Compressed Models are NOT Trust-equivalent to Their Large Counterparts

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Rai, Rohit Raj, Kothari, Chirag, Shelke, Siddhesh, Awekar, Amit
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912543821266944
author Rai, Rohit Raj
Kothari, Chirag
Shelke, Siddhesh
Awekar, Amit
author_facet Rai, Rohit Raj
Kothari, Chirag
Shelke, Siddhesh
Awekar, Amit
contents Large Deep Learning models are often compressed before being deployed in a resource-constrained environment. Can we trust the prediction of compressed models just as we trust the prediction of the original large model? Existing work has keenly studied the effect of compression on accuracy and related performance measures. However, performance parity does not guarantee trust-equivalence. We propose a two-dimensional framework for trust-equivalence evaluation. First, interpretability alignment measures whether the models base their predictions on the same input features. We use LIME and SHAP tests to measure the interpretability alignment. Second, calibration similarity measures whether the models exhibit comparable reliability in their predicted probabilities. It is assessed via ECE, MCE, Brier Score, and reliability diagrams. We conducted experiments using BERT-base as the large model and its multiple compressed variants. We focused on two text classification tasks: natural language inference and paraphrase identification. Our results reveal low interpretability alignment and significant mismatch in calibration similarity. It happens even when the accuracies are nearly identical between models. These findings show that compressed models are not trust-equivalent to their large counterparts. Deploying compressed models as a drop-in replacement for large models requires careful assessment, going beyond performance parity.
format Preprint
id arxiv_https___arxiv_org_abs_2508_13533
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Compressed Models are NOT Trust-equivalent to Their Large Counterparts
Rai, Rohit Raj
Kothari, Chirag
Shelke, Siddhesh
Awekar, Amit
Computation and Language
Machine Learning
Large Deep Learning models are often compressed before being deployed in a resource-constrained environment. Can we trust the prediction of compressed models just as we trust the prediction of the original large model? Existing work has keenly studied the effect of compression on accuracy and related performance measures. However, performance parity does not guarantee trust-equivalence. We propose a two-dimensional framework for trust-equivalence evaluation. First, interpretability alignment measures whether the models base their predictions on the same input features. We use LIME and SHAP tests to measure the interpretability alignment. Second, calibration similarity measures whether the models exhibit comparable reliability in their predicted probabilities. It is assessed via ECE, MCE, Brier Score, and reliability diagrams. We conducted experiments using BERT-base as the large model and its multiple compressed variants. We focused on two text classification tasks: natural language inference and paraphrase identification. Our results reveal low interpretability alignment and significant mismatch in calibration similarity. It happens even when the accuracies are nearly identical between models. These findings show that compressed models are not trust-equivalent to their large counterparts. Deploying compressed models as a drop-in replacement for large models requires careful assessment, going beyond performance parity.
title Compressed Models are NOT Trust-equivalent to Their Large Counterparts
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2508.13533