Research map / Time series / Long-term forecasting
Multi-Horizon Time Series Forecasting of Non-Parametric CDFs with Deep Lattice Networks
Niklas Erdmann, Lars Bentsen, Roy Stenbro, Heine Nygard Riise, Narada Dilp Warakagoda, Paal E. Engelstad
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 |
|---|---|---|---|
| constrain quantile monotonicity with a DLN instead of an unconstrained NN output (ablation) | SQR LSTM forecaster · Solar irradiance day-ahead | MAE / RMSE: MAE 42.612 -> 43.266; RMSE 72.930 -> 74.520 | No clear effect |
| use DLN instead of unconstrained NN output (ablation) | SQR LSTM forecaster · Solar irradiance day-ahead | ACE: 0.075 -> 0.061 | Improved |
| use DLN instead of unconstrained NN output (ablation) | SQR LSTM forecaster · Solar irradiance day-ahead | CRPS: 27.690 -> 29.175 | Hurt |
| use DLN instead of SMNN as monotonic output model (ablation) | SQR LSTM forecaster · Solar irradiance day-ahead | MAE: 47.548 -> 43.266 | Improved |
| use DLN instead of SMNN (ablation) | SQR LSTM forecaster · Solar irradiance day-ahead | CRPS / ACE: CRPS 29.333 -> 29.175 | No clear effect |
| use non-linear DLN instead of linear output layer (ablation) | SQR LSTM forecaster · Solar irradiance day-ahead | ACE: 0.173 -> 0.061 | Improved |
| use non-linear DLN instead of linear output layer (ablation) | SQR LSTM forecaster · Solar irradiance day-ahead | CRPS: 24.427 -> 29.175 | Hurt |
| use constrained linear output instead of linear (ablation) | SQR LSTM forecaster · Solar irradiance day-ahead | ACE: 0.173 -> 0.190 | No clear effect |
| SQR with DLN instead of QR with 11 fixed quantiles | LSTM forecaster · Solar irradiance day-ahead | ACE: 0.052 -> 0.061; CRPS 30.506 -> 29.175 | Mixed |
| SQR probabilistic training instead of point prediction | LSTM forecaster · Solar irradiance day-ahead | RMSE: 74.933 -> 72.930 (LSTM-NN) | Improved |
| use LSTM-DLN instead of smart persistence | LSTM-DLN · Solar irradiance day-ahead | MAE: 52.903 -> 43.266 | Improved |
| constrain with DLN to prevent quantile crossover | SQR LSTM forecaster · Solar irradiance day-ahead | quantile crossover: 3.075 % -> 0 (guaranteed monotonic) | Improved |
Closest papers by meaning
- RNN with Particle Flow for Probabilistic Spatio-temporal Forecasting (ICML 2021 · similarity 0.84)
- ProbTS: Benchmarking Point and Distributional Forecasting across Diverse Prediction Horizons (NeurIPS 2024 · similarity 0.84)
- Less is More: Unlocking Specialization of Time Series Foundation Models via Structured Pruning (NeurIPS 2025 · similarity 0.84)
- Relational Conformal Prediction for Correlated Time Series (ICML 2025 · similarity 0.84)
- TimePFN: Effective Multivariate Time Series Forecasting with Synthetic Data (AAAI 2025 · similarity 0.83)
- TStruct: Learning Shared Temporal Structures for Long-Term Time-Series Forecasting (NeurIPS 2026 · 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.
