Research map / Time series / Causal and generative time series
Neural Architecture and Hyperparameter Selection Through Meta-Learning on Time Series
Erfan Moeini, Christopher Vox, Marie Anastacio, Wadie Skaf, Mitra Baratchi, Holger H. Hoos
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 |
|---|---|---|---|
| use one-hot architecture encoding instead of arch2vec (ablation) | Meta-predictor (MFE data encoding) · Meta-predictor encodings | Spearman rank correlation: 0.848 -> 0.868 | Improved |
| use one-hot architecture encoding instead of AdjOp+GCN (ablation) | Meta-predictor (MFE data encoding) · Meta-predictor encodings | Spearman rank correlation: 0.841 -> 0.868 | Improved |
| use meta-feature dataset encoding instead of VAE (ablation) | Meta-predictor (OHE architecture encoding) · Meta-predictor encodings | Spearman rank correlation: 0.862 -> 0.868 | Improved |
| add transformer on top of one-hot architecture encoding (ablation) | Meta-predictor (MFE data encoding) · Meta-predictor encodings | Spearman rank correlation: 0.868 -> 0.868 | No clear effect |
| train with combined ranking loss (MSE + margin ranking) instead of single loss (ablation) | Meta-predictor · Meta-predictor loss comparison | Spearman rank correlation and MSE | Mixed |
| replace HEBO tuning with meta-predictor recommendation | Meta-predictor · Forecasting datasets (8) | MSE: 0.410 -> 0.268 | Mixed |
| replace HEBO tuning with meta-predictor recommendation | Meta-predictor · UEA classification (28 datasets) | accuracy: 44.2 -> 66.2 | Mixed |
| replace HEBO tuning with meta-predictor recommendation | Meta-predictor · UCR classification (45 datasets) | accuracy | Mixed |
| replace HEBO tuning with meta-predictor recommendation | Meta-predictor · Forecasting and classification | statistical significance (Wilcoxon, 0.05) | No clear effect |
| replace HEBO tuning with meta-predictor recommendation | Meta-predictor · Forecasting and classification | computation time: 5% and 10% of HEBO time | Improved |
| replace HEBO tuning with meta-predictor recommendation | Meta-predictor · Classification datasets | memory | Hurt |
| replace HEBO tuning with meta-predictor recommendation | Meta-predictor · Forecasting datasets | memory: 78% | Improved |
Closest papers by meaning
- Meta-Learning Framework with Applications to Zero-Shot Time-Series Forecasting (AAAI 2021 · similarity 0.89)
- Learning Deep Time-index Models for Time Series Forecasting (ICML 2023 · similarity 0.87)
- SwiftTS: A Swift Selection Framework for Time Series Pre-trained Models via Multi-task Meta-Learning (ICLR 2026 · similarity 0.87)
- Large Pre-trained time series models for cross-domain Time series analysis tasks (NeurIPS 2024 · similarity 0.85)
- Cherry-Picking in Time Series Forecasting: How to Select Datasets to Make Your Model Shine (AAAI 2025 · similarity 0.85)
- Adaptive Meta-Learning Probabilistic Inference Framework for Long Sequence Prediction (AAAI 2024 · similarity 0.85)
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.
