Pietro Liò

University of Cambridge · United Kingdom

Unclaimed · AI-structured profileORCID OpenAlex

University of Cambridge machine-learning researcher: graph neural networks, topological analysis, explainable AI

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—

Output data is insufficient: the OpenAlex snapshot lacks a reliable full publication record for this identity, so no output score is computed. This axis would measure total publications, venue quality, and yearly production volume.

Impact—

Impact data is insufficient: no dependable citation totals are available for the resolved profile, so no impact score is given. This axis would measure citations received, citation trajectory over time, and uptake by other researchers.

Focus0/100

Focus scores 0, the lowest possible, because the field distribution is split exactly 50/50 between machine learning / NLP (LLM reasoning analysis) and machine learning — explainable AI for clinical prediction. Topic counts reinforce the breadth: Advanced Graph Neural Networks (16), Topological and Geometric Data Analysis (13), Graph Theory and Algorithms (12), Explainable AI (8), and Computational Drug Discovery Methods (6) — a deliberately exploratory, multi-line agenda rather than one narrow specialty.

Collaboration48/100

Collaboration scores 48, mid-range, built on 10 distinct coauthors in the recent window. The frequent-partner list includes Haitz Sáez de Ocáriz Borde, Davide Murari, Carola-Bibiane Schönlieb, Fabrizio Silvestri, George Montañez, Mario Severino, and others, suggesting a Cambridge-anchored core with Italian and international branches. None of the three catalog papers lists him as corresponding author; he appears last on DynSHAP and the sheaf-network paper and middle on the LLM-reasoning study.

Activity100/100

Activity scores 100, the maximum. The snapshot records 43 works in the recent three-year window, and the counts-by-year series shows 43 in 2026 against just 7 in 2022 and 4 in 2023 — an extreme concentration of production in the current cycle. All three Evolve-catalog papers are 2026 arXiv preprints, corroborating an intense, preprint-first publishing phase spanning explainable survival analysis, LLM reasoning, and sheaf-based graph learning.

Reproducibility55/100

Reproducibility scores 55, the mean across 2 assessed papers in the Evolve catalog. Both reviewed items carry a medium repro-readiness rating: DynSHAP is described as promising and honest with model-specific gains and mixed faithfulness, while the answer-distribution-trajectories paper has a sound framework but noisy estimates from only 16 rollouts and no demonstrated utility. Reproducibility evidence is therefore real but limited to a small, recent sample.

Research summary

Pietro Liò is a University of Cambridge researcher whose OpenAlex-resolved profile centers on machine learning, with activity split evenly between LLM-reasoning analysis and explainable AI for clinical prediction. Topic counts show graph-centric foundations — Advanced Graph Neural Networks (16 papers), Topological and Geometric Data Analysis (13), and Graph Theory and Algorithms (12) — alongside Explainable AI (8) and Computational Drug Discovery Methods (6). The Evolve catalog holds three 2026 arXiv papers: DynSHAP (last author), which extends SHAP to dynamic survival analysis through time–feature players and was judged promising though model-specific; Answer-Distribution Trajectories (middle author), a stochastic-dynamics view of chain-of-thought reasoning judged sound but limited by noisy 16-rollout estimates; and Equivariant Sheaf Neural Networks on geometric transport over graphs (last author), not yet reviewed. Activity is the standout axis: 43 works in the recent three-year window, nearly all stamped 2026 in the counts-by-year series (7 in 2022, 4 in 2023). Collaboration is mid-range with 10 distinct recent coauthors, including Haitz Sáez de Ocáriz Borde, Davide Murari, Carola-Bibiane Schönlieb, and Fabrizio Silvestri. Reproducibility averages 55 across two assessed papers, both rated medium readiness. Output and impact axes remain unscored pending fuller citation data.

Collaboration pattern

Collaboration is mid-range (radar 48) with 10 distinct recent coauthors. The recurring list spans Haitz Sáez de Ocáriz Borde, Davide Murari, Carola-Bibiane Schönlieb, Fabrizio Silvestri, George Montañez, Mario Severino, Alessio Borgi, Mar Gonzàlez I Català, Nastasya Anokhina, and Jonas Jürß — a Cambridge-anchored network with Italian and European ties. Authorship patterns point to a senior-collaborator role: last position on two catalog papers and middle on the third, none as corresponding author.

Research keywords

Graph neural networksGeometric deep learningSheaf neural networksTopological data analysisExplainable AI (XAI)SHAP and Shapley valuesDynamic survival analysisChain-of-thought reasoningUncertainty quantificationLLM evaluationComputational drug discoveryGraph theory and algorithms

Highlights

Extending SHAP to dynamic survival analysis
DynSHAP introduces time–feature players to bring Shapley-value explanations into dynamic survival analysis on longitudinal data. Evolve's verdict: promising and honest, but gains are model-specific and faithfulness is mixed; repro-readiness rated medium.
A stochastic-dynamics lens on LLM reasoning
This paper tracks evolving answer distributions across chain-of-thought rollouts, offering a finer signal than entropy or accuracy alone. Evolve finds the framework sound but notes that 16-rollout estimates are noisy and no downstream utility is shown; repro-readiness medium.
Equivariant sheaf networks for geometric transport on graphs
The newest catalog entry, with Liò in the last author position, applies sheaf-theoretic constructions to learning geometric transport on graphs. It extends the graph-and-geometry line visible in the topic profile but has not yet received an Evolve review or reproducibility assessment.

Timeline

20227 publications recorded for the OpenAlex-resolved profile in the counts-by-year series.
2023Publication count dips to 4, the lowest point in the provided series.
2026Output surges to 43 recorded works — essentially the entire recent three-year total lands in this single year.
2026Three papers enter the Evolve catalog: DynSHAP (last author), Answer-Distribution Trajectories (middle author), and Equivariant Sheaf Neural Networks (last author), all arXiv preprints.

Papers in the Evolve catalog

Frequent co-authors

Alessio Borgi×1 Mario Severino×1 Fabrizio Silvestri×1 Mar Gonzàlez I Català×1 Haitz Sáez de Ocáriz Borde×1 Davide Murari×1 Carola-Bibiane Schönlieb×1 George Montañez×1 Nastasya Anokhina×1 Jonas Jürß×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.

All bibliometrics (h-index 5, i10 index 4, counts by year, topics, coauthors) come from an OpenAlex-resolved identity; the low reported indices suggest the record may be incomplete or partially name-resolved.
Only 3 papers, all from 2026, are in the reviewed Evolve catalog; 2 carry reproducibility assessments, and the sheaf-network paper has no verdict yet.
No output or impact scores are computed — citation totals were insufficient in this snapshot.
All catalog papers are arXiv preprints; peer-reviewed venue status and funding/award information fall outside the provided data.