Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Grants Program - Individual

University of Manitoba

$31,000.00

Bayesian Methods and Computing with Applications

EF 2024-25

Résumé
We are moving towards a data-rich digital world and open Artificial Intelligence (AI) continues to evolve and integrate into our daily lives. Bayesian methods are becoming more important in this era in making inference and decision-making in real life problems and applications. Bayesian methods allow AI and machine learning models to update their beliefs and make decisions based on observed data and prior knowledge in a continuous fashion in real time and space. More specifically, Bayesian state space models with clustering and classification techniques can be used to address certain complex problems for feature extraction and inferential purposes. In this proposal, I will focus on developing Bayesian state space models, Bayesian classification and clustering methods which provide a flexible framework to incorporate prior knowledge and iteratively refine predictions in machine learning, as more data are observed.  I will develop new Bayesian state space modelling frameworks which are computationally efficient and account for temporal and spatial variations as well. In high dimensional settings, I will develop new Bayesian classification and clustering methods for feature selections. It is particularly useful when dealing with uncertain or noisy, sparse or imbalanced data and allows a way to incorporate prior knowledge about the classes. This involves specifying prior probabilities for each class and likelihood functions that describe how the features are distributed for each class. When the data are imbalance in the applications such as as fraud detection, medical diagnosis, rare event prediction, and anomaly detection, I will develop Bayesian and Dirichlet process based classification methods which handles the imbalance with higher prediction accuracy. I will compare the performance of the methods against existing benchmark resampling methods such as random oversampling, synthetic minority over-sampling techniques, random under sampling and ensemble methods. I will also develop Bayesian and non-paramteric Bayesian model-based clustering methods that only require a dissimilarity matrix as input, while also accounting for the measurement error that may be present within the recorded data. This requires extending upon the existing Bayesian multidimensional scaling methods by proposing several additional constraints, model selection criteria, and strategies for reducing computation time. These extensions will ensure that the algorithms are effective even in high-dimensional spaces and provides a wide range of choices to the practitioner that can be used with a variety of data. In terms of assessing the models, I will develop gradient-free statistical goodness-of-fit tests. These methods will not require the calculations of derivatives which will be impactful in evaluating the performance of AI generative models. The new methods will also be tested on applications arising from environmental, health and engineering sciences for improved inferences.
Programme
Discovery Grants Program - Individual
Emplacement
Manitoba, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
RGPIN-2024-05460
Name nom
Muthukumarana, Palavinnage
Department département
Statistics
Organization id
19
Institution établissement
University of Manitoba
Province en
Manitoba
Province fr
Manitoba
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
31000
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
1200
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Advancement of knowledge
Area of application fr
Avancement des connaissances
Research subject code
3009
Research subject group en
Statistics and probability
Research subject group fr
Statistique et probabilité
Research subject en
Parametric inference
Research subject fr
Inférence paramétrique
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Bayesian Methods and Computing with Applications
Keywords
markov chain monte carlo; bayesian inference; dirichlet process; simulation based inference; state space models; smooth dirichlet distribution; hierarchical models; goodness of fit testing; clustering; classification
Application summary
We are moving towards a data-rich digital world and open Artificial Intelligence (AI) continues to evolve and integrate into our daily lives. Bayesian methods are becoming more important in this era in making inference and decision-making in real life problems and applications. Bayesian methods allow AI and machine learning models to update their beliefs and make decisions based on observed data and prior knowledge in a continuous fashion in real time and space. More specifically, Bayesian state space models with clustering and classification techniques can be used to address certain complex problems for feature extraction and inferential purposes. In this proposal, I will focus on developing Bayesian state space models, Bayesian classification and clustering methods which provide a flexible framework to incorporate prior knowledge and iteratively refine predictions in machine learning, as more data are observed.  I will develop new Bayesian state space modelling frameworks which are computationally efficient and account for temporal and spatial variations as well. In high dimensional settings, I will develop new Bayesian classification and clustering methods for feature selections. It is particularly useful when dealing with uncertain or noisy, sparse or imbalanced data and allows a way to incorporate prior knowledge about the classes. This involves specifying prior probabilities for each class and likelihood functions that describe how the features are distributed for each class. When the data are imbalance in the applications such as as fraud detection, medical diagnosis, rare event prediction, and anomaly detection, I will develop Bayesian and Dirichlet process based classification methods which handles the imbalance with higher prediction accuracy. I will compare the performance of the methods against existing benchmark resampling methods such as random oversampling, synthetic minority over-sampling techniques, random under sampling and ensemble methods. I will also develop Bayesian and non-paramteric Bayesian model-based clustering methods that only require a dissimilarity matrix as input, while also accounting for the measurement error that may be present within the recorded data. This requires extending upon the existing Bayesian multidimensional scaling methods by proposing several additional constraints, model selection criteria, and strategies for reducing computation time. These extensions will ensure that the algorithms are effective even in high-dimensional spaces and provides a wide range of choices to the practitioner that can be used with a variety of data. In terms of assessing the models, I will develop gradient-free statistical goodness-of-fit tests. These methods will not require the calculations of derivatives which will be impactful in evaluating the performance of AI generative models. The new methods will also be tested on applications arising from environmental, health and engineering sciences for improved inferences.