CAREER: Developing Evaluation Methods for Network-based Findings

点击翻译按钮获取中文摘要

Networks are all around us: transportation networks, power networks, communication networks, social networks, biological networks, the Internet, and many others. Networks are studied by various disciplines within science and engineering to glean the invaluable insights that networks can carry. However, most studies of networks look at networks that are uncertain, e.g., are noisy. Generalizing findings derived from such uncertain networks can lead to various erroneous findings. The goal of this project is to develop methods that can evaluate findings derived in studies of networks under these uncertainties. The developed techniques contribute towards a future in network science research, where findings are validated, reproducible, and comparable. The outcomes of this project will contribute to the development of a diverse globally-competitive STEM workforce, full participation of women and underrepresented minorities, enhanced research and education infrastructure, and increased public scientific literacy and engagement. The education plan involves four groups: K-12, undergraduate students, graduate students, and beyond university/public. Educational objectives are designed for these groups by integrating research into education, curricular development, presenting tutorials and seminars, and development of software and public data repositories.The project is motivated by the challenges and the research questions that exist around evaluating findings in studies of networks, e.g., whether the findings are specific to the network study, or if the data was correctly collected for the study. Addressing these challenges and answering these questions, the project will lead to the development of new knowledge on evaluating findings derived from networks, advancing the state of research on reproducibility in network science. It will provide a multiangle perspective towards evaluation of network science findings by developing evaluation methods and metrics to assess findings in terms of general scientific evaluation metrics, e.g., authenticity of findings. New network embedding methods will be designed especially for evaluation purposes. In particular, the project will develop specificity metrics for network studies to determine how specific are the findings to a network study. These metrics will be introduced by designing "graph identification" mechanisms, which rely on network embedding methods and designing an ``identity" for a graph. The project will further develop methods that can provide acceptance/rejection likelihoods for network-based findings, as a way to assess their authenticity. While the problems are in general NP-Complete, alternative practical solutions with acceptable performances are pursued by designing "network-based authentication" mechanisms that are inspired by biometrics research. The methods introduced contribute to the broader research areas of network representation and embedding, facilitate new applications such as network identification and authentication, and extend well-known techniques from statistics to networks.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.