Winner-takes-all for Multivariate Probabilistic Time Series Forecasting

Adrien Cortes, Remi Rehm, Victor Letzelter

TL;DR: We introduce TimeMCL, leveraging the Multiple Choice Learning (MCL) paradigm to forecast multiple plausible time series futures.

Abstract: We introduce $\texttt{TimeMCL}$, 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.

Lay Summary:

When we try to predict what might happen in the future based on past data, we often find that there isn’t just one “right” answer — there could be several possible future scenarios. In this work, we introduce TimeMCL, a method that helps machines to predict multiple plausible futures for a time series, such as weather data or stock prices.

TimeMCL builds on a technique called Multiple Choice Learning (MCL), which trains a computer model to generate a diverse set of predictions rather than focusing on a single outcome. To make sure these predictions are truly different and not just minor variations, we use a special “Winner-Takes-All” approach that updates only the best-performing prediction for each example.

We tested this idea with both simulated and real-world data and found that TimeMCL can provide accurate and varied predictions without needing a lot of computing power.

Link To Code: https://github.com/Victorletzelter/timeMCL

Primary Area: Deep Learning->Sequential Models, Time series

Keywords: Time-series forecasting, Quantization, Probabilistic methods, Conditional Distribution Estimation, Winner-takes-all, Diversity, Multiple Choice Learning

Submission Number: 12563