Research map / Time series / Causal and generative time series
Adjusting for Autocorrelated Errors in Neural Networks for Time Series
Fan-Keng Sun, Chris Lang, Duane S Boning
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| learn autocorrelation coefficient rho jointly with model parameters (adjust for autocorrelated errors) | LSTM · Traffic | RRMSE: .4936 -> .3643 | Improved |
| add autocorrelation adjustment | TCN · 17 forecasting datasets | avg relative improvement (%): 25.3% | Improved |
| add autocorrelation adjustment | AGCRN · Traffic | RRMSE: .3326 -> .3339 | Hurt |
| add autocorrelation adjustment | DSANet · M5-L9 | RRMSE: .1823 -> .1824 | No clear effect |
| add autocorrelation adjustment with grid-searched hyperparameters | AGCRN, DSANet · M4-Hourly, Traffic, M5-L9, Electricity | RRMSE: .0275 -> .0267 (M4-Hourly) | Mixed |
| adjust only the input (rho on input) (ablation) | DSANet · 6 datasets | avg relative improvement (%): 3.52% | Improved |
| adjust only the output (ablation) | DSANet · 6 datasets | avg relative improvement (%): -1.21% | Hurt |
| adjust both input and output instead of only one (ablation) | TPA · 6 datasets | avg relative improvement (%): 10.4% -> 14.4% | Improved |
| learn rho jointly instead of fixing it by grid search (ablation) | adjustment method · four model-dataset combinations | RRMSE | Improved |
| learn rho vs fixing rho=1 (differencing) (ablation) | adjustment method · four model-dataset combinations | RRMSE | Improved |
| increase LSTM hidden units (less model misspecification) (ablation) | LSTM adjustment · Traffic | relative improvement of adjustment | Hurt |
Closest papers by meaning
- Multivariate Probabilistic Time Series Forecasting with Correlated Errors (NeurIPS 2024 · similarity 0.87)
- FreDF: Learning to Forecast in the Frequency Domain (ICLR 2025 · similarity 0.85)
- AutoTimes: Autoregressive Time Series Forecasters via Large Language Models (NeurIPS 2024 · similarity 0.83)
- Temporal Query Network for Efficient Multivariate Time Series Forecasting (ICML 2025 · similarity 0.83)
- ARM: Refining Multivariate Forecasting with Adaptive Temporal-Contextual Learning (ICLR 2024 · similarity 0.83)
- Time-o1: Time-Series Forecasting Needs Transformed Label Alignment (NeurIPS 2025 · similarity 0.83)
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.
