Attack LLMs in TS

To expose the vulnerabilities of LLMs in time series forecasting by three distinct black-box adversarial attacks.

(To be continued)

This project aims to fully understand the vulnerabilities of LLMs in time series forecasting and is composed of three papers:

Adversarial Vulnerabilities in Large Language Models for Time Series Forecasting AISTATS 2025.
Temporally Sparse Attack for Fooling Large Language Models in Time Series Forecasting Building Trust in Language Models and Applications Workshop at ICLR 2025.

1. Black-box attack against LLMs in time series forecasting.

Attack LLMs

We manipulate LLM-based time series forecasting.