NLP researcher at the University of Edinburgh working on vision-language robustness and reasoning-model faithfulness.
2
papers here
1
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
10
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 yet scored. This axis measures total publication volume and productivity trends; the current OpenAlex snapshot lacks the complete, validated output aggregates Evolve needs, so no score is given, even though 46 works appear in 2024–2026.
Impact—
Not yet scored. This axis measures citation-based influence — citations, citation velocity and citing works. Although the snapshot records an h-index of 10 and an i10-index of 10, the full citation aggregates needed to score impact are unavailable.
Focus100/100
Score 100. The catalog field distribution is concentrated — machine learning (vision-language models) is the sole assigned field — and the two reviewed 2026 papers target vision-language prompt robustness and chain-of-thought faithfulness. The wider topic profile stays NLP-centric: Topic Modeling (44), NLP Techniques (20), Multimodal ML Applications (12), with applied strands in Biomedical Text Mining (11) and Pharmacovigilance (9).
Collaboration50/100
Score 50. The record lists 11 distinct coauthors, a moderately sized network for this output volume. Recurring partners include Pasquale Minervini, who appears twice in the frequent-coauthor list and thus marks the most durable tie, alongside Neel Rajani, Rohit Saxena, Wai-Chung Kwan, Farooq Ahmad Wani, Maria Sofia Bucarelli, Mujtaba Hussain Mirza, Oleksandr Pryymak, Iacopo Masi and Fabrizio Silvestri.
Activity100/100
Score 100. The counts-by-year record shows productivity accelerating sharply: 1–3 works annually from 2017 through 2023, then 20 works in 2024, 17 in 2025 and 9 already in 2026, giving 46 works in the last three years (recent3yWorks). This sustained surge places the profile in the top activity band.
Reproducibility100/100
Score 100, from a single assessed paper (reproPapers: 1, reproMean: 100). The reviewed 2026 vision-language paper carries a high reproducibility-readiness rating; Evolve's verdict credits a strong experimental design while noting that its mechanism evidence is mixed across runs. With only one paper assessed, the perfect score should be read as promising rather than established.
Research summary
Aryo Pradipta Gema is a natural language processing researcher at the University of Edinburgh whose OpenAlex-resolved record spans 2017–2026 and shows a sharp acceleration in output: after 1–3 works per year from 2017 through 2023, the counts-by-year series jumps to 20 works in 2024, 17 in 2025 and 9 in 2026 to date — 46 works across the last three years. The topic profile is NLP-centered, led by Topic Modeling (44), Natural Language Processing Techniques (20) and Multimodal Machine Learning Applications (12), with applied strands in Biomedical Text Mining and Ontologies (11) and Pharmacovigilance and Adverse Drug Reactions (9). Current work centers on machine learning for vision-language models: Evolve's catalog holds two 2026 arXiv papers — a middle-authored study arguing that verbose prompts improve robustness under image corruption via a cross-modal attention frequency filter, rated high for reproducibility readiness despite mixed mechanism evidence across runs, and a first-authored study showing that chain-of-thought faithfulness in reasoning models varies with where and how preference cues are delivered. The record lists 11 distinct coauthors, with Pasquale Minervini, Neel Rajani, Rohit Saxena and Wai-Chung Kwan among recurring collaborators. Collaboration scores mid-scale at 50, while activity, focus and reproducibility each score 100. Citation-based impact remains unscored pending full aggregates, though the snapshot records an h-index of 10 and an i10-index of 10.
Collaboration pattern
Gema's recorded network comprises 11 distinct coauthors. Pasquale Minervini appears twice in the frequent-coauthor list, marking the most repeated tie; other recurring names include Neel Rajani, Rohit Saxena, Wai-Chung Kwan, Fabrizio Silvestri, Maria Sofia Bucarelli, Iacopo Masi, Oleksandr Pryymak, Farooq Ahmad Wani and Mujtaba Hussain Mirza. The small-team authorship positions in the catalog papers fit a mid-scale collaboration score of 50 rather than large-consortium work.
Research keywords
Highlights
Verbose prompts as a frequency filter for VLM robustness
First-authored study: chain-of-thought faithfulness shifts with preference-cue placement
2026 arXiv preprint with Gema as first author (non-corresponding), showing that the chain-of-thought faithfulness of reasoning models varies with where and how preference cues are delivered. Evolve has not yet published a review verdict or reproducibility rating for this paper.
Productivity surge: 46 works in the last three years
Counts by year jump from 1–3 works annually across 2017–2023 to 20 in 2024, 17 in 2025 and 9 in 2026 to date. Topic counts over the same record are led by Topic Modeling (44) and Natural Language Processing Techniques (20), with Multimodal ML Applications (12) next.
Timeline
- 2017First works appear in the bibliometric record, with 2 publications in the year.
- 20233 works recorded — the highest annual count of the 2017–2023 period before the surge.
- 2024Output surges to 20 works, the largest annual total in the record.
- 202517 further works, sustaining the elevated publication pace for a second year.
- 20269 works to date, including both catalog papers: the first-authored chain-of-thought faithfulness study and the vision-language frequency-filter paper.
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 figures are computed from an OpenAlex snapshot; the profile is name-resolved from public bibliographic data tied to ORCID 0009-0007-1163-3531.
- Output and impact axes are unscored because full citation aggregates are unavailable; the h-index (10) and i10-index (10) quoted here come from the snapshot as recorded.
- The reproducibility score rests on a single reviewed paper; only one of the two catalog papers carries a full Evolve verdict, and the other has no slug or rating yet.
- Counts-by-year figures come from the snapshot, and the 2026 total covers only part of the year.