Research map / Time series / Long-term forecasting
Designing Time Series Experiments in A/B Testing with Transformer Reinforcement Learning
Xiangkun Wu, Qianglin Wen, Yingying Zhang, Hongtu Zhu, Ting Li, Chengchun Shi
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 |
|---|---|---|---|
| Replace LSTM with full-history transformer for the DDQN state encoder (ablation) | TRL · Synthetic Setting (i) | MSE: 0.002975 -> 0.002750 | Improved |
| Increase transformer attention window from 6 to 12 toward full history (ablation) | TRL · Synthetic Setting (i) | MSE: 0.003002 -> 0.002915 -> full 0.002750 | Improved |
| Use a 7-day reward warm-up instead of 2 days (ablation) | TRL · Real-data simulator (n=35, M=12, 5% lift) | MSE x 1e-4: 42.6 -> 37.9 | Improved |
| Extend warm-up from 7 to 14 days (ablation) | TRL · Real-data simulator (n=35, M=12, 5% lift) | MSE x 1e-4: 37.9 -> 39.2 | Hurt |
| Use TRL instead of best baseline design (TMDP) | Experimental design for ATE · Synthetic Setting (i) | MSE: 0.0036 -> 0.0027 | Improved |
| Use TRL instead of WSY switchback in real-data simulator | Experimental design for ATE · Real-data simulator (n=35, M=12, 5% lift) | MSE x 1e-4: 46.5 -> 37.9 | Improved |
| Misspecify the simulator used for RL training (five shift types) (ablation) | TRL · Synthetic Setting (i) | MSE: 0.0027 -> 0.0028 to 0.0030 | No clear effect |
Closest papers by meaning
- Unraveling the Interplay between Carryover Effects and Reward Autocorrelations in Switchback Experiments (ICML 2025 · similarity 0.86)
- Optimal Dynamic Regret by Transformers for Non-Stationary Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- In-Context Reinforcement Learning From Suboptimal Historical Data (ICML 2025 · similarity 0.84)
- Rethinking Decision Transformer via Hierarchical Reinforcement Learning (ICML 2024 · similarity 0.84)
- Optimizing Test-Time Compute via Meta Reinforcement Finetuning (ICML 2025 · similarity 0.84)
- ACT: Empowering Decision Transformer with Dynamic Programming via Advantage Conditioning (AAAI 2024 · similarity 0.84)
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.
