Grant: Discovery Grants Program - Individual
McGill University
$46,000.00
Noise and Uncertainty Quantification in Statistical Network Analysis
EF 2023-24
Résumé
My proposed research will focus on development of statistical methods and theory in the context of noisy networks. The statistical analysis of network data faces inherent and, in many ways, unique challenges, including the basic relational aspect of the data; complex statistical dependencies among the data without a natural geometry to leverage; and, typically, the sheer volume of the data (i.e., frequently high dimensional and often massive in quantity). Novel statistical methodologies have necessarily been developed to meet these challenges and others for both canonical and domain driven network problems. Yet the vast majority of methodologies used in practice assume the connectivity structure of the network(s) at hand is obtained error free. Such an assumption is untrue in almost all contexts to a greater or lesser extent, particularly in settings where network topology is inferred, such as computational biology or neuroscience. My own work in statistical network analysis (going back almost 20 years) has most recently focused on beginning to characterize and account for how uncertainty at the level of network topology propagates to higher level inferential tasks and how we can best quantify that uncertainty. The research I propose would substantially deepen and broaden this work. The long-term vision of my Discovery Grant program is to develop a comprehensive framework for uncertainty propagation in noisy networks, unified through the notion of a network `signal plus noise' model. The standard version of this model assumes that we observe a network that is the result of flipping the (non)edge status for every pair of nodes independently with some common probabilities. This perspective captures the notion of observing `noisy networks' with Type I and II statistical errors (i.e., `edges' observed between nodes i and j when no such edge truly exists, and `no edges' observed when an edge does actually exist). In the short term, my five-year plan is to further develop the theory and methods for uncertainty quantification for network inference in three contexts: (i) under the signal plus noise model; (ii) under generative models corrupted by signal plus noise; and (iii) for noisy dynamic networks. The emergence and growth of the multidisciplinary field of complex network analysis over the past two decades has revolutionized the way we analyze complex systems across the sciences, humanities, and medicine. Statistical concepts, methods, and theory, not surprisingly, lie at the heart of much of the development in this area of research. Impact of the statistical development pursued under this application will be illustrated through existing collaborations in a number of areas, including bioinformatics, computational neuroscience, and political science. Furthermore, adoption will be facilitated by open distribution of software and data artifacts via github, a long-standing practice in my research group.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Quebec, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data
Tel que publié
Application id
RGPIN-2023-03566
Name nom
Kolaczyk, Eric
Department département
Mathematics & Statistics
Organization id
61
Institution établissement
McGill University
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2023
Competition year année de concours
2023
Award amount
46000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1508
Committee name en
Mathematics and Statistics
Committee name fr
Mathématiques et statistique
Area of application code
1205
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Mathematical sciences
Area of application fr
Sciences mathématiques
Research subject code
9999
Research subject group en
Not available
Research subject group fr
Not available
Research subject en
Non disponible
Research subject fr
Non disponible
Application title
Noise and Uncertainty Quantification in Statistical Network Analysis
Keyword
Bayesian inference; bootstrap; complex networks; dynamic network; hidden Markov model; random graph model; signal plus noise; subgraph count; uncertainty quantification
Application summary
My proposed research will focus on development of statistical methods and theory in the context of noisy networks. The statistical analysis of network data faces inherent and, in many ways, unique challenges, including the basic relational aspect of the data; complex statistical dependencies among the data without a natural geometry to leverage; and, typically, the sheer volume of the data (i.e., frequently high dimensional and often massive in quantity). Novel statistical methodologies have necessarily been developed to meet these challenges and others for both canonical and domain driven network problems. Yet the vast majority of methodologies used in practice assume the connectivity structure of the network(s) at hand is obtained error free. Such an assumption is untrue in almost all contexts to a greater or lesser extent, particularly in settings where network topology is inferred, such as computational biology or neuroscience. My own work in statistical network analysis (going back almost 20 years) has most recently focused on beginning to characterize and account for how uncertainty at the level of network topology propagates to higher level inferential tasks and how we can best quantify that uncertainty. The research I propose would substantially deepen and broaden this work. The long-term vision of my Discovery Grant program is to develop a comprehensive framework for uncertainty propagation in noisy networks, unified through the notion of a network `signal plus noise' model. The standard version of this model assumes that we observe a network that is the result of flipping the (non)edge status for every pair of nodes independently with some common probabilities. This perspective captures the notion of observing `noisy networks' with Type I and II statistical errors (i.e., `edges' observed between nodes i and j when no such edge truly exists, and `no edges' observed when an edge does actually exist). In the short term, my five-year plan is to further develop the theory and methods for uncertainty quantification for network inference in three contexts: (i) under the signal plus noise model; (ii) under generative models corrupted by signal plus noise; and (iii) for noisy dynamic networks. The emergence and growth of the multidisciplinary field of complex network analysis over the past two decades has revolutionized the way we analyze complex systems across the sciences, humanities, and medicine. Statistical concepts, methods, and theory, not surprisingly, lie at the heart of much of the development in this area of research. Impact of the statistical development pursued under this application will be illustrated through existing collaborations in a number of areas, including bioinformatics, computational neuroscience, and political science. Furthermore, adoption will be facilitated by open distribution of software and data artifacts via github, a long-standing practice in my research group.