Researchers have explored methods for detecting time series data generated by diffusion models, particularly when the specific generator is unknown. A white-box detection approach, which requires access to the generator, proved effective in in-distribution scenarios but struggled with generator shift. In contrast, a simple black-box classifier performed significantly better, achieving an average F1 score of 79.2, outperforming the white-box method by 22.1%. This suggests that diffusion-generated time series detection differs from its image domain counterpart. AI
RANK_REASON This is a research paper published on arXiv detailing a new method for detecting AI-generated time series data. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
