EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Metamorphic Evidence as a Runtime Reliability Signal for Small Language Models

Bongjin Jung iD, Joongho Chang iD

Sensors · 31.10.2026 · 26 (21) · 6918

DOI: 10.3390/s26216918 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Sensors, online erschienen am 31.10.2026. Verfasst von Bongjin Jung und 1 weiteren Autor:innen. Beteiligte Einrichtung: Program in AI Convergence Engineering, aSSIST University, Seoul 03767, Republic of Korea. Hauptthemen laut MEDLINE: Language. Die Arbeit nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Small language models often run locally, where no larger model can check their answers, and the runtime exposes tokens and log-probabilities but not hidden states. MOME (Module of Metamorphic Evidence) is a label-free runtime gate. It edits each problem twice by deterministic rules, once keeping the correct answer and once changing a condition on which that answer depends. A predictor fitted offline on other backbones scores the answers. A two-stage policy serves high-scoring answers and verifies the rest by a vote over further samples or declines them. Mathematical word problems were used because every answer can be checked exactly. On five four-bit backbones and three benchmarks, the gate raised selective accuracy from 69.5% to 85.8% with two stage-2 samples and to 88.1% with four, at 4.34 and 5.70 generations per item. Correct yield stayed at the ungated level. The MOME score ranked answers with an area under the ROC curve (AUROC) of 0.863, against 0.734 for mean token log-probability. On four backbones of further families, it reached 0.829 against 0.629, and correct yield changed by −1.81 percentage points at stage-2 temperature 0.7 and by −0.19 at 0.3. The fitted components apply to mathematical problems, but the procedure can be reused in other domains.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

LanguageAlgorithmsLarge Language ModelsReproducibility of Results

Stichwörter der Autor:innen

black-box monitoringedge inferencelabel-free confidencemathematical reasoningmetamorphic testingquantized inferenceselective predictionsmall language model

Autor:innen und Institutionen

  • Program in AI Convergence Engineering, aSSIST University, Seoul 03767, Republic of Korea.
  • Program in AI Convergence Engineering, SDG Management School, 1202 Geneva, Switzerland.

ORCID-Profile: Bongjin Jung, Joongho Chang

Zitieren

APA

Jung, B., & Chang, J. (2026). Metamorphic Evidence as a Runtime Reliability Signal for Small Language Models. Sensors, 26(21), 6918. https://doi.org/10.3390/s26216918

Vancouver

Jung B, Chang J. Metamorphic Evidence as a Runtime Reliability Signal for Small Language Models. Sensors. 2026;26(21):6918. doi:10.3390/s26216918

BibTeX
@article{Jung2026,
  author = {Jung, Bongjin and Chang, Joongho},
  title = {Metamorphic Evidence as a Runtime Reliability Signal for Small Language Models},
  journal = {Sensors},
  year = {2026},
  volume = {26},
  number = {21},
  pages = {6918},
  doi = {10.3390/s26216918}
}
Kostenlos starten →