arXiv AI recent: TimeVista: Exploring and Exploiting Vision-Language Models as Judges for Time Series Forecasting
The authors propose using Vision-Language Models (VLMs) as judges for evaluating time series forecasting.,They introduce a benchmark called TimeVista, which contains 5,563 time series sam...
The paper argues that traditional point‑wise metrics often fail to capture complex temporal patterns and do not align well with human intuitive preferences.,It describes a novel framework that integrates micro‑ and macro‑level judgments informed by contextual information to assess time series for...