Nadja Klein

Research profile

Research summary

Nadja Klein is a researcher at Karlsruhe Institute of Technology whose record, name-resolved from OpenAlex bibliographic data, bridges core statistical methodology and two fast-moving application domains. The topic profile is dominated by Bayesian inference (84 topic mentions), statistical distribution estimation (65), Gaussian processes (63), general statistical methods (61) and spatial and panel data analysis (41), with the field split evenly, 50/50, between space weather / solar physics machine-learning forecasting and explainable AI / machine learning evaluation. Headline bibliometrics include an h-index of 24 and an i10-index of 47, and output has accelerated sharply: annual works climbed from single digits in the mid-2010s to a 2024 peak of 47, with 101 works in the most recent three-year window — the basis for a maximum activity score. Evolve's catalog contains two 2026 arXiv papers, both with Klein in a middle author position and neither as corresponding author: PROSWIN, a deep distributional-regression pipeline producing the first fully probabilistic machine-learning forecast of solar wind speed from solar images with claimed sub-1% calibration deviation (high reproducibility readiness), and XAI-Arena, an LLM-as-a-judge framework benchmarking five XAI methods across eight explanation-quality dimensions (medium readiness). The overall picture is a statistician supplying probabilistic modelling and evaluation rigor to interdisciplinary teams of eight-plus distinct coauthors spanning solar physics and machine learning.

Research keywords

Highlights

• PROSWIN: first fully probabilistic ML forecast of solar wind speed

  This 2026 arXiv paper reports a deep distributional regression pipeline that derives probabilistic solar wind speed forecasts directly from solar images using a Swin Transformer backbone. Evolve's review finds the evidence strong for the claim of the first fully probabilistic machine-learning forecast in this space, with under 1% calibration deviation, though high-speed stream peaks still trail persistence; reproducibility readiness is rated high.

• XAI-Arena: can LLMs judge explanation quality?

  This 2026 arXiv paper introduces an LLM-as-a-judge framework using GPT-5.4 to benchmark five explainable-AI methods, including SHAP, LIME and counterfactual explanations, across eight explanation-quality dimensions. Evolve judges the design solid but flags thin human validation and pseudoreplicated statistics; reproducibility readiness is medium.

• Sustained productivity acceleration

  Annual work counts in the bibliometric record climb from a single work in 2009 to 21 in 2020, 34 in 2023 and a peak of 47 in 2024, with 25 already logged for 2026 and 101 works in the most recent three-year window. This trajectory underlies the maximum activity score and mirrors the expansion into space weather and XAI application work.

Timeline

1. 2009First year with recorded works (1) in the OpenAlex bibliometric record.
2. 2020Output reaches 21 works, beginning a steep multi-year climb in publication volume.
3. 202334 works logged as the productivity acceleration continues.
4. 2024Peak year in the record with 47 works.
5. 2026Two arXiv preprints enter Evolve's catalog with Klein as middle author: PROSWIN (probabilistic solar wind speed forecasting) and XAI-Arena (LLM-as-a-judge evaluation of XAI methods), with 25 works already logged this year.

Papers in the Evolve catalog

• 2026-09-22

  PROSWIN: Probabilistic Solar Wind Speed Forecasting Using Deep Distributional Regression From Solar Images

• 2026-09-08

  XAI-Arena: Can LLMs Assess the Quality of XAI Explanations?

Frequent co-authors