When in Doubt: Neural Non-Parametric Uncertainty Quantification for Epidemic Forecasting

Authors: Harshavardhan Kamarthi, Lingkai Kong, Alexander Rodriguez, Chao Zhang, B. Aditya Prakash

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments in a real-time flu forecasting setting show that EPIFNP significantly outperforms state-of-the-art models in both accuracy and calibration metrics, up to 2.5x in accuracy and 2.4x in calibration.
Researcher Affiliation Academia College of Computing Georgia Institute of Technology {harsha.pk,lkkong,arodriguezc,chaozhang,badityap}@gatech.edu
Pseudocode Yes The pseudocode for Autoregressive inference is given in the Appendix.
Open Source Code No The code is implemented using Pytorch and will be released for research purposes.
Open Datasets Yes The CDC uses the ILINet surveillance system to gather flu information from public health labs and clinical institutions across the US. It releases weekly estimates of weighted influenza-like illness (w ILI)1: out-patients with flu-like symptoms aggregated for US national and 10 different regions (called HHS regions).1https://www.cdc.gov/flu/weekly/flusight/index.html
Dataset Splits No The paper states, 'During training phase of our supervised learning task, EPIFNP is trained to predict x(t+k) i given x(1...t) i as input for i N. Therefore, we define the training set M as set of partial sequences and their forecast ground truths from historical data H, i.e, M = {(x(1...t) i , y(t) i ) : i N, t + k T, y(t) i = x(t+k) i }.' It also mentions evaluating on 'w ILI data of 17 seasons from 2003/04 to 2019/20' and 'new unseen partial sequence x' for testing, but does not provide explicit percentages, counts, or a detailed methodology for how the dataset is split into training, validation, and test sets beyond the time-based separation of seasons.
Hardware Specification Yes All experiments were done on an Intel i5 4.8 GHz CPU with Nvidia GTX 1650 GPU.
Software Dependencies No The code is implemented using Pytorch and will be released for research purposes. The paper mentions 'Pytorch' but does not specify a version number or other software dependencies with versions.
Experiment Setup No The paper mentions that 'The model typically takes around 20 minutes to train' and refers to 'Adam' as the optimizer, but it states 'Supplementary contains additional details and results (e.g. hyperparameters, results on additional metrics (MAPE), additional case and ablation studies)', indicating that specific hyperparameter values and detailed training configurations are not present in the main text.