Nils Feldhus — Explainable NLP researcher at Groningen auditing LLM explanation faithfulness and causal grounding
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—
No reliable output totals are available yet. This axis would measure aggregate publication volume and its spread across venues and years, but the underlying data is insufficient in this snapshot, so it is left unscored rather than estimated.
Impact—
Impact data is not yet available: citation counts, citation velocity, and field-normalized influence could not be computed for this profile. The axis would measure how widely the work is cited and used, and it remains unscored.
Focus0/100
The focus dimension scores 0, reflecting a field distribution split exactly 50/50 between 'nlp / mechanistic interpretability' and 'natural language processing (explainable AI evaluation)'. In practice these are closely adjacent subfields, so the low score signals breadth within a narrow explanatory-NLP band rather than scatter across unrelated disciplines; five assigned topics led by XAI (45 works) keep the thematic center coherent.
Collaboration55/100
The collaboration dimension scores 55, computed from 15 distinct coauthors across 48 works in the recent three-year window. The recurring partner list is concentrated rather than broad: Qianli Wang, Simon Ostermann, Yilong Wang, Jingyi Sun, Jing Yang, Isabelle Augenstein, Pepa Atanasova, Antonin Poché, and Fanny Jourdan appear repeatedly. Together with non-corresponding author roles on both reviewed papers, this points to stable small-team co-authorship rather than large distributed consortia.
Activity100/100
Activity scores a maximum 100, driven directly by the yearly output curve in the raw data: 3 works in 2020, rising to 7 in 2022, 12 in both 2024 and 2025, and 24 recorded in 2026 — roughly an eightfold increase across the observed window. With 48 works falling in the most recent three-year span, the profile shows sustained and accelerating publication activity rather than sporadic bursts.
Reproducibility100/100
Reproducibility scores 100, the mean across 2 papers with reproducibility assessments in the catalog. Both reviewed 2026 arXiv papers are rated 'high' on reproducibility readiness: the chain-of-thought faithfulness study validates its causal metric Δp across five LLMs, and the simulatability critique proposes actionable fixes for its evaluation protocol, consistent with artifact- and evaluation-transparent working habits.
Research summary
Nils Feldhus is an NLP and explainable-AI researcher at the University of Groningen (Netherlands), profiled from OpenAlex bibliographic records linked to ORCID 0009-0008-7408-7483. His topic profile is dominated by Explainable Artificial Intelligence (45 assigned works), followed by Topic Modeling (36), Natural Language Processing Techniques (18), Adversarial Robustness in Machine Learning (16), and Speech and Dialogue Systems (15); his field distribution splits evenly (0.5/0.5) between NLP/mechanistic interpretability and NLP with an explainable-AI-evaluation emphasis. Output has accelerated sharply: yearly work counts rise from 3 in 2020 and 5 in 2021 to 12 in both 2024 and 2025, and 24 in 2026, with 48 works in the recent three-year window and an h-index of 10 (i10-index 10). Evolve has reviewed two 2026 arXiv papers: a last-author intervention study reframing chain-of-thought faithfulness as sparse-autoencoder concept grounding, validating the causal metric Δp across five LLMs, and a middle-author self-critique showing that LLM simulators can bypass explanations by leaning on task priors and leaking anonymized labels. Both papers carry high reproducibility readiness, consistent with a computed reproducibility score of 100 (mean across 2 assessed papers), while a collaboration score of 55 reflects 15 distinct recent coauthors and small-team working patterns.
Collaboration pattern
With 15 distinct coauthors across 48 recent works (collaboration score 55), Feldhus works in recurring small teams rather than large rotating consortia. Qianli Wang and Simon Ostermann appear most often, alongside Yilong Wang, Jingyi Sun, Jing Yang, Isabelle Augenstein, Pepa Atanasova, Antonin Poché, and Fanny Jourdan. On the two Evolve-reviewed papers he holds non-corresponding senior and middle positions, consistent with contributing inside multi-author groups.
Research keywords
Highlights
Reframing chain-of-thought faithfulness as causal concept grounding
Showing LLM simulators can bypass explanations
Timeline
- 20203 works recorded in the OpenAlex yearly counts — the earliest year in the observed window.
- 2022Output reaches 7 works, continuing the gradual climb from 5 works in 2021.
- 2024A step change: 12 works recorded, well above the 5 works of 2023 and double the 2022 level.
- 2025Activity holds at 12 works, with 48 works now falling inside the most recent three-year window.
- 202624 works recorded; two arXiv preprints on chain-of-thought faithfulness and the limits of automated simulatability enter Evolve's reviewed catalog.
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 figures (h-index 10, i10-index 10, yearly counts, topics) come from an OpenAlex-resolved identity and may undercount or merge records.
- The output and impact radar axes are unscored because reliable totals and citation data are not available in this snapshot.
- Both Evolve-reviewed papers are 2026 arXiv preprints; peer-review status is not reflected in this profile.
- The name is resolved from public bibliographic data; affiliation and coauthor links reflect the matched OpenAlex profile.