A case study in contrastive learning information combination: Application to technical forensics of additive manufacturing filament source identification
Hagen, Alexander R., Nizinski, Cody A., Yaros, Jessica L. et al. · 2025
Read A case study in contrastive learning information combination: Application to technical forensics of additive manufacturing filament source identification by Hagen, Alexander R., Nizinski, Cody A., Yaros, Jessica L. et al. (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/doe-osti-3013601-a-case-study-in-contrastive-learning-inf.txt.
Introduction
Pacific Northwest National Laboratory's November 2025 case study applies contrastive learning to a forensic problem. The authors are Alex Hagen, Charlie Hayden, Cody Nizinski, Cuong Ly, and Jessie Yaros. The problem is identifying which filament source produced a 3D-printed part. Different measurement types describe the same object in different languages. Investigators need those views combined into one identification.
The study tests methods for combining information, drawn from machine learning. They include attention-based models. In those models, the mixing of inputs shifts with the data itself. The authors argue that this dynamic weighting fits technical forensics well. The model can focus on relevant signals and ignore irrelevant ones. One modality can also give context to another. The filament case is the worked example. It supports a general framework for multi-instrument identification problems.
Catalog record
| CREATOR | Hagen, Alexander R., Nizinski, Cody A., Yaros, Jessica L. et al. |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 22 KB |
| LICENSE | Public Domain |
| RIGHTS | PUBLIC DOMAIN |
| SOURCE | DOE OSTI |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A case study in contrastive learning information combination: Application to technical forensics of additive manufacturing filament source identification November 2025 Alex Hagen Charlie Hayden Cody Nizinski Cuong Ly Jessie Yaros Prepared for the U.S. Department of Energy under Contract DE-AC05-76RL01830 This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor Battelle Memorial Institute, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof, or Battelle Memorial Institute. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof. PACIFIC NORTHWEST NATIONAL LABORATORY operated by BATTELLE for the UNITED STATES DEPARTMENT OF ENERGY under Contract DE-AC05-76RL01830 Printed in the United States of America Available to DOE and DOE contractors from the Office of Scientific and Technical Information, P.O. Box 62, Oak Ridge, TN 37831-0062 Contents Acknowledgments ......................................................................................................................... ii Acronyms and Abbreviations ........................................................................................................iii Contents....................................................................................................................................... iv 1.0 Introduction ....................................................................................................................... 1 2.0 Method .............................................................................................................................. 2 3.0 Data ................................................................................................................................... 4 4.0 Results .............................................................................................................................. 5 5.0 References ........................................................................................................................ 7 Figures Figure 1. Example reconstructions from validation sample from diKale .... Error! Bookmark not defined. Figure 2. Receiver operating Characteristic (ROC) curve showing the proportion of times the same manufacturer scored above a varying threshold versus the proportion of the time a different manufacturer scored above that same threshold. Separate lines indicate different ROC curves for increasing number of modalities included. ............................................................ 5 Figure 3. Receiver Operating Characteristic (ROC) curve showing the proportion of times the same manufacturer scored above a varying threshold versus the proportion of the time a different manufacturer scored above that same threshold. Separate lines indicate different ROC curves for samples including the indicated modality. ........................................................................... 6 Contents iv 1.0 Introduction Combination of information from disparate data sources into a single decision is a core challenge in many fields, including the field of technical forensics. Technical forensics (TF) utilizes technical characterization of questioned samples to determine properties of that sample; these properties are then used to infer information of forensic interest, such as provenance, age, or attribution. TF is utilized in traditional forensic applications, such as the attribution of material fragments from an explosive, and in nuclear forensic applications, such as the attribution of actinides which have been interdicted out of regulatory control. The challenge of combining information from disparate sources, described alternately by many terms including “Data Fusion” and “Data Integration”, is exacerbated in the technical forensics domain due to at least two factors: the challenge of interpreting each information source singularly, and the relatively small data set sizes available. Extensive literature exists attempting to combine technical forensics information sources, both in manual and automated processes. These attempts are often bespoke to the specific information sources (such as the bi-, tri-, or quad-isotope chart (Moody, Grant, and Hutcheon 2005)), with some emerging examples of simple early- and late- fusion (, respectively). Simultaneous to the information combination efforts described in the previous paragraph, the field of natural language processing attempted (and largely succeeded) in combining information from multiple non-technical information sources. The ecosystem of “multi-modal” language models, which can take text and images as input, and generate text and images as output, became large and diverse by 2025 (Khan et al. 2025). In a generalized sense, many of these methods are trained by learning neural networks which can convert raw text or images into a vector of numbers describing the text or image, hereafter called “embeddings” and the neural networks performing the conversion are called “embedders”. By using a separate embedder for text and images, finding coincident text and images (such as images with their captions), and optimizing the parameters of the embedders such that the embeddings for the text and the image are similar, the field has found a bridge between text and images (Girdhar et al. 2023). It is the contention of the authors of this report that this insight is not limited to text and images but instead can be extended to any modality which can be found coincidently. The subject of the rest of this report is the application of this method to example multi-modal technical forensic data. Some details about the data used in this report are not appropriate for this report, and are included in a companion report (PNNL-38669). Introduction 1 2.0 Method Prior work applying neural network to TF analysis (TFA) showed the utility of unsupervised learning for representation of actinide samples, due to the dearth of labeled data suitable for supervised learning frameworks (Girard* et al. 2021). A modified technique from (Girard* et al. 2021) was developed for the data anticipated in TFA. In TFA, several types of data are anticipated: • Bulk Measurements: Scalar quantities, such as density, which represent the entire sample. • Spectra Like: One dimensional quantities, which represent the excitation of the sample against an excitation variable, such as energy or wavelength. Note that these may be measured in several different “channels” (such as in Raman spectroscopy where the excitation wavelength is changed), so they are represented with two dimensions (i.e. ["channel" ×"spectra length" ]). • Image Like: Two dimensional quantities, which represent measurements of the sample at different spatial locations projected into a single plane, usually in a raster scan. Like spectra-like, these may be measured in several different “channels” (such as red, green, and blue for visible light imagery), or they are represented with three dimensions (i.e. ["channel" ×"image height" ×"image width" ]). • Volume Like: Three dimensional quantities, which represent measurements of the sample at different spatial locations, usually in a voxel scan or tomographic reconstruction. Like image-like, these may be measured in several different “channels” (such as different energies of x-ray in x-ray computed tomography), so they are represented with four dimensions (i.e. ["channel" ×"volume depth" ×"volume height" ×"volume width" ]). Convolutional neural networks were developed to handle the types of data described above. These convolutional neural networks were trained to reconstruct their input after compression through blocks of convolutional neural network blocks, and decompression through blocks of convolutional up-sampling neural network blocks. In this way, they are trained to concentrate the information in their input into a small, single dimensional representation, called the embedding1. The spatial average at the last layer of convolutional blocks was used as the embedding for each measurement. With embeddings computed for each measurement from a sample, combination of information from each modality is then possible. The approached proposed in this report is to cast all measurements of a sample as a sequence, therefore each sample can be represented by a matrix of size [number of measurements × embedding dimension] 2. This approach aligns the 1F combination of information from each modality into sequence modeling, a topic of enormous amounts of