NHITS: Neural Hierarchical Interpolation for Time Series Forecasting

• DOI: 10.1609/aaai.v37i6.25854
• Publication Date: 2023-06-27T17:03:36Z
• Authors: Cristian Challu; Kin G. Olivares; Boris N. Oreshkin; Federico Garza Ra...; Max Mergenthaler ...; Artur Dubrawski
• Keywords: Interpolation

Recent progress in neural forecasting accelerated improvements the performance of large-scale systems. Yet, long-horizon remains a very difficult task. Two common challenges afflicting task are volatility predictions and their computational complexity. We introduce NHITS, model which addresses both by incorporating novel hierarchical interpolation multi-rate data sampling techniques. These techniques enable proposed method to assemble its sequentially, emphasizing components with different frequencies scales while decomposing input signal synthesizing forecast. prove that technique can efficiently approximate arbitrarily long horizons presence smoothness. Additionally, we conduct extensive dataset experiments from literature, demonstrating advantages our over state-of-the-art methods, where NHITS provides an average accuracy improvement almost 20% latest Transformer architectures reducing computation time order magnitude (50 times). Our code is available at https://github.com/Nixtla/neuralforecast.