Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Selective Learning for Deep Time Series Forecasting
Authors: Yisong Fu, Zezhi Shao, Chengqing Yu, Yujie Li, Zhulin An, Qi (Cheems) Wang, Yongjun Xu, Fei Wang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments across eight real-world datasets demonstrate that selective learning can significantly improve the predictive performance for typical state-of-the-art deep models, including 37.4% MSE reduction for Informer, 8.4% for Times Net, and 6.5% for i Transformer. |
| Researcher Affiliation | Academia | 1State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences 2University of Chinese Academy of Sciences 3Department of Automation, Tsinghua University EMAIL, EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1 The workflow of selective learning. |
| Open Source Code | Yes | Code: https://github.com/Gestalt Cog Team/selective-learning https://github.com/Gestalt Cog Team/Basic TS |
| Open Datasets | Yes | We thoroughly evaluate the effectiveness of the proposed selective learning on 8 realworld datasets, including Electricity, Exchange, Weather, ILI, and 4 ETT datasets (ETTh1, ETTh2, ETTm1, ETTm2), which have been extensively used for benchmarking [42, 64, 37]. |
| Dataset Splits | Yes | We follow the same data processing and train-validation-test set split protocol used in Times Net[63], where the train, validation, and test datasets are strictly divided according to chronological order to ensure no data leakage issues. The statistics of the datasets are provided in Table 5. Table 5: Statistics of the datasets. ... Split 6:2:2 ... Split 7:1:2 |
| Hardware Specification | Yes | All experiments are implemented with Py Torch and conducted on 8 NVIDIA Ge Force RTX 4090 24GB GPUs. |
| Software Dependencies | No | The paper mentions 'Py Torch' but does not specify a version number or any other software dependencies with version numbers. |
| Experiment Setup | Yes | All baselines follow the same experimental setup with prediction lengths F {24, 36, 48, 60} for ILI and F {96, 192, 336, 720} for others [63]. We search for the lookback window L and report the best results. For fair evaluation, when training baselines with selective learning to enhance their performance, we follow their original hyperparameter settings and only tune the masking ratios ra and rn. We utilize Adam [20] for the model optimization. |