Sapienza University of Rome researcher in machine learning: vision-language models, multimodal learning and NLP
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
1
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 measures publication volume over time, but the available data is insufficient to score it: the OpenAlex snapshot indexes only 17 works in total (2021: 1, 2025: 2, 2026: 14), a record too sparse to characterize sustained output.
Impact—
This axis would measure citation-based influence, but the signals are too thin to score: the record lists h-index 1 and i10-index 0 with no aggregate citation counts, so no meaningful impact estimate is possible yet.
Focus100/100
Evolve's focus score is 100, the maximum. The field distribution in the record contains a single field — machine learning (vision-language models) — and the topic profile stays within one broad area: topic modeling (8), multimodal machine learning applications (7), natural language processing techniques (5), explainable AI (4) and advanced neural network applications (3).
Collaboration48/100
The collaboration score is 48, mid-range. The record shows 10 distinct coauthors, including recurring collaborators such as Alessio Borgi, Mario Severino, Pietro Liò and Pasquale Minervini. Both 2026 catalog papers are multi-authored, with Silvestri in a middle position on one and last position on the other; no corresponding-author role is recorded.
Activity83/100
The activity score is 83, reflecting a clear recent ramp-up: 16 works in the last three years, with the indexed yearly counts rising from 1 (2021) and 2 (2025) to 14 in 2026 — the year that also produced both Evolve-catalog arXiv papers.
Reproducibility100/100
The reproducibility score is 100 (reproMean 100), but it rests on a small base: only 1 paper has been assessed. That paper, the cross-modal attention frequency-filter study, carries Evolve's highest reproducibility readiness rating ('high'), consistent with its strong experimental design. The score's wider reliability awaits further assessed papers.
Research summary
Fabrizio Silvestri is a researcher at Sapienza University of Rome whose Evolve profile is name-resolved from OpenAlex bibliographic data. The snapshot is recent-heavy and fast-moving: 16 works in the last three years, with indexed yearly counts concentrated in 2026 (14 works) after 2 in 2025 and 1 in 2021. His topical profile sits squarely inside machine learning and NLP — topic modeling (8), multimodal machine learning applications (7), natural language processing techniques (5), explainable AI (4) and advanced neural network applications (3) — and the record's field distribution names a single field, machine learning (vision-language models), producing Evolve's maximum focus score of 100. Two 2026 arXiv papers anchor the Evolve catalog: 'Cross-Modal Attention Acts as a Frequency Filter: Why Verbose Prompts Improve Robustness in Vision-Language Models', where he is last author, and 'Equivariant Sheaf Neural Networks: Learning Geometric Transport on Graphs', where he is a middle author. Evolve's review of the first judged the design strong while noting mixed mechanism evidence across runs, and rated its reproducibility readiness high — the basis of his 100 reproducibility score, although that rests on a single assessed paper. He collaborates with 10 distinct coauthors, including Pietro Liò and Pasquale Minervini. Citation-based indicators are too thin to score (h-index 1, i10-index 0), so output and impact remain unscored in this profile.
Collaboration pattern
Silvestri's collaboration footprint in this snapshot is moderate: 10 distinct coauthors, with recurring partners including Alessio Borgi, Mario Severino, Pietro Liò, Farooq Ahmad Wani, Maria Sofia Bucarelli, Mujtaba Hussain Mirza, Oleksandr Pryymak, Aryo Pradipta Gema, Iacopo Masi and Pasquale Minervini. Both 2026 catalog papers are collaborative, placing him in middle and last authorship positions; no corresponding-author role appears in the record.
Research keywords
Highlights
Verbose prompts as a frequency filter for VLM robustness
Equivariant sheaf neural networks for geometric transport on graphs
A second 2026 arXiv preprint, with Silvestri as a middle author, explores equivariant sheaf neural networks that learn geometric transport on graphs — a geometric deep learning direction adjacent to his neural-network and NLP topic profile. Evolve has not yet published a verdict or reproducibility assessment for it.
Concentrated machine learning focus with a 2026 output surge
The OpenAlex record shows a single-field distribution (machine learning, vision-language models) and topics clustered in topic modeling (8), multimodal ML applications (7), NLP techniques (5), XAI (4) and neural network applications (3), with 14 of 17 indexed works appearing in 2026 — yielding a maximum focus score of 100 and an activity score of 83.
Timeline
- 2021Earliest indexed work in this OpenAlex snapshot: 1 publication recorded for the year.
- 20252 indexed works, the first step of the recent ramp-up visible in this sparse record.
- 2026Most productive indexed year with 14 works, including both Evolve-catalog arXiv papers: the verbose-prompt robustness study (last author) and the equivariant sheaf neural networks paper (middle author).
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 only; an h-index of 1 and i10-index of 0 alongside 16 recent three-year works suggest a sparse or incomplete author record rather than a full career profile.
- Only one catalog paper carries an Evolve verdict and reproducibility assessment, so the reproducibility score of 100 rests on a single data point.
- The output and impact dimensions are unscored because publication-volume and citation data in this snapshot are insufficient.
- The profile is name-resolved from public bibliographic data (OpenAlex identity); biography, awards and career history outside the indexed works are not covered.