Model editing for distribution shifts in uranium oxide morphological analysis

TL;DR

This paper demonstrates that model editing techniques significantly improve the generalization of deep learning models for classifying uranium ore concentrate micrographs under distribution shifts caused by different measurement instruments and conditions.

Contribution

It introduces the application of model editing to scientific data classification, outperforming finetuning for handling distribution shifts in uranium oxide morphological analysis.

Findings

Model editing outperforms finetuning on curated datasets.

Significant improvement in generalization to different measurement instruments.

Effective handling of distribution shifts in scientific micrograph data.

Abstract

Deep learning still struggles with certain kinds of scientific data. Notably, pretraining data may not provide coverage of relevant distribution shifts (e.g., shifts induced via the use of different measurement instruments). We consider deep learning models trained to classify the synthesis conditions of uranium ore concentrates (UOCs) and show that model editing is particularly effective for improving generalization to distribution shifts common in this domain. In particular, model editing outperforms finetuning on two curated datasets comprising of micrographs taken of U$_{3}$O$_{8}$ aged in humidity chambers and micrographs acquired with different scanning electron microscopes, respectively.