Reconstructing an Epidemic Outbreak Using Steiner Connectivity
Authors: Ritwick Mishra, Jack Heavey, Gursharn Kaur, Abhijin Adiga, Anil Vullikanti
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we evaluate our formulation and algorithms for several synthetic and realistic contact networks, including a contact network for the University of Virginia (UVA) hospital, constructed using Electronic Health Record (EHR) data. Our results show improved performance compared to a prior baseline in identifying missing infections. |
| Researcher Affiliation | Academia | Ritwick Mishra1,2, Jack Heavey1,2, Gursharn Kaur1, Abhijin Adiga1, Anil Vullikanti1,2 1 Biocomplexity Institute & Initiative, University of Virginia 2 Department of Computer Science, University of Virginia {mbc7bu, jch7jm, fug3aj, abhijin, vsakumar}@virginia.edu |
| Pseudocode | Yes | Algorithm 1: MINCOSTSTEINERTREE Input: An undirected contact graph G = (V, E) with edge probabilities pe and a set of observed infected nodes S Output: Tree Tr consistent with S ... Algorithm 2: MINCOSTSTEINERTREE-OBS-UNINFECTED Input: An undirected contact graph G = (V, E) with edge probabilities pe, set of observed uninfected nodes S0, a set of observed infected nodes S1 Output: Tree Tr consistent with S0, S1. |
| Open Source Code | No | The paper does not include an explicit statement or a link indicating that the source code for their proposed method (MINCOSTSTEINERTREE) is publicly available. |
| Open Datasets | Yes | 1. ar Xiv High Energy Physics-Theory (HEP-TH): This is an academic collaboration network in the High Energy Physics-Theory community based on the citations in the ar Xiv preprints published between January 1993 and April 2004 (Gehrke, Ginsparg, and Kleinberg 2003; Leskovec and Krevl 2014). Taking the largest connected component, we generate a subgraph with n = 500 nodes, obtained by BFS starting from a random node. We refer to it as arxiv. ... SNAP Datasets: Stanford Large Network Dataset Collection. http://snap.stanford.edu/ data. Accessed: 2022-7-25. |
| Dataset Splits | No | The paper mentions generating infection cascades and observed node sets, including randomly sampling a fixed percentage of nodes or choosing nodes at a certain distance from the source. It also states that "All reported values are averaged over 100 trials." However, it does not specify explicit train/validation/test splits by percentages or counts, nor does it explicitly mention a "validation set" for hyperparameter tuning or model selection. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used to run the experiments. |
| Software Dependencies | No | The paper mentions using "Klein and Ravi's (1995) algorithm" for the node-weighted Steiner tree problem and comparing against "CULT (Rozenshtein et al. 2016)", but it does not specify any software names with version numbers (e.g., Python, PyTorch, specific libraries or solvers with their versions) that would be needed to replicate the experimental setup. |
| Experiment Setup | Yes | We use the homogeneous probability setting for our experiments where we set the diffusion probability p across all edges to be the same. ... varying the probability p from 0.05 to 0.49. ... we have considered cascade sizes to be within (0.02n, 0.1n), where n is the network size so that sufficient number of observed nodes can be extracted from the cascade. ... We compare our method against CULT (Rozenshtein et al. 2016) which is the state-of-the-art Steiner tree-based cascade reconstruction method. ... We choose Matthews correlation coefficient (MCC) (Matthews 1975) and F1-score as in (Rozenshtein et al. 2016; Jang et al. 2021), to evaluate the quality of the reconstructed cascades with the ground-truth. All reported values are averaged over 100 trials. |