Advancing Text Adversarial Example Generation Using Large Language Models

N. Madrueño Sierro, A. Fernández-Isabel, R. R. Fernández, I. Martín de Diego

Text adversarial example generation is a powerful technique for identifying and analyzing the vulnerabilities present in natural language processing models. These adversarial attacks introduce subtle text perturbations that cause victim models to make incorrect predictions, while preserving the original semantic meaning from a human perspective. In this context, a novel method for generating adversarial text examples through the usage of large language models (LLMs) is presented. This approach utilizes the remarkable text generation capabilities of LLMs to modify the original text at different text levels.

Palabras clave: Adversarial Attack, Text Adversarial Example, Large Language Model, Natural Language Processing, Text Classification

Programado

Software II
10 de junio de 2025  15:30
Sala VIP Jaume Morera i Galícia

Assessing the Impact of Percentage of Missing Data and Imputation Methods on Youden Index Estimation

S. Sabroso-Lasa, L. M. Esteban Escaño, N. Malats, J. T. Alcalá Nalvaiz

CUtools: an R package for clinical utility analysis of predictive models

M. Escorihuela Sahún, L. M. Esteban Escaño, Á. Borque-Fernando, G. Sanz Sáiz