Hinrich Schütze

Ludwig-Maximilians-Universität München · Germany

Unclaimed · AI-structured profileOpenAlex

NLP researcher at LMU Munich: LLM robustness evaluation, cross-lingual consistency, and quantization-aware training.

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—

Output is not scored: the snapshot carries no complete publication-count series for this author, so productivity cannot be computed yet. This axis would measure total volume and cadence of outputs — papers, preprints, and other research artefacts over time.

Impact—

Impact is not scored: no citation counts, citation distribution, or influence metrics are available in this snapshot. The axis would measure how widely the work is cited and taken up — citations per paper, citation trajectory, and uptake across subfields.

Focus0/100

The OpenAlex field distribution splits exactly 0.5/0.5 between natural language processing (LLM robustness evaluation) and natural language processing (computational linguistics), and topic counts fan out across five clusters — topic modeling (17), NLP techniques (8), adversarial robustness (7), advanced neural network applications (6), and multimodal machine learning (6). That spread produces a focus score of 0: a deliberately broad portfolio rather than a single dominant niche.

Collaboration64/100

Twenty-four distinct coauthors appear in the last three years, and twelve recur across the catalog — including Jiamin Song, Mengjie Zhao, Zijing Wang, Yongkang Liu, Shi Feng, Feiliang Ren, and Hazel H. Kim. He sits mid-list on the two benchmark/evaluation papers and takes the last position on the quantization survey. The collaboration dimension scores 64, a mid-high reading consistent with steady team-based output across groups.

Activity95/100

Activity scores 95, the strongest axis in the snapshot. OpenAlex records 25 works over the last three years, with the countsByYear series showing all 25 landing in 2026 — an intense recent cluster of arXiv-first evaluation and efficiency work. All three catalog papers (FramingQA, the cross-lingual consistency evaluation, and the quantization-aware training survey) are dated 2026, matching the surge the bibliometric data captures.

Reproducibility55/100

Reproducibility scores 55, computed as a mean over 2 scored papers. Both reviewed papers carry medium reproducibility readiness: FramingQA ships a 54k-question benchmark with a documented construction process, while the cross-lingual consistency study is controlled and useful but lacks significance tests. A 55 points to reproducible-with-effort work — artefacts and protocol are described, yet independent replication would need additional detail.

Research summary

Hinrich Schütze is a natural language processing researcher at Ludwig-Maximilians-Universität München whose recent catalog centers on rigorous evaluation of large language models. An OpenAlex-resolved snapshot records 25 works in the last three years (25 in 2026 alone), 24 distinct coauthors, and a topic profile led by topic modeling (17), NLP techniques (8), adversarial robustness (7), advanced neural network applications (6), and multimodal machine learning (6). His field distribution splits evenly between LLM robustness evaluation and computational linguistics (0.5 each), yielding a focus score of 0 — a broad, multi-cluster portfolio rather than a narrow niche. Three reviewed 2026 arXiv papers illustrate the range: FramingQA, a 54k-question benchmark showing LLM answers flip with question framing (binary strict accuracy 0%); a systematic evaluation of cross-lingual consistency enhancement methods finding post-training most reliable; and a target-centric survey of quantization-aware training, where he is last author. Evolve scores activity at 95, collaboration at 64, and reproducibility at 55 (mean across 2 scored papers, both rated medium readiness), while output and impact lack sufficient data and are left unscored.

Collaboration pattern

Collaboration is team-based and recurring: 24 distinct coauthors in the last three years, with twelve frequent collaborators including Jiamin Song, Mengjie Zhao, Zijing Wang, Yongkang Liu, Shi Feng, Feiliang Ren, Daling Wang, and Hazel H. Kim. Authorship sits mid-list on the two evaluation papers (FramingQA, cross-lingual consistency) and last on the quantization-aware training survey. The collaboration dimension scores 64, consistent with steady cross-group work spanning robustness evaluation, multilinguality, and efficiency.

Research keywords

framing effectLLM robustnessbenchmark evaluationprompt sensitivitysycophancycross-lingual consistencymultilingual language modelspost-training alignmentrepresentation interventionquantization-aware trainingtopic modelingcomputational linguistics

Highlights

FramingQA: a 54k-question benchmark exposing framing-driven answer flips in LLMs
The benchmark shows LLM answers flip with question framing, with binary strict accuracy falling to 0%. Evolve calls it a well-built 54k-question benchmark while noting that the novelty and scale claims need firmer support.
Systematic evaluation finds post-training the most reliable route to cross-lingual consistency
A unified comparison of cross-lingual consistency enhancement methods in multilingual language models concludes that post-training is the most reliable approach; Evolve rates it controlled and useful, but flags the absence of significance tests and that DCO is the authors' own method.
Target-centric survey of quantization-aware training, from the last author position
A 2026 arXiv survey of quantization-aware training organized around target objectives, with this researcher in the last author position. It appears in the catalog without an Evolve verdict, keyword set, or reproducibility score yet.

Timeline

2026Publishes FramingQA, a 54k-question benchmark measuring the compositional framing effect in LLM answers.
2026Releases a systematic evaluation of cross-lingual consistency enhancement methods, finding post-training the most reliable.
2026Authors a target-centric survey of quantization-aware training in the last author position.
2026OpenAlex records 25 works for the year, with 24 distinct coauthors across the three-year window.

Papers in the Evolve catalog

Frequent co-authors

Jiamin Song×1 Mengjie Zhao×1 Zijing Wang×1 Yongkang Liu×1 Qian Li×1 Shi Feng×1 Feiliang Ren×1 Daling Wang×1 Hazel H. Kim×1 Andrew M. Bean×1 Guilherme Affonso Ferreira de Camargo×1 Shanyu Chauhan×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 an OpenAlex snapshot only; the profile is name-resolved from public bibliographic data.
The snapshot reports h-index 1 and i10-index 0, which conflicts with the paper catalog and likely reflects a name-resolution artifact; treat impact as unmeasured.
Output and impact dimensions have insufficient underlying data and are intentionally left unscored.
Only three catalog papers are reviewed; the quantization-aware training survey has no Evolve verdict, keywords, or reproducibility score yet.