Research map / Time series / Foundation models for time series
MemCast: Memory-Driven Time Series Forecasting with Experience-Conditioned Reasoning
Xiaoyu Tao, Mingyue Cheng, Ze Guo, Shuo Yu, Yaguo Liu, Qi Liu, Shijin Wang
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 historical pattern memory (ablation) | MemCast · PJM | MSE: 39.259 -> 30.238 | Improved |
| add reasoning wisdom memory (ablation) | MemCast · ETTh | MSE: 8.215 -> 6.551 | Improved |
| add general law reflection (ablation) | MemCast · NP | MSE: 46.463 -> 36.871 | Improved |
| add all memory components (vs none) (ablation) | MemCast · NP | MSE: 65.275 -> 36.871 | Improved |
| add general law reflection (ablation) | MemCast · WP | MSE: w/o Law 2712.801 vs full 2,170.456; but MAE 35.937 vs 31.715 | Mixed |
| add dynamic confidence adaptation (ablation) | MemCast · NP, PJM, ETTh, ETTm | MSE / MAE | Improved |
| select trajectory by semantic consistency instead of mean ensemble or single pass (ablation) | MemCast · NP, PJM, ETTh, ETTm | MSE / MAE | Improved |
| mean ensemble of sampled paths instead of single pass (ablation) | MemCast · NP, PJM, ETTh, ETTm | MSE / MAE | Mixed |
| retrieve top-k=3 cases instead of 1, 5 or 7 (ablation) | MemCast · PJM | MSE: 37.120 (k=1), 45.221 (k=7) -> 30.238 (k=3) | Improved |
| add dynamic contextual features (ablation) | MemCast · PJM | MSE: 37.374 -> 30.238 | Improved |
| add contextual features (text) to MemCast (ablation) | MemCast · NP | MAE: 3.403 -> 3.567 | Hurt |
| swap GPT-5 backbone for other LLMs (ablation) | MemCast · PJM | MSE: 30.238 (GPT-5) vs 41.077 (Gemini 2.5 Pro), 145.995 (DeepSeek-V3.1) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Time-LLM — Time-LLM: Time Series Forecasting by Reprogramming Large Language Models (ICLR 2024) · better on 2 of 2 shared results
- TimeXer — TimeXer: Empowering Transformers for Time Series Forecasting with Exogenous Variables (NeurIPS 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting (NeurIPS 2026 · similarity 0.88)
- Time-LLM: Time Series Forecasting by Reprogramming Large Language Models (ICLR 2024 · similarity 0.88)
- ForecastCompass: Guiding Agentic Forecasting with Adaptive Factor Memory (NeurIPS 2026 · similarity 0.88)
- TS-RAG: Retrieval-Augmented Generation based Time Series Foundation Models are Stronger Zero-Shot Forecaster (NeurIPS 2025 · similarity 0.87)
- TimeXL: Explainable Multi-modal Time Series Prediction with LLM-in-the-Loop (NeurIPS 2025 · similarity 0.87)
- Learning Pattern-Specific Experts for Time Series Forecasting Under Patch-level Distribution Shift (NeurIPS 2025 · similarity 0.86)
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.
