Hallucination Detection and Mitigation with Diffusion in Multi-Variate Time-Series Foundation Models

TMLR Paper8075 Authors

Abstract: Foundation models (FMs) for natural language processing have many coherent definitions of hallucination and methods for its detection and mitigation. However, analogous definitions and methods do not exist for multi-variate time-series (MVTS) FMs. We propose new definitions for MVTS hallucination, along with new detection and mitigation methods using a diffusion model to estimate hallucination levels. We derive relational datasets from popular time-series datasets to benchmark these relational hallucination levels. Using these definitions and models, we find that open-source pre-trained MVTS imputation FMs relationally hallucinate on average up to 59.5\% as much as a weak baseline. The proposed mitigation method reduces this by up to 47.7\% for these models. The definition and methods may improve adoption and safe usage of MVTS FMs.

Submission Type: Regular submission (no more than 12 pages of main content)

Assigned Action Editor: ~Zhen_Fang2

Submission Number: 8075