Get Noema

Papers / explorations

Noema Research.

Understanding models.
Opening up what comes next.

Research from NoemaAI and our collaborators, shared openly. A growing collection of papers, experiments, and ideas.

Explore publications

Publications

01 paper

Preprint / Open access

  • Probability aggregation
  • Generalization
  • Semantic parsing

Reusing Model Probabilities for New Queries

  • Armin StamateNoemaAI
  • Victoria Paraoulaki de MirandaHarvard University
  • Wyatt WunkerStanford University

Can one set of model probabilities answer many different questions? This paper studies how fine-grained predictions can be combined into broader categories, keeping answers consistent across label groupings without querying a model again for each grouping.

Experiments examine intent classification, logical queries, and semantic parsing. The practical benefits depend on prompt detail, training, and the questions being asked: consistency is guaranteed by construction, while better predictive performance is not. Reuse was faster in the measured workloads with two or more groupings of the same input.

Read on Zenodo Download paper (PDF)
DOI: 10.5281/zenodo.23584994
Cite this paper

Stamate, A., Paraoulaki de Miranda, V., & Wunker, W. (2026). Reusing Model Probabilities for New Queries [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.23584994