Google's TimesFM-3 AI model beats rivals in forecasting with sales data and weather
google releases timesfm-3, an ai model that forecasts future sales from time series data like daily sales figures, weather, and discount schedules, outperforming rivals on three benchmarks.
google research has released timesfm-3, an ai model that forecasts the future from time series data like daily sales figures, drawing on related data and known upcoming events. timesfm-3 is built on a transformer architecture and was trained on over one trillion data points. it handles three types of supplementary data, predicts multiple related variables, and incorporates factors known only for the past and known future events. timesfm-3 outperforms competitors, including amazon's chronos-2 and google's own timesfm-2.5, on three benchmarks: gift-eval, fev-bench, and time. it works zero-shot and needs no extra training for new tasks.