TimeVista: Exploring and Exploiting Vision-Language Models as Judges for Time Series Forecasting
Researchers have introduced TimeVista, a new framework that utilizes Vision-Language Models (VLMs) to evaluate time series forecasting. This approach leverages VLMs' ability to interpret time series plots alongside textual information, offering a more human-aligned judgment than traditional point-wise metrics. The TimeVista benchmark includes 5563 time series samples and has demonstrated that VLMs provide reliable and consistent evaluations, outperforming conventional methods in aligning with human preferences. This framework has been used to assess recent Time Series Foundation Models, revealing VLMs as robust and interpretable judges for this domain. AI
IMPACT VLMs offer a more human-aligned evaluation standard for time series forecasting, potentially improving model development and comparison.