data science

Can We Trust This Evaluation? Instance-Level Reliability of LLM-as-a-Judge

LLM-as-a-judge evaluation commonly aggregates multiple judges'decisions into a majority vote, yet this panel decision process does not guarantee that the evaluation of an individual item is reliable. We propose Structured Cross-fitted Uncertainty Risk (SCUR), a label-free approach for estimating instance-level evaluati...

Predicting how model propensities generalize out of distribution

How do language models generalize from their training data? This is a central challenge for alignment: models reliably learn their training distribution given enough samples, but we want models that behave well when deployed in new contexts. We study out-of-distribution (OOD) generalization of propensities. Our central...

Impact in 2025

The Consortium of European Social Science Data Archives (CESSDA ERIC) is a distributed research infrastructure that operates through a network of 22 Service Providers and 12 partners across Europe. In 2025, the consortium worked collectively to deliver on its strategic priorities: Data, People, and Landscape. It preser...

A RISK-BASED QUALITY ASSURANCE FRAMEWORK FOR FHIR-BASED CLINICAL APPLICATIONS: ALIGNING VERIFICATION EFFORT WITH CLINICAL CONSEQUENCE

Verification effort for FHIR-based clinical systems can be allocated by clinical consequence rather than by test-suite convention, extending quality obligations into the operational-behavior layer where specification-conformant systems have been shown to fail.

delhirolls2002-43

Unprocessed raw data: the Delhi electoral roll of the 2002 special intensive revision (the reference roll for the later revision), constituency 43 in the constituency numbering valid until the 2008 delimitation. One PDF per polling part, named U05_043_ .pdf (three-digit constituency and part numbers). Per-part PDFs ret...

delhirolls2002-55

Unprocessed raw data: the Delhi electoral roll of the 2002 special intensive revision (the reference roll for the later revision), constituency 55 in the constituency numbering valid until the 2008 delimitation. One PDF per polling part, named U05_055_ .pdf (three-digit constituency and part numbers). Per-part PDFs ret...

Accuracy of retinal-image artificial intelligence for atherosclerosis on coronary CT and vascular ultrasound: a systematic review and meta-analysis

RETROSPECTIVE REGISTRATION. This registration documents the protocol and methods of a completed review; it was created after the review was finished. OBJECTIVE To estimate the accuracy, as the area under the ROC curve (AUC), of artificial intelligence applied to retinal images for detecting atherosclerosis defined by i...

delhirolls2002-35

Unprocessed raw data: the Delhi electoral roll of the 2002 special intensive revision (the reference roll for the later revision), constituency 35 in the constituency numbering valid until the 2008 delimitation. One PDF per polling part, named U05_035_ .pdf (three-digit constituency and part numbers). Per-part PDFs ret...

CarbideMatterGen: code, model and data for the ε-W2C/WOx CO2-reduction study

Code, fine-tuned model weights and data accompanying "Sacrificial rare-earth extraction closes the C1 channel on an epsilon-W2C/WOx catalyst that converts CO2 into ethanol and acetone" (Cell Reports Physical Science, 2026). CarbideMatterGen is MatterGen fine-tuned on 5,275 transition-metal carbides with nine conditioni...

Revealing parasitoid-pest interactions using multi-marker metabarcoding: a nationwide assessment of parasitism in cabbage stem flea beetle larvae

Management of insect pests of oilseed rape (Brassica napus L.) such as cabbage stem flea beetle (CSFB; Psylliodes chrysocephala L.) is increasingly challenging given a lack of effective permitted insecticides. Conservation biocontrol, including through naturally-occurring parasitoids of CSFB, is one of the first lines...

Metadata vs Arsenic: a Case Study of Digital Library for the Conservation of Physical Collections

Libraries worldwide have begun to uncover the presence of arsenic in their collections, posing risks to collections, staff, and the broader public. KB National Library of the Netherlands is at the forefront of addressing this issue, having tested a large sample of items for arsenic. Building on these testing practices...

Bringing topological heterogeneities into the study of time-series data on militarized international disputes: a deep learning approach