Conformal Prediction with Temporal Quantile Adjustments

Authors: Zhen Lin, Shubhendu Trivedi, Jimeng Sun

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate TQA s performance through extensive experimentation: TQA generally obtains efficient PIs and improves longitudinal coverage while preserving cross-sectional coverage.
Researcher Affiliation Academia Zhen Lin1 Shubhendu Trivedi2 Jimeng Sun1,3 1 Department of Computer Science, University of Illinois at Urbana-Champaign 2 Massachusetts Institute of Technology 3 Carle Illinois College of Medicine, University of Illinois at Urbana-Champaign
Pseudocode No The paper describes algorithmic steps and equations (e.g., Eq. 9) but does not present them in a formally labeled 'Pseudocode' or 'Algorithm' block.
Open Source Code Yes Our code is available at https://github.com/zlin7/TQA.
Open Datasets Yes Datasets We test our methods and baselines on the following datasets: Electronic health records data for white blood cell counts (WBCC) prediction (MIMIC [23, 18, 22]), COVID-19 cases prediction (COVID [10]), Electroencephalography trajectory prediction after visual stimuli (EEG [51]), energy load forecasting (GEFCom [20]), and healthcare claim amount prediction (CLAIM) using data from a large American healthcare data provider.
Dataset Splits Yes To construct a PI for Yi,t, we first split our data {Si}N i=1 into a proper training set and a calibration set [41]. Table 1: Number of TSs in each dataset along with the length. # train/cal/test 192/100/100 2393/500/500 200/100/80 300/100/200 1198/200/700
Hardware Specification No The provided paper text does not contain specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions 'RNN as the base point estimator' and 'LSTM ([19])' and 'Adam [24]', but does not specify specific software library names with version numbers (e.g., PyTorch 1.9, TensorFlow 2.x).
Experiment Setup Yes We use RNN as the base point estimator due to its flexibility and for comparison with [48]. We use α = 0.1, and a LSTM ([19]) similar to that in [48] (full implementation details in the Appendix). For TQA-E, we use γ = 0.005 following [17].