Leakage-free, efficient neural architecture search for cross-subject physiological pain assessment
2
papers here
2
reviewed by Evolve
2
first-author
1
corresponding
—
works · OpenAlex
—
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—
Insufficient data: no publication counts or venue records are available in this snapshot, so research output cannot be scored. This axis would measure total production — number of papers per year and their spread across journals, conferences and preprint servers.
Impact—
Insufficient data: citation counts and related influence metrics are not available in this snapshot, so impact is left unscored. This axis would measure how often the work is cited and taken up by others in neural architecture search and affective computing.
Focus0/100
The focus score is 0 because the field distribution splits exactly evenly (0.5/0.5) between two labelled areas: 'machine learning — affective computing / physiological signal processing' and 'machine learning (AutoML / affective computing)'. In practice both catalogued 2026 papers target the same problem — architecture search under leave-one-subject-out evaluation on the BioVid heat-pain dataset — so the even split likely reflects label granularity more than genuine topic scattering.
Collaboration14/100
Collaboration scores 14: the snapshot records only 1 distinct co-author, with Friedhelm Schwenker listed as the frequent collaborator. Both catalogued papers are first-authored (one also corresponding), consistent with a small, stable two-person working pattern rather than a broad network. This is measured over a very thin data scope — two 2026 arXiv preprints — so the score says more about sample size than about an unwillingness to collaborate.
Activity—
Insufficient data: only two 2026 papers are recorded, with no multi-year publication history, so research activity cannot be scored. This axis would measure publishing cadence and sustained output over recent years.
Reproducibility55/100
Reproducibility averages 55/100 across the 2 papers Evolve has reviewed, both rated 'medium' readiness. This aligns with the work's own agenda — explicit leakage control and cutting nested LOSO search cost from ~90h to ~12h (≈8x) on BioVid makes reruns cheaper — while single-dataset evaluation and an untested +0.60pp margin temper confidence in exact replication.
Research summary
Heinke Hihn is a machine learning researcher at IU International University of Applied Sciences working where neural architecture search (NAS) meets affective computing on physiological signals. The Evolve catalog holds two first-authored 2026 arXiv preprints, one also corresponding-authored, both tackling a concrete methodological bottleneck: nested leave-one-subject-out (LOSO) evaluation makes architecture search for cross-subject pain assessment expensive and, if done naively, prone to data leakage. The proposed block-shared, leakage-controlled search reduces nested LOSO search cost on the BioVid Heat Pain dataset from roughly 90 hours to about 12 hours (around 8x) at comparable accuracy. Evolve's reviews judge the approach sound while noting limits: evaluation rests on a single dataset, the gains are efficiency rather than accuracy, and a reported +0.60pp improvement over baseline lacks significance testing. The field distribution splits evenly between machine learning — affective computing / physiological signal processing and machine learning (AutoML / affective computing). The measured co-author network in this snapshot is minimal — one distinct co-author, with Friedhelm Schwenker the frequent collaborator — and reproducibility readiness averages 55/100 across the two reviewed papers, both rated medium.
Collaboration pattern
The measured collaboration footprint is minimal: 1 distinct co-author in the snapshot and a collaboration score of 14. Friedhelm Schwenker appears as the frequent co-author, and both catalogued 2026 papers are first-authored by Hihn (one also corresponding), suggesting a stable two-person collaboration on NAS for physiological signals. With only two papers in scope, this should be read as a small-sample pattern rather than a definitive collaboration style.
Research keywords
Highlights
Nested LOSO search cut from ~90h to ~12h on BioVid
~8x cheaper leakage-controlled NAS at equal accuracy
Multimodal physiological signals with early fusion for automated pain assessment
Timeline
- 2026Publishes a first-authored arXiv preprint on efficient architecture search under leave-one-subject-out evaluation for automated pain assessment.
- 2026Publishes a first- and corresponding-authored arXiv preprint on leakage-free NAS, reporting nested LOSO search cost cut from ~90h to ~12h on BioVid.
- 2026Both catalogued papers consolidate one research line — block-shared, leakage-controlled NAS for cross-subject physiological pain assessment — with Friedhelm Schwenker as co-author.
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 dimension scores come from public bibliometric data only (OpenAlex-scope records); unindexed output is not reflected.
- Only the two Evolve-reviewed 2026 arXiv preprints feed this profile; output, impact and activity dimensions are missing because no citation or production data was available.
- This profile is name-resolved from public bibliographic data (ORCID 0000-0002-3244-3661); only the provided affiliation is asserted and no biographical details are inferred.
- Reproducibility rests on just 2 medium-readiness papers; per Evolve review, the leakage-free NAS paper's +0.60pp margin lacks significance testing and evaluation uses a single dataset (BioVid).