Statistical inference on black-box generative models in the data kernel perspective space

TL;DR

This paper develops statistical inference methods for black-box generative models using data kernel perspective, enabling analysis without access to model internals, and demonstrates their effectiveness across various tasks.

Contribution

It extends recent representation techniques to model-level inference, providing new tools for analyzing black-box generative models without internal access.

Findings

Model-level representations are effective for multiple inference tasks.

The methods work without access to model pre-training data or weights.

The approach broadens the ability to analyze generative models in practical scenarios.

Abstract

Generative models are capable of producing human-expert level content across a variety of topics and domains. As the impact of generative models grows, it is necessary to develop statistical methods to understand collections of available models. These methods are particularly important in settings where the user may not have access to information related to a model's pre-training data, weights, or other relevant model-level covariates. In this paper we extend recent results on representations of black-box generative models to model-level statistical inference tasks. We demonstrate that the model-level representations are effective for multiple inference tasks.