Gitta Kutyniok

Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR) · Germany

Unclaimed · AI-structured profileORCID OpenAlex

Applied harmonic analyst turned deep-learning theorist: expressivity, spiking networks, and pruning at DLR

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—

Unscored: this snapshot lacks the normalized publication-volume data Evolve needs. The output axis would measure productivity relative to field norms over time, such as total and per-year publication counts adjusted for field and career stage; no estimate is offered here.

Impact—

Unscored: although raw h-index (47) and i10-index (136) appear in the snapshot, the field-normalized citation components the impact axis requires are not populated, so Evolve reports no impact score. This axis would measure citation influence relative to field norms.

Focus100/100

Scores 100 for focus. The OpenAlex topic profile is strongly concentrated: 184 works tagged Mathematical Analysis and Transform Methods and 167 in Sparse and Compressive Sensing Techniques, followed by Image and Signal Denoising (84), Neural Networks and Applications (44), and Numerical methods in inverse problems (34). The catalog's field distribution adds a single recent line, machine learning theory (neural network expressivity), showing a tight core now extending into deep-learning theory.

Collaboration41/100

Scores 41, a mid-range value. Over the most recent three-year window the snapshot records 68 works with only 7 distinct coauthors, and the recurring-collaborator list holds seven names: Pascal Esser, Miriam Kranzlmüller, Manjot Singh, Guido Montúfar, Maria Matveev, Ayush Bharadwaj, and Lucius Bushnaq. Both reviewed catalog papers place her as last author (not corresponding), a pattern of small, stable teams rather than large ad-hoc consortia.

Activity100/100

Scores 100. The snapshot records 68 works in the last three years, and the year-by-year counts stay high through the current window: 16 in 2024, 26 in 2025, and 26 in 2026, against a record reaching back to 1998 (1 count). This is sustained, near-continuous production across nearly three decades with a visibly intensified recent pace.

Reproducibility55/100

Scores 55, computed from a reproducibility mean of 55 over just one assessed paper, so coverage is thin. Within the catalog, the functional-degeneracy paper is rated high reproducibility readiness while the polyhedral time-to-first-spike paper is rated medium; both are arXiv preprints. Treat this axis as directional rather than settled until more papers are assessed.

Research summary

Gitta Kutyniok, based at the German Aerospace Center (DLR) in Germany, is an applied mathematician whose record spans nearly three decades: the OpenAlex snapshot traces activity back to 1998 and reports an h-index of 47 and an i10-index of 136. Her topic profile is anchored in Mathematical Analysis and Transform Methods (184 works) and Sparse and Compressive Sensing Techniques (167), extending through Image and Signal Denoising (84), Neural Networks and Applications (44), and Numerical methods in inverse problems (34). Production remains intense, with 68 works in the last three years, and recent focus (scored 100) has shifted toward deep-learning theory: expressivity of time-to-first-spike spiking networks via polyhedral geometry, functional degeneracy measured by behavioral recovery rank for pruning, and an expressivity-normalized energy comparison of ANNs and SNNs. Evolve's catalog holds three 2026 arXiv papers, two already reviewed, on all of which she is last author. Collaboration (scored 41) is compact: 7 distinct coauthors recently, including Guido Montúfar and Pascal Esser.

Collaboration pattern

Recent collaboration is compact: 68 works over three years with only 7 distinct coauthors, giving a mid-range collaboration score of 41. Recurring partners include Pascal Esser, Miriam Kranzlmüller, Manjot Singh, Guido Montúfar, Maria Matveev, Ayush Bharadwaj, and Lucius Bushnaq. On all three catalog papers she is the last author (not corresponding), a pattern pointing to a senior supervisory position in small theory-oriented teams rather than large consortia.

Research keywords

applied harmonic analysissparse approximationcompressive sensingspiking neural networkstime-to-first-spike codingneural network expressivitypolyhedral geometrylinear regionsfunctional degeneracyneural network pruningmodel compressioninverse problems

Highlights

Polyhedral bounds on time-to-first-spike spiking networks
This 2026 arXiv paper uses polyhedral geometry to bound the causal regions of TTFS spiking networks, proving tight shallow-network counts and depth-folding results. Evolve's verdict: rigorous, but the deep bounds are loose and the empirical checks are initialization-based; reproducibility readiness is rated medium.
Behavioral recovery rank: measuring and exploiting functional degeneracy
This paper introduces behavioral recovery rank, built on a behavioral Hessian, to quantify functional degeneracy and guide pruning and teacher-student compression. Evolve calls it a clean small-scale study while noting that the pruning-gap headline is softened by its own refitting ablation; reproducibility readiness is high.
Expressivity-normalized energy comparison of ANNs and SNNs
A 2026 arXiv preprint proposing a comparison of artificial and spiking networks under equalized expressivity when assessing energy demand. It is listed in the Evolve catalog without keywords, verdict, or reproducibility rating yet, so no evaluation is available.

Timeline

2026Last-authored arXiv paper on polyhedral geometry of time-to-first-spike spiking networks, reviewed by Evolve with medium reproducibility readiness.
2026Last-authored arXiv paper introducing behavioral recovery rank for functional degeneracy measurement and pruning, rated high for reproducibility readiness.
2026arXiv preprint towards an expressivity-normalized energy-demand comparison of ANNs and SNNs enters the Evolve catalog, review pending.

Papers in the Evolve catalog

2026-09-10
Polyhedral Geometry of Time-to-First-Spike Neural Networks
2026-08-31
Functional Degeneracy in Neural Networks: Measurement and Pruning

Frequent co-authors

Pascal Esser×2 Miriam Kranzlmüller×1 Manjot Singh×1 Guido Montúfar×1 Maria Matveev×1 Ayush Bharadwaj×1 Lucius Bushnaq×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.

Bibliometrics come from a single OpenAlex snapshot; identity is name-resolved from public bibliographic data, and no other sources were consulted.
Output and impact dimensions lack the data Evolve needs and are left unscored in this profile.
The reproducibility score rests on a single assessed paper (mean 55 from 1 paper), so its coverage is thin.
Only three 2026 arXiv papers from this researcher are in the Evolve catalog, and one of them has not yet been reviewed.