The article introduces ForecastPFN, a novel zero-shot forecasting model trained on synthetic data distribution. This approach is designed to overcome the limitations of many forecasting methods that require substantial initial data, which is often not available in real-world applications. The ForecastPFN model uses a prior-data fitted network to approximate Bayesian inference and can make predictions on a new time series dataset in a single pass. The model’s performance has been found to be more accurate and faster than other state-of-the-art forecasting methods, even when those methods are allowed to train on hundreds of additional in-distribution data points.
Publication date: 3 Nov 2023
Project Page: https://arxiv.org/abs/2311.01933v1
Paper: https://arxiv.org/pdf/2311.01933
