Federated learning at HPC scale: first-author systems work on sharded data parallelism and C/C++ FL frameworks
2
papers here
2
reviewed by Evolve
2
first-author
2
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) 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 output score is available: the snapshot provides no publication-volume or venue-count data beyond the 2 cataloged preprints. This axis would measure total research output over time, including all venues and years, and should not be estimated from current inputs.
Impact—
No impact score can be computed: no citation counts, h-index, or venue metrics appear in the input data. This axis would measure community uptake of the work — citations and adoption of the proposed frameworks — and is left unscored.
Focus0/100
Focus scores 0, reflecting an even 50/50 split in the field distribution between 'high-performance computing / distributed machine learning systems' and 'distributed machine learning / federated learning systems'. In practice these two clusters are adjacent facets of one agenda — scaling distributed training — so the zero reads more as metric sensitivity to a tiny two-paper sample than as genuine scatter; more cataloged work would clarify true topical concentration.
Collaboration32/100
Collaboration scores 32 on the 0-100 radar, built on a raw count of 4 distinct co-authors in the cataloged record. The pattern is small and stable: repeated first-author partnerships with Marco Aldinucci and Robert Birke dominate. This is typical of a tightly-knit academic group rooted at the University of Turin, but the limited breadth means the score measures a narrow slice of this researcher's collaborative surface; broader co-authorship data would likely revise it upward.
Activity—
No activity score exists because the snapshot lacks year-by-year productivity data; both cataloged papers fall in 2026. This axis would measure publishing cadence and sustained research rhythm across years, and cannot be inferred from a single-year, two-paper sample.
Reproducibility60/100
Reproducibility averages 60 across the 2 papers with a repro-readiness assessment, matching the radar score of 60. The spread is wide: the federated sharded-data-parallelism study is rated high, backed by 512-GPU experiments and an explicit cost model, while the FFL framework paper is rated low, a vision piece whose claimed 2.5–3.69x speedups over OpenFL rest on thin evaluation. The mean therefore blends one solid artifact with one aspirational one.
Research summary
Gianluca Mittone is a researcher at the University of Turin (ORCID 0000-0002-1887-6911) working at the intersection of high-performance computing and distributed machine learning, with federated learning as the connecting thread. The Evolve catalog holds two 2026 arXiv preprints, both first-authored and corresponding-authored. The first, 'Accelerating Sharded Data Parallelism at Scale with Federated Learning', proposes hybrid FL+FSDP/FL+HSDP training and reports strong results on a 512-GPU cluster alongside an explicit cost model; Evolve's review rates its reproducibility readiness as high while flagging that perplexity gains are confounded by unequal effective batch sizes. The second, 'Efficiently Distributed Federated Learning', is a vision paper for FFL, a C/C++ federated-learning framework built on FastFlow with customizable communication graphs, claiming 2.5–3.69x speedups over OpenFL across x86, ARM, and RISC-V; its evaluation is acknowledged as thin and largely future work, with low reproducibility readiness. Bibliometrically the profile is compact: 4 distinct co-authors, a collaboration score of 32, an even field split between HPC/distributed-ML systems and distributed-ML/FL systems, and a reproducibility mean of 60 over two assessed papers. Output, impact, and activity are unscored for lack of data. Taken together, the record describes a systems-oriented researcher pushing federated learning toward HPC-grade scale, with one substantial experimental contribution and one forward-looking framework agenda whose promise still awaits full empirical backing.
Collaboration pattern
The cataloged record shows a compact collaboration footprint: 4 distinct co-authors (collaboration score 32), anchored by repeated first-author work with Marco Aldinucci and Robert Birke. Both 2026 preprints carry Mittone as first and corresponding author, consistent with a driver role inside a small, stable Turin-based systems group rather than a wide international network. Collaboration breadth metrics from a two-paper sample should be read as a floor, not a ceiling.
Research keywords
Highlights
512-GPU federated sharded data parallelism with a cost model
FFL: a C/C++ federated learning framework vision on FastFlow
Timeline
- 2026Publishes the vision preprint 'Efficiently Distributed Federated Learning', laying out FFL, a FastFlow-based C/C++ federated-learning framework with customizable communication graphs.
- 2026Publishes 'Accelerating Sharded Data Parallelism at Scale with Federated Learning', demonstrating hybrid FL+FSDP/FL+HSDP training with strong 512-GPU results and a cost model.
- 2026Both first-author, corresponding-author preprints enter the Evolve review catalog, anchoring a profile concentrated on federated learning and distributed ML systems.
Papers in the Evolve catalog
- first authorcorresponding
Accelerating Sharded Data Parallelism at Scale with Federated Learning
- first authorcorresponding
Efficiently Distributed Federated Learning
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.
- Radar scores are computed from a small bibliometric snapshot covering only the 2 cataloged 2026 arXiv preprints; focus=0 and collaboration=32 are highly sample-sensitive.
- Output, impact, and activity axes are unscored: no citation counts, h-index, publication-volume, or year-by-year productivity data were provided.
- All cataloged works are arXiv preprints; no peer-reviewed venue record appears in the input, so venue-level quality signals are absent.
- Country and career-stage information are not provided; the profile is name-resolved via ORCID 0000-0002-1887-6911 and public bibliographic data.