A new model called Informer has been proposed for multivariate probabilistic time series forecasting. It uses a self-attentive mechanism to capture long-term dependencies and a probabilistic framework to output a distribution of possible future values. The authors claim that Informer outperforms existing methods on several benchmark datasets. --- Why it matters: This matters because accurate time series forecasting is crucial in many applications, including demand prediction for businesses and weather forecasting. Informer's ability to capture long-term dependencies and provide probabilistic forecasts could lead to more reliable predictions. Source: https://huggingface.co/blog/informer