Artificial Interrogation for Attributing Language Models

TL;DR

This paper addresses the challenge of attributing fine-tuned language models to their base models by developing interrogation strategies and multiple response similarity measures, achieving effective model attribution under restricted API access.

Contribution

The paper introduces four novel approaches for model attribution using response similarity metrics, advancing the methodology for identifying model origins in a restricted API setting.

Findings

Response similarity metrics effectively distinguish models

Transformer-based classifiers improve attribution accuracy

Multiple approaches outperform baseline methods

Abstract

This paper presents solutions to the Machine Learning Model Attribution challenge (MLMAC) collectively organized by MITRE, Microsoft, Schmidt-Futures, Robust-Intelligence, Lincoln-Network, and Huggingface community. The challenge provides twelve open-sourced base versions of popular language models developed by well-known organizations and twelve fine-tuned language models for text generation. The names and architecture details of fine-tuned models were kept hidden, and participants can access these models only through the rest APIs developed by the organizers. Given these constraints, the goal of the contest is to identify which fine-tuned models originated from which base model. To solve this challenge, we have assumed that fine-tuned models and their corresponding base versions must share a similar vocabulary set with a matching syntactical writing style that resonates in their generated outputs. Our strategy is to develop a set of queries to interrogate base and fine-tuned models. And then perform one-to-many pairing between them based on similarities in their generated responses, where more than one fine-tuned model can pair with a base model but not vice-versa. We have employed four distinct approaches for measuring the resemblance between the responses generated from the models of both sets. The first approach uses evaluation metrics of the machine translation, and the second uses a vector space model. The third approach uses state-of-the-art multi-class text classification, Transformer models. Lastly, the fourth approach uses a set of Transformer based binary text classifiers, one for each provided base model, to perform multi-class text classification in a one-vs-all fashion. This paper reports implementation details, comparison, and experimental studies, of these approaches along with the final obtained results.