Statistical physicist theorizing neural learning: SGD dynamics, auxiliary tasks, and coarse-grained representations
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
10
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 scored yet: this record lacks verified author-level output totals (full-count publications, position-weighted productivity). The axis would measure the volume and consistency of publications over time, here only partly visible via yearly counts.
Impact—
Not scored yet: verified citation-impact aggregates (field-normalized citations, percentile benchmarks) are missing for this record, though a raw h-index of 10 appears in the snapshot. This axis would measure influence on subsequent research.
Focus0/100
The radar assigns focus a score of 0/100, yet the raw field distribution reads 50% machine-learning theory (statistical physics of learning) and 50% statistical physics of machine learning — two adjacent labels for essentially one research program — and the topic list (neural networks 24, neural dynamics 11, theoretical physics 8, stochastic gradient optimization 7, complex networks 5) stays within theoretical/computational themes. The zero likely reflects the even split across synonymous labels rather than genuine dispersion; treat it with caution.
Collaboration39/100
Collaboration scores 39/100, the lowest of the scored axes: only 6 distinct coauthors appear in the recent window covering 16 works, and both catalog papers place him in middle and last (non-corresponding) positions rather than a large-team lead. Recurring partners include Federico Milanesio, Matteo Osella, Margherita Mele, Andrea Castagna, Roberto Menichetti, and Raffaello Potestio, suggesting small, stable theory-oriented working groups.
Activity83/100
Activity is the strongest axis at 83/100. Yearly OpenAlex counts run from 2014 through 2026 with acceleration in the recent period: 6 works in 2024 and 8 already in 2026, against 2 in 2025, and 16 works in the last three years. The 2026 figure — the highest single-year count in the record — is consistent with the two arXiv preprints Evolve has reviewed this year.
Reproducibility78/100
Reproducibility scores 78/100 (mean over the 2 Evolve-reviewed papers). The analytical auxiliary-learning paper rates high readiness — closed-form steady-state results for teacher-student online SGD, though idealized with fixed readouts, Gaussian inputs, and small μ — while the mapping-entropy neuron-selection paper rates medium, with coherent small-scale synthetic results but thin baselines. Two scored papers is a small base, so the score should be read as provisional.
Research summary
Alessandro Ingrosso, based at Radboud University Nijmegen (Netherlands), works at the intersection of statistical physics and machine-learning theory. His OpenAlex-resolved record (h-index 10, i10-index 10) shows a publication curve that began around 2014, accelerated to 6 works in 2016, and is now at its most active: 8 works already recorded in 2026, 16 in the last three years, and an activity score of 83/100. Topic labels concentrate on neural networks and applications (24 works), neural dynamics and brain function (11), theoretical and computational physics (8), stochastic gradient optimization (7), and complex network analysis (5); the reported field split is even between machine-learning theory (statistical physics of learning) and statistical physics of machine learning — two views of one research program. Two 2026 arXiv papers in the Evolve catalog illustrate the method: a closed-form, steady-state analytical theory of auxiliary learning in teacher-student online SGD built on mean-field ODEs (high reproducibility readiness, though idealized with fixed readouts, Gaussian inputs, and small μ), and mapping-entropy-based unsupervised neuron selection, judged coherent on small synthetic tasks but narrow in scope. Reproducibility reads 78/100 over 2 scored papers. Collaboration is comparatively thin (39/100) with 6 distinct recent coauthors, and citation-based output and impact axes are not yet scored.
Collaboration pattern
Collaboration metrics point to small-team theory work: a radar score of 39/100, 6 distinct coauthors over 16 recent works, and recurring partners including Federico Milanesio, Matteo Osella, Margherita Mele, Andrea Castagna, Roberto Menichetti, and Raffaello Potestio. On both reviewed 2026 arXiv papers he holds a middle or last, non-corresponding position, consistent with a contributing theorist inside stable multi-author groups rather than a large-lab corresponding lead.
Research keywords
Highlights
An analytical theory of auxiliary learning
Mapping entropy for unsupervised neuron selection
Timeline
- 2014First works appear in the OpenAlex record: 1 publication in 2014 and 1 in 2015.
- 2016First output peak: 6 works recorded in 2016.
- 2024Output re-accelerates: 6 works recorded in 2024.
- 2026Most active year on record: 8 works so far, including two arXiv preprints reviewed by Evolve — the analytical theory of auxiliary learning and mapping-entropy-based neuron selection.
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.
- Bibliometrics come from a single OpenAlex identity snapshot (identity: openalex); counts may over- or under-represent the person if author disambiguation merges or splits records.
- The focus axis reads 0/100 even though the raw field split (0.5/0.5) contrasts two near-synonymous statistical-physics-of-learning labels; this score-versus-raw discrepancy is unresolved in the snapshot.
- Output and impact axes are unscored; a raw h-index of 10 and yearly counts exist, but normalized impact benchmarks and verified output totals are absent.
- Evolve coverage is thin: only 2 papers (both 2026 arXiv preprints) carry review verdicts and reproducibility scores.