Giambattista Amati

Fondazione "Ugo Bordoni"

Unclaimed · AI-structured profileORCID OpenAlex

Learning-to-Optimize architect at Fondazione Ugo Bordoni: MILP-supervised GNNs for AI-native 6G and NR-V2X networks

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-15) 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—

No publication-volume metric is available yet. This axis would measure research productivity over time — yearly output counts and consistency across venues. With only two 2026 arXiv preprints in the Evolve catalog, the data is insufficient and the axis is left unscored.

Impact—

No citation or influence metrics are available for this profile yet. This axis would measure community uptake — citations, usage, and downstream follow-up built on the catalogued papers. Since such data is absent, the axis is honestly left unscored rather than estimated.

Focus0/100

The focus score is 0, reflecting an even 0.5/0.5 field split between AI-native 6G network architecture/network optimization and wireless networking/AI for communications. In practice this is two adjacent facets of one agenda rather than scatter: both 2026 first-author papers share a MILP-plus-GNN template, and the architecture paper explicitly reuses the NR-V2X case from the relay-optimisation study. The low score captures breadth across two labelled fields, not unrelated topic-hopping.

Collaboration32/100

Collaboration scores 32, consistent with a compact footprint: four distinct coauthors in the snapshot, all recurring — Federica Mangiatordi, Emiliano Pallotti, Simone Angelini, and Pierpaolo Salvo. Both catalogued 2026 papers are first-authored, suggesting a stable small team built around Fondazione Ugo Bordoni rather than a broad external network. No cross-institution or international co-authorship appears in the available data.

Activity—

Publishing cadence cannot be assessed yet. This axis would measure research activity over time — the frequency and recency of outputs across venues and years. The current snapshot covers only 2026 arXiv preprints, so there is no basis for scoring this axis.

Reproducibility38/100

Reproducibility scores 38, the mean readiness across the two assessed papers. Evolve rates the NR-V2X relay-optimisation study medium: a credible MILP-supervised GINE simulation claiming roughly 100x speed-up, but with no baselines or error bars. The Learning-to-Optimize architecture paper is rated low because its results are imported from the authors' prior papers. Releasing code, baselines, and uncertainty figures would likely lift this score.

Research summary

Giambattista Amati is a researcher at Fondazione Ugo Bordoni whose Evolve-catalogued work, all from arXiv in 2026, concentrates on making mathematical optimization a native, learnable component of future wireless networks. He is first author on both catalogued papers. One frames Learning-to-Optimize as the missing architectural layer of AI-native 6G networks, demonstrated through an NR-V2X graph-neural-network case study; the other trains a MILP-supervised GINE network for real-time relay selection in smart urban NR-V2X, claiming near-optimal decisions with roughly 100x speed-up. The methodological signature is consistent: exact optimization as teacher, learned surrogate as fast proxy. Evolve's field distribution splits evenly (0.5/0.5) between AI-native 6G network architecture/network optimization and wireless networking/AI for communications — a tight two-field agenda. The collaboration footprint is compact, with four distinct coauthors (Federica Mangiatordi, Emiliano Pallotti, Simone Angelini, Pierpaolo Salvo). Reproducibility sits at a mean of 38 across two assessed papers, with Evolve noting imported results and missing baselines or error bars. Output, impact, and activity metrics are not yet available.

Collaboration pattern

Collaboration is compact and stable: the snapshot records 4 distinct coauthors, all recurring — Federica Mangiatordi, Emiliano Pallotti, Simone Angelini, and Pierpaolo Salvo — yielding a collaboration score of 32. Both 2026 papers are first-authored at Fondazione Ugo Bordoni, suggesting a small in-house team pattern rather than distributed or international co-authorship; no cross-institution partners appear in the available data.

Research keywords

Learning-to-OptimizeAI-native networks6G architectureSurrogate modelsGraph neural networksMILP (mixed-integer linear programming)NR-V2XRelay selectionGINE graph isomorphism networkReal-time network optimization

Highlights

Learning-to-Optimize proposed as the missing architectural layer of AI-native 6G
In a 2026 first-author arXiv paper, Amati argues that Learning-to-Optimize should sit inside the network stack as a native architectural layer, demonstrated through an NR-V2X graph-neural-network case study. Evolve rates reproducibility low because results are imported from the authors' prior papers.
MILP-supervised GINE delivers near-optimal NR-V2X relay selection with ~100x speed-up claim
A 2026 first-author simulation study trains a GINE graph neural network under MILP supervision for real-time relay selection in smart urban NR-V2X networks, claiming near-optimal decisions at roughly 100x the speed of exact optimization. Evolve calls the study credible but notes the absence of baselines and error bars (reproducibility: medium).
A consistent method: exact optimization as teacher, learned surrogate as fast proxy
Across both catalogued 2026 papers, MILP serves as the supervising teacher while surrogate models and graph neural networks act as fast proxies — a template the architecture paper generalizes from the NR-V2X relay case toward AI-native 6G design.

Timeline

2026First-author arXiv paper on AI-driven real-time relay optimisation in smart urban NR-V2X networks via learning-to-optimise graph neural networks (MILP-supervised GINE; ~100x speed-up claim; reproducibility medium).
2026First-author arXiv paper framing Learning-to-Optimize as the missing architectural layer of AI-native networks, with an NR-V2X GNN case study (reproducibility low; results imported from prior papers).
2026Catalogued field distribution splits 0.5/0.5 between AI-native 6G network architecture/network optimization and wireless networking/AI for communications, with a four-person recurring coauthor circle.

Papers in the Evolve catalog

2026-09-18first author
Learning-to-Optimize as the Missing Architectural Layer of AI-Native Networks
2026-09-18first author
AI-Driven Real-Time Relay Optimisation in Smart Urban NR-V2X Networks via Learning-to-Optimise Graph Neural Networks

Frequent co-authors

Federica Mangiatordi×2 Emiliano Pallotti×2 Simone Angelini×2 Pierpaolo Salvo×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 scores are computed from a limited OpenAlex-derived snapshot; the output, impact, and activity dimensions lack sufficient data and are intentionally left unscored.
Only two 2026 arXiv preprints underlie this profile; no peer-reviewed venue record, citation history, or career facts beyond the Fondazione Ugo Bordoni affiliation are included.
Reported performance claims (e.g., the ~100x speed-up) are the authors' own results as assessed by Evolve, not independently verified.
The profile is name-resolved from public bibliographic data (ORCID 0000-0002-7067-4583); identity or award information outside this data is not covered.