Frictionless knowledge injection for few-shot learning

Jordan Stomps; Kenneth Dayman; Christian Randolph; Jason Hite

doi:10.1007/s10967-025-10548-7

Frictionless knowledge injection for few-shot learning

Research output: Contribution to journal › Article › peer-review

Abstract

Cutting-edge machine learning methods often require large volumes of curated training data, precluding their use in national security problems with rare events in massive datasets. We present a method for incorporating abstract knowledge into models tailored for sparse data. A subject matter expert defines salient concepts using data examples, which are encoded in the model’s embedding space. Models are then trained to respect these concepts. This method enables knowledge injection, yielding effective models with limited labeled data and the ability to assess model sensitivity for subject matter expertise across the nonproliferation mission space, as demonstrated with Raman spectra analysis.

Original language	English
Pages (from-to)	9073-9083
Number of pages	11
Journal	Journal of Radioanalytical and Nuclear Chemistry
Volume	334
Issue number	12
DOIs	https://doi.org/10.1007/s10967-025-10548-7
State	Published - Dec 2025

Funding

This research was funded by the Laboratory Directed Research and Development Program of Oak Ridge National Laboratory, managed by UT-Battelle, LLC, for the US Department of Energy. A concrete example of these challenges is the Multi-Informatics for Nuclear Operations Scenarios venture [] funded by the Office of Defense Nuclear Nonproliferation Research and Development within the US Department of Energy’s National Nuclear Security Administration. During the project’s approximately 5-year life cycle, the venture team instrumented the High Flux Isotope Reactor, the Radiochemical Engineering Development Center, and the surrounding area with multiple sensor networks spanning multiple measurement modalities. Despite the substantial budget and time invested in collecting data, only a handful of plutonium and californium processing campaigns were performed and observed. The scarcity of these events hindered motivated targeted research in developing effective analytical models for sparse datasets []. This manuscript has been authored by UT-Battelle LLC under contract DE-AC05-00OR22725 with the US Department of Energy (DOE). The US government retains and the publisher, by accepting the article for publication, acknowledges that the US government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for US government purposes. DOE will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan ( https://www.energy.gov/doe-public-access-plan ).

Keywords

Concept activation vectors
Few-shot learning
Nuclear nonproliferation
Raman spectra
Sparse data

Access to Document

10.1007/s10967-025-10548-7

Cite this

@article{1f97041018ee4494af24cf3e306bfb13,

title = "Frictionless knowledge injection for few-shot learning",

abstract = "Cutting-edge machine learning methods often require large volumes of curated training data, precluding their use in national security problems with rare events in massive datasets. We present a method for incorporating abstract knowledge into models tailored for sparse data. A subject matter expert defines salient concepts using data examples, which are encoded in the model{\textquoteright}s embedding space. Models are then trained to respect these concepts. This method enables knowledge injection, yielding effective models with limited labeled data and the ability to assess model sensitivity for subject matter expertise across the nonproliferation mission space, as demonstrated with Raman spectra analysis.",

keywords = "Concept activation vectors, Few-shot learning, Nuclear nonproliferation, Raman spectra, Sparse data",

author = "Jordan Stomps and Kenneth Dayman and Christian Randolph and Jason Hite",

note = "Publisher Copyright: {\textcopyright} UT-Battelle, LLC 2025.",

year = "2025",

month = dec,

doi = "10.1007/s10967-025-10548-7",

language = "English",

volume = "334",

pages = "9073--9083",

journal = "Journal of Radioanalytical and Nuclear Chemistry",

issn = "0236-5731",

publisher = "Akad{\'e}miai Kiad{\'o}",

number = "12",

}

TY - JOUR

T1 - Frictionless knowledge injection for few-shot learning

AU - Stomps, Jordan

AU - Dayman, Kenneth

AU - Randolph, Christian

AU - Hite, Jason

N1 - Publisher Copyright: © UT-Battelle, LLC 2025.

PY - 2025/12

Y1 - 2025/12

N2 - Cutting-edge machine learning methods often require large volumes of curated training data, precluding their use in national security problems with rare events in massive datasets. We present a method for incorporating abstract knowledge into models tailored for sparse data. A subject matter expert defines salient concepts using data examples, which are encoded in the model’s embedding space. Models are then trained to respect these concepts. This method enables knowledge injection, yielding effective models with limited labeled data and the ability to assess model sensitivity for subject matter expertise across the nonproliferation mission space, as demonstrated with Raman spectra analysis.

AB - Cutting-edge machine learning methods often require large volumes of curated training data, precluding their use in national security problems with rare events in massive datasets. We present a method for incorporating abstract knowledge into models tailored for sparse data. A subject matter expert defines salient concepts using data examples, which are encoded in the model’s embedding space. Models are then trained to respect these concepts. This method enables knowledge injection, yielding effective models with limited labeled data and the ability to assess model sensitivity for subject matter expertise across the nonproliferation mission space, as demonstrated with Raman spectra analysis.

KW - Concept activation vectors

KW - Few-shot learning

KW - Nuclear nonproliferation

KW - Raman spectra

KW - Sparse data

UR - https://www.scopus.com/pages/publications/105025373970

U2 - 10.1007/s10967-025-10548-7

DO - 10.1007/s10967-025-10548-7

M3 - Article

AN - SCOPUS:105025373970

SN - 0236-5731

VL - 334

SP - 9073

EP - 9083

JO - Journal of Radioanalytical and Nuclear Chemistry

JF - Journal of Radioanalytical and Nuclear Chemistry

IS - 12

ER -

Frictionless knowledge injection for few-shot learning

Abstract

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this