Anomalous sound detection researcher at Aalborg University deriving AUC bounds for anomaly-free model selection
2
papers here
2
reviewed by Evolve
2
first-author
0
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—
This axis is not yet scored: no publication counts or venue history are available in the current snapshot. It would measure total and per-year output across the researcher's career; the present record covers only two reviewed 2026 preprints.
Impact—
Impact is not yet scored: the snapshot contains no citation or influence data. This axis would measure community uptake of the AUC-bound and pseudo-anomaly methods; without citation records in scope, no score or estimate is attempted.
Focus0/100
The radar records a focus score of 0 on a 0–100 scale. The underlying field distribution splits evenly, 0.5/0.5, between two nearly identical labels — machine learning (audio anomaly detection) and machine learning — anomaly detection / anomalous sound detection — and both catalogued 2026 preprints sit inside anomalous sound detection. The low score therefore appears driven by the label split rather than genuine topical scatter.
Collaboration28/100
The collaboration dimension scores 28/100. The underlying raw metric lists 3 distinct co-authors across the reviewed record, with Zheng-Hua Tan the recurring collaborator in the frequent-co-author data. Both catalogued 2026 preprints are first-author, non-corresponding papers. Together this depicts a small, stable working pattern rather than broad, networked collaboration across many groups.
Activity—
Activity is not yet scored: the snapshot has no multi-year publication timeline. This axis would measure publishing cadence and sustained engagement over time; with only two 2026 preprints in scope, the record is too short to characterize an activity trajectory.
Reproducibility55/100
Reproducibility scores 55/100, a mean of 55 computed over 2 papers — all of the catalogued 2026 preprints. Evolve rates each as medium reproducibility readiness: the core objects (AUC bounds, a lower bound for kNN detectors, differentiable ensemble objectives) are precisely stated, while end-to-end replication of the reported DCASE results remains only moderately ready.
Research summary
Kevin Wilkinghoff is a researcher at Aalborg University working on machine learning for anomaly detection, focused on anomalous sound detection (ASD). His Evolve-catalogued output consists of two 2026 arXiv preprints, both with him in first-author position, that approach ASD evaluation from a theoretical angle built on AUC bounds. 'Anomaly-Free Self-Optimization via AUC Bounds' recasts AUC bounds as a differentiable objective for tuning ASD ensembles without labeled anomalies, reporting strong DCASE gains under LDN scores but no improvement for plain neural networks. 'Can We Predict Anomaly Detection Performance from Embedding-Space Geometry?' derives an AUC lower bound for kNN detectors and shows that pseudo-anomaly probes enable anomaly-free model selection on DCASE, though the strongest constructions require metadata. Evolve rates both papers at medium reproducibility readiness. Bibliometric coverage is still thin: the record lists 3 distinct co-authors, with Zheng-Hua Tan as the recurring collaborator, and the output, impact, and activity radar axes remain unscored pending broader data.
Collaboration pattern
Collaboration is compact: the reviewed record lists only 3 distinct co-authors, yielding a radar collaboration score of 28/100. Zheng-Hua Tan is the recurring partner in the frequent-co-author data. Both 2026 preprints are first-author, non-corresponding papers, suggesting Wilkinghoff drives the technical core within a small, stable team rather than a large distributed network.
Research keywords
Highlights
AUC bounds turned into anomaly-free self-optimization
Predicting detection performance from embedding-space geometry
Pseudo-anomaly probes for anomaly-free model selection on DCASE
Timeline
- 2026arXiv preprint 'Can We Predict Anomaly Detection Performance from Embedding-Space Geometry?' derives an AUC lower bound for kNN detectors and enables pseudo-anomaly-based model selection.
- 2026arXiv preprint 'Anomaly-Free Self-Optimization via AUC Bounds' recasts AUC bounds as a differentiable objective, with strong DCASE gains under LDN scores.
- 2026Both 2026 first-author preprints concentrate on anomalous sound detection, marking a single-year, tightly focused research agenda on anomaly-free evaluation and model 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.
- Bibliometric scores are computed from an OpenAlex-based snapshot only; no other bibliographic sources were used.
- No ORCID or country data was available, so this profile is name-resolved from public bibliographic records.
- Output, impact, and activity radar axes are unscored because the snapshot lacks publication counts, citations, and multi-year history.
- Both catalogued papers are 2026 arXiv preprints; peer-reviewed venue status is not captured in the input data.