Deep Factors for Forecasting

Abstract

Producing probabilistic forecasts for large collections of similar and/or dependent time series is a practically relevant and challenging task. Classical time series models fail to capture complex patterns in the data, and multivariate techniques struggle to scale to large problem sizes. Their reliance on strong structural assumptions makes them data-efficient, and allows them to provide uncertainty estimates. The converse is true for models based on deep neural networks, which can learn complex patterns and dependencies given enough data. In this paper, we propose a hybrid model that incorporates the benefits of both approaches. Our new method is data-driven and scalable via a latent, global, deep component. It also handles uncertainty through a local classical model. We provide both theoretical and empirical evidence for the soundness of our approach through a necessary and sufficient decomposition of exchangeable time series into a global and a local part. Our experiments demonstrate the advantages of our model both in term of data efficiency, accuracy and computational complexity.

http://proceedings.mlr.press/v97/wang19k/wang19k.pdf. arXiv admin note: substantial text overlap with arXiv:1812.00098

References in corpus (9)

• Auto-Encoding Variational Bayes
• MXNet: A Flexible and Efficient Machine Learning Library for Heterogeneous Distributed Systems
• A Recurrent Latent Variable Model for Sequential Data
• A Multi-Horizon Quantile Recurrent Forecaster
• State Space LSTM Models with Particle MCMC Inference
• ARMDN: Associative and Recurrent Mixture Density Networks for eRetail Demand Forecasting
• Deep Factors with Gaussian Processes for Forecasting
• Auto-Differentiating Linear Algebra
• A Unified Framework for Long Range and Cold Start Forecasting of Seasonal Profiles in Time Series

Cited by in corpus (21)

• Time Series Forecasting With Deep Learning: A Survey
• Recurrent Neural Networks for Time Series Forecasting: Current Status and Future Directions
• Deep Learning for Time Series Forecasting: Tutorial and Literature Survey
• Bayesian Temporal Factorization for Multidimensional Time Series Prediction
• N-BEATS: Neural basis expansion analysis for interpretable time series forecasting
• Think Globally, Act Locally: A Deep Neural Network Approach to High-Dimensional Time Series Forecasting
• Real-time forecasting of metro origin-destination matrices with high-order weighted dynamic mode decomposition
• Time Series Analysis and Modeling to Forecast: a Survey
• Forecasting: theory and practice
• Domain Adaptation for Time Series Forecasting via Attention Sharing
• Topological Attention for Time Series Forecasting
• Sequential Estimation of Gaussian Process-based Deep State-Space Models
• Learning Quantile Functions without Quantile Crossing for Distribution-free Time Series Forecasting
• A Multi-view Multi-task Learning Framework for Multi-variate Time Series Forecasting
• Intermittent Demand Forecasting with Deep Renewal Processes
• A Worrying Analysis of Probabilistic Time-series Models for Sales Forecasting
• Synergetic Learning of Heterogeneous Temporal Sequences for Multi-Horizon Probabilistic Forecasting
• Neural forecasting at scale
• Graph Deep Factors for Forecasting
• Relation-aware Meta-learning for Market Segment Demand Prediction with Limited Records
• Meta-Forecasting by combining Global Deep Representations with Local Adaptation