Research map / Time series / Causal and generative time series
Information Geometry Loss for Time Series Forecasting
Jiayu Fang, Xuande Liu, Sangsha Fang, Ernie Tian, Hongwei Ma, Zhiqi Shao, Junbin Gao
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 |
|---|---|---|---|
| add InfoGeo loss on top of MSE | iTransformer · ETTh1 avg | MSE: 0.454 -> 0.420 | Improved |
| add InfoGeo loss on top of MSE | DLinear · ETTh1 avg | MSE: 0.433 -> 0.398 | Improved |
| add InfoGeo loss on top of MSE | iTransformer · Exchange T=96 | MSE: 0.086 -> 0.088 | Hurt |
| use InfoGeo instead of Patch-wise loss | iTransformer · Weather T=96 | MSE: 0.163 -> 0.165 | Hurt |
| use InfoGeo instead of FreDF | iTransformer · ETTh1 T=96 | MSE: 0.370 -> 0.361 | Improved |
| add Fisher metric term (ablation) | InfoGeo with DLinear · ETTm1 T=96 | MSE: 0.294 -> 0.286 | Improved |
| add Bregman divergence term (ablation) | InfoGeo with DLinear · ETTm1 T=96 | MSE: 0.291 -> 0.286 | Improved |
| add natural gradient weighting (ablation) | InfoGeo with DLinear · ETTm1 T=96 | MSE: 0.289 -> 0.286 | Improved |
| add distribution parameterization (ablation) | InfoGeo with DLinear · ETTm1 T=96 | MSE: 0.297 -> 0.286 | Improved |
| add Fisher metric term (ablation) | InfoGeo with DLinear · ETTm1 T=720 | MSE: 0.418 -> 0.401 | Improved |
| train with InfoGeo, test zero-shot | iTransformer · ETTh1 to ETTm1 zero-shot | MSE: 0.367 -> 0.348 | Improved |
| loss weight lambda in 0.5-2.0 vs below 0.3 (ablation) | InfoGeo with iTransformer · ETTh1 T=96 | MSE | Mixed |
Closest papers by meaning
- Patch-wise Structural Loss for Time Series Forecasting (ICML 2025 · similarity 0.86)
- DBLoss: Decomposition-based Loss Function for Time Series Forecasting (NeurIPS 2025 · similarity 0.83)
- Loss Shaping Constraints for Long-Term Time Series Forecasting (ICML 2024 · similarity 0.83)
- Wrapped Gaussian on the manifold of Symmetric Positive Definite Matrices (ICML 2025 · similarity 0.83)
- Multivariate Time Series Forecasting needs Cross Variable Loss (NeurIPS 2026 · similarity 0.83)
- Multivariate Probabilistic Time Series Forecasting with Correlated Errors (NeurIPS 2024 · similarity 0.82)
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.
