New method audits time series foundation models for data contamination

Researchers have introduced TSFMAudit, a novel method designed to detect data contamination in time series foundation models (TSFMs). This is the first study to address pretraining contamination auditing specifically for TSFMs, which are increasingly trained on vast datasets. TSFMAudit operates by analyzing probe adaptation dynamics, identifying contamination through unusually rapid loss reduction and minimal backbone movement during fine-tuning probes. The method was evaluated on six TSFMs and 187 datasets, outperforming ten existing baselines adapted from large language model research. AI

RANK_REASON The cluster contains an academic paper introducing a new method for auditing AI models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• time series foundation models
• TSFMAudit

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →