VERBA: Verbalizing Model Differences Using Large Language Models

2025-07-02

AI Summary

In machine learning, the “model lake” phenomenon—where numerous models exhibit comparable performance yet divergent behavioral patterns—hampers systematic model comparison. Method: This paper introduces VERBA, the first framework enabling automated, fine-grained, natural-language description of behavioral differences among models. VERBA integrates output sampling, simulation-based evaluation protocols, and structured prompting to elicit high-fidelity behavioral analyses from large language models (LLMs), circumventing the O(N²) complexity of manual pairwise comparison. Contribution/Results: By incorporating structural information modeling—including decision-path alignment and error-pattern categorization—VERBA significantly improves descriptive accuracy: it achieves 90% accuracy on decision-tree model pairs, outperforming the baseline (80%). VERBA provides a scalable, end-to-end solution for enhancing model transparency, interpretability assessment, and trustworthy model selection.

Technology Category

• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success

Abstract

In the current machine learning landscape, we face a "model lake" phenomenon: Given a task, there is a proliferation of trained models with similar performances despite different behavior. For model users attempting to navigate and select from the models, documentation comparing model pairs is helpful. However, for every $N$ models there could be $O(N^2)$ pairwise comparisons, a number prohibitive for the model developers to manually perform pairwise comparisons and prepare documentations. To facilitate fine-grained pairwise comparisons among models, we introduced $ extbf{VERBA}$. Our approach leverages a large language model (LLM) to generate verbalizations of model differences by sampling from the two models. We established a protocol that evaluates the informativeness of the verbalizations via simulation. We also assembled a suite with a diverse set of commonly used machine learning models as a benchmark. For a pair of decision tree models with up to 5% performance difference but 20-25% behavioral differences, $ extbf{VERBA}$ effectively verbalizes their variations with up to 80% overall accuracy. When we included the models' structural information, the verbalization's accuracy further improved to 90%. $ extbf{VERBA}$ opens up new research avenues for improving the transparency and comparability of machine learning models in a post-hoc manner.

Problem

• Automating pairwise comparisons of ML models efficiently
• Verbalizing model differences using LLMs for transparency
• Reducing manual effort in model documentation generation

Innovation

• Leverages LLM to verbalize model differences
• Establishes protocol for evaluating verbalization informativeness
• Improves accuracy by including structural information