XU-RS: Explaining Credal Width in Random-Set Language Models
Abstract
Uncertainty estimates tell us how unsure a model is, but not why. Without knowing which parts of an input influences a model's uncertainty, we cannot tell whether that uncertainty score depends on input features that are relevant for the task. We study this problem in randomset classifiers built using pretrained language models. These classifiers assign probability to individual answers and to groups of answers, producing lower and upper probabilities for each answer; The difference between these probabilities, called credal width, is used to represent epistemic uncertainty about an answer arising from limited training data. We propose XU-RS, a framework that attributes an answer's credal width to the input tokens (words or word pieces) supplied to a language model. XU-RS uses Expected Gradients (a standard feature attribution method) to estimate how input tokens contribute to credal width. The proposed framework is evaluated on a MedQA dataset using SmolLM3-3B and Llama-2-7B models, demonstrating that setting the embedding of a token ranked highly by XU-RS to zero (zero-masking) causes larger changes in credal width than zero-masking randomly selected tokens. In addition, we show that normalisation can cause other answer groups to influence an answer's width, reveal how token attribution can mask numerical errors, and provide diagnostic checks to verify whether a token ranked highly by XU-RS meaningfully explains model uncertainty.
Related
- Credal Large Language Models for Semantic Commitment under Uncertainty
- Random Width and Brightness: Polyhedral Density Theory, Reconstruction, and Gaussian Identifiability
- Credal Machine Learning for Risk-Averse Decision Making
- A Hilbert-Valued Functional Decomposition Framework for Explaining Time-Dependent Outputs
- Explaining Heterogeneous Treatment Effects with Causal Forests and the Hoeffding Functional Decomposition
BibTeX
@misc{achara2026xurs,
title = {{XU-RS: Explaining Credal Width in Random-Set Language Models}},
author = {David Achara and Maryam Sultana and Alexander D. Rast and Fabio Cuzzolin},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.37594},
eprint = {2609.37594},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.37594},
note = {CC BY-SA 4.0},
}
RIS
TY - UNPB
TI - XU-RS: Explaining Credal Width in Random-Set Language Models
AU - David Achara
AU - Maryam Sultana
AU - Alexander D. Rast
AU - Fabio Cuzzolin
PY - 2026
DA - 2026/09/29
PB - arXiv
DO - 10.48550/arxiv.2609.37594
UR - https://arxiv.org/abs/2609.37594
AB - Uncertainty estimates tell us how unsure a model is, but not why. Without knowing which parts of an input influences a model's uncertainty, we cannot tell whether that uncertainty score depends on input features that are relevant for the task. We study this problem in randomset classifiers built using pretrained language models. These classifiers assign probability to individual answers and to groups of answers, producing lower and upper probabilities for each answer; The difference between these probabilities, called credal width, is used to represent epistemic uncertainty about an answer arising from limited training data. We propose XU-RS, a framework that attributes an answer's credal width to the input tokens (words or word pieces) supplied to a language model. XU-RS uses Expected Gradients (a standard feature attribution method) to estimate how input tokens contribute to credal width. The proposed framework is evaluated on a MedQA dataset using SmolLM3-3B and Llama-2-7B models, demonstrating that setting the embedding of a token ranked highly by XU-RS to zero (zero-masking) causes larger changes in credal width than zero-masking randomly selected tokens. In addition, we show that normalisation can cause other answer groups to influence an answer's width, reveal how token attribution can mask numerical errors, and provide diagnostic checks to verify whether a token ranked highly by XU-RS meaningfully explains model uncertainty.
ER -