Research map / Time series / Long-term forecasting
Deep Switching Auto-Regressive Factorization: Application to Time Series Forecasting
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use higher-order nonlinear autoregressive dynamics with lag set ℓ={1,2,3} instead of first-order switching dynamics | DSARF temporal model · Toy Example test set | short-term NRMSE: 13.81% vs 78.45% and 98.01% | Helped |
| use higher-order nonlinear autoregressive dynamics | DSARF temporal model · Lorenz Attractor test set | short-term NRMSE: 0.88% compared to 1.14% and 2.70% | Helped |
| add lag 2 to the autoregressive lag set (ablation) | DSARF · Double Pendulum test set | short-term NRMSE: 9.79% -> 4.38% | Helped |
| increase the number of switching states from S=3 to S=10 (ablation) | DSARF · Double Pendulum test set | short-term NRMSE: 4.38% -> 4.16% | Helped |
| use DSARF's deep generative prior for spatial factors instead of a Matrix Normal prior | DSARF spatial factor model · Real-data experiments | spatial factor test-set log-likelihood: -1.37 nats -> -1.02 nats | Helped |
| add the switching feature (ablation) | DSARF · Birmingham, Bat flight, PST, Precipitation, and Google flu | short-term prediction error | Helped |
| use DSARF for short-term forecasting | DSARF · Google flu and Dengue | short-term prediction error | Mixed |
| use DSARF for short-term forecasting | DSARF · Guangzhou and PST | short-term prediction error | Hurt |
| use DSARF for long-term forecasting | DSARF · Birmingham, Hangzhou, Seattle, PST, Google flu, and Dengue | long-term prediction error | Helped |
| use DSARF for long-term forecasting | DSARF · Guangzhou | long-term prediction error | No clear effect |
Closest papers by meaning
- Deep Functional Factor Models: Forecasting High-Dimensional Functional Time Series via Bayesian Nonparametric Factorization (ICML 2024 · similarity 0.88)
- Deep Factors for Forecasting (ICML 2019 · similarity 0.88)
- HDT: Hierarchical Discrete Transformer for Multivariate Time Series Forecasting (AAAI 2025 · similarity 0.87)
- Time Series Generation Under Data Scarcity: A Unified Generative Modeling Approach (NeurIPS 2025 · similarity 0.86)
- TARFVAE: Efficient One-Step Generative Time Series Forecasting via TARFLOW based VAE (NeurIPS 2025 · similarity 0.86)
- SynTSBench: Rethinking Temporal Pattern Learning in Deep Learning Models for Time Series (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.