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 [1].
EARTH: Epidemiology-Aware Neural ODE with Continuous Disease Transmission Graph
Authors: Guancheng Wan, Zewen Liu, Xiaojun Shan, Max Sy Lau, B. Aditya Prakash, Wei Jin
ICML 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show EARTH superior performance in forecasting real-world epidemics compared to state-of-the-art methods. ... In this section, we comprehensively evaluate our proposed EARTH by answering the main questions: Q1: Performance. Does EARTH outperforms the existing state-of-the-art epidemic forecasting methods? Q2: Resilience. Is EARTH stable on different settings? Q3: Effectiveness. Are proposed two key components: EANO and GLTG both effective? Q4: Sensitivity. What is the performance of the proposed method with different hyper-parameters? The answers of Q1-Q4 are illustrated as follows. |
| Researcher Affiliation | Academia | 1Department of Computer Science, Emory University, USA 2Department of Computer Science, University of California, Los Angeles 3Department of Electrical and Computer Engineering, University of California, San Diego 4Department of Biostatistics and Bioinformatics, Emory University, USA 5College of Computing, Georgia Institute of Technology, USA. Correspondence to: Wei Jin <EMAIL>. |
| Pseudocode | No | The paper describes the methodology using mathematical equations and textual explanations without presenting any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code is available at https://github.com/ Guancheng Wan/EARTH. |
| Open Datasets | Yes | We leverage three datasets to examine the validity of our EARTH, to ensure fair and consistent comparisons with prior work (Liu et al., 2023; Kamarthi & Prakash), including COVID-19 and influenza-like illness: Australia-COVID, US-Regions, and US-States. Please see Appendix A for dataset details. ... Australia-COVID. Provided by JHU-CSSE1, this dataset records daily new COVID-19 cases, including 6 states and 2 territories, from January 2020 to August 2021. 1https://github.com/CSSEGISand Data/COVID-19 |
| Dataset Splits | No | The paper mentions a window size T of 20 and a prediction horizon h of 5, 10, and 15, but does not provide specific details on how the datasets are split into training, validation, and test sets, such as percentages or sample counts for each split. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running the experiments, such as GPU or CPU models, or memory specifications. |
| Software Dependencies | No | The paper mentions using SGD as the optimizer but does not specify any software library names with version numbers (e.g., PyTorch, TensorFlow, or Python versions) that would be needed to replicate the experiment. |
| Experiment Setup | Yes | In all experimental setups, we set the learning rate to 1e 3 and use SGD (Robbins & Monro, 1951) as the optimizer with a momentum of 0.9 and weight decay of 1e 5 (Bi et al., 2025a;b). The default hidden size is 64, and the window size T is 20. Considering that decision-makers need time to allocate prevention resources in epidemic modeling, we set the horizon h to 5, 10, and 15. We repeat each experiment five times for each dataset and record the average results. |