration scores 58 with 17 distinct recent coauthors and a recurring circle (Dernoncourt, Kveton, Eldardiry, Pal, Mukherjee). Output and impact axes lack sufficient scored data in this snapshot and are reported as missing rather than estimated.

Collaboration pattern

Rossi's collaboration footprint is mid-range (score 58): 17 distinct coauthors appear in the recent three-year window, with a stable recurring circle of 12 named collaborators — including Franck Dernoncourt, Branislav Kveton, Hoda Eldardiry, Soumyabrata Pal, Koyel Mukherjee, Wang Wei, Samyadeep Basu, Hongjie Chen, Yue Zhao, Tiankai Yang, Zhengzhong Tu, and Xiyang Hu. All three reviewed 2026 papers place him in a middle author position rather than first or corresponding, consistent with a senior-collaborator role inside larger teams.

Research keywords

LLM agentsskill routingcontextual banditsonline learningdeterminantal point processesdiversity-aware retrievalsparse attentionlong-context LLM inferencegraph neural networkscomplex network analysistopic modelingregret bounds

Highlights

Diversity-aware skill routing for LLM agents via determinantal point processes
In 'Beyond Top-k Skill Retrieval' (2026), DSR adds DPP query-residual diversity reranking to LLM skill routing; ablations support the design, though gains are small, untested, and shown on only 75 queries.
Routing prompts to LLM experts under a limited feedback budget
'Online Learning with LLM Experts from Limited Feedback' (2026) applies contextual and linear bandits to LLM routing with regret bounds; the theory is promising, but proofs have gaps and Nectar Table 1 gains sit within noise.
CRISP: structural-mass sparse prefilling for long-context inference
CRISP (2026) replaces JSD-based routing with a structural-mass proxy and sink-aware thresholding, matching dense attention on InfiniteBench; evidence is single-run and no code is released.

Timeline

2012Indexed output reaches 10 works — the highest to that point, up from 1–3 per year in 2007–2011 — during the early graph and network-analysis period.
2018Output jumps to 22 works, double 2017's 11, as the publication record accelerates.
202129 works are indexed, extending a multi-year climb that continues through the graph-ML and multimodal period.
202339 works are indexed, the strongest year in the snapshot to that point.
202588 works are indexed — the highest annual count in the snapshot.
2026Three Evolve-reviewed arXiv papers appear — DPP-based skill routing, bandit routing of LLM experts, and CRISP sparse prefilling — all with Rossi in a middle author position.

Papers in the Evolve catalog

Frequent co-authors

Franck Dernoncourt×3 Wang Wei×2 Hoda Eldardiry×2 Soumyabrata Pal×1 Koyel Mukherjee×1 Branislav Kveton×1 Tiankai Yang×1 Samyadeep Basu×1 Hongjie Chen×1 Yue Zhao×1 Zhengzhong Tu×1 Xiyang Hu×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 counts, h-index, topics, and yearly output come from a single OpenAlex snapshot and may undercount or misresolve works.
The output and impact radar axes are unscored in this snapshot; no citation counts or venue-normalized metrics were provided, so neither is estimated here.
Only the three 2026 arXiv papers reviewed by Evolve were available; no venue, citation, or funding details accompany them.
Identity is name-resolved from public bibliographic data (OpenAlex); affiliation is as recorded in the snapshot and has not been independently verified.