Winner-takes-all for Multivariate Probabilistic Time Series Forecasting
This addresses the need for diverse probabilistic forecasts in time series applications, though it appears incremental as an adaptation of an existing paradigm.
The paper tackles the problem of forecasting multiple plausible futures in multivariate time series by introducing TimeMCL, a method based on Multiple Choice Learning with a Winner-Takes-All loss to promote prediction diversity. It demonstrates promising performance on real-world data with light computational cost.
We introduce TimeMCL, a method leveraging the Multiple Choice Learning (MCL) paradigm to forecast multiple plausible time series futures. Our approach employs a neural network with multiple heads and utilizes the Winner-Takes-All (WTA) loss to promote diversity among predictions. MCL has recently gained attention due to its simplicity and ability to address ill-posed and ambiguous tasks. We propose an adaptation of this framework for time-series forecasting, presenting it as an efficient method to predict diverse futures, which we relate to its implicit quantization objective. We provide insights into our approach using synthetic data and evaluate it on real-world time series, demonstrating its promising performance at a light computational cost.