Research map / Time series / Long-term forecasting
Adversarial Sparse Transformer for 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 |
|---|---|---|---|
| add adversarial sequence-level training (ablation) | Sparse Transformer · traffic1d | Q50 loss: 0.109 -> 0.093 | Helped |
| add adversarial sequence-level training (ablation) | Sparse Transformer · elect1d | Q50 loss: 0.055 -> 0.042 | Helped |
| use alpha-entmax sparse attention instead of softmax (ablation) | vanilla Transformer · elect1d | Q50 loss: 0.064 -> 0.055 | Helped |
| use alpha-entmax sparse attention instead of softmax (ablation) | vanilla Transformer · elect7d | Q90 loss: 0.039 -> 0.041 | Hurt |
| use 1.5-entmax instead of softmax (ablation) | AST · traffic1d | Q50 loss: 0.120 -> 0.093 | Helped |
| use 1.5-entmax instead of softmax (ablation) | AST · elect1d | Q50 loss: 0.040 -> 0.042 | Hurt |
| use 1.5-entmax instead of sparsemax (ablation) | AST · traffic1d | Q50 loss: 0.123 -> 0.093 | Helped |
| add adversarial training | DeepAR · traffic1d | Q50 loss: 0.161 -> 0.155 | Helped |
| encoder-decoder instead of decoder-only autoregressive (ablation) | AST · traffic1d | Q50 loss: 0.096 -> 0.093 | Helped |
| encoder-decoder instead of decoder-only autoregressive (ablation) | AST · elect1d | Q50 loss: 0.048 -> 0.042 | Helped |
| AST vs ConvTrans | time series forecasting · electricity and traffic | Q50 loss: 26.4% on average | Helped |
Closest papers by meaning
- TEMPO: Prompt-based Generative Pre-trained Transformer for Time Series Forecasting (ICLR 2024 · similarity 0.86)
- Timer: Generative Pre-trained Transformers Are Large Time Series Models (ICML 2024 · similarity 0.85)
- SAMformer: Unlocking the Potential of Transformers in Time Series Forecasting with Sharpness-Aware Minimization and Channel-Wise Attention (ICML 2024 · similarity 0.85)
- A Closer Look at Transformers for Time Series Forecasting: Understanding Why They Work and Where They Struggle (ICML 2025 · similarity 0.85)
- Time-MoE: Billion-Scale Time Series Foundation Models with Mixture of Experts (ICLR 2025 · similarity 0.85)
- Loss Shaping Constraints for Long-Term Time Series Forecasting (ICML 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.