Graduate Students Seminar

Speaker 1: Emoke Galambos

Title

Solving a linear-quadratic parabolic optimal control problem using the cG(1) dG(1) method

Abstract

PDE constrained optimization is a mathematical field that couples complex optimization problems with partial differential equations to address the challenges of optimizing system performance, while adhering to various constraints. It is widely applied in areas, such as engineering, medical imaging or environmental modeling.

In my talk, I would like to introduce the implementation of the cCG(1) dG(1) (continuous - discontinuous Galerkin) method, to solve a time-dependent, linear-quadratic parabolic optimal control problem, modeling heat distribution. The discretization of the problem involves the application of finite elements, using linear basis functions, which are continuous in space and discontinuous in time.

Speaker 2: Vishal Subedi

Title

Prediction Using Machine Learning and Data Mining

Abstract

Accurate rainfall prediction is crucial for agriculture, disaster management, and water resource planning. This project applies machine learning and data mining techniques to enhance prediction accuracy using the WeatherAUS dataset from Kaggle, which contains 145,460 entries and 23 features. We address challenges like missing values through KNN imputation and tackle class imbalance using oversampling and undersampling.  Among five models tested, Random Forest with KNN imputation and oversampling performed best. We also explore imputation using correlated features but find no significant improvement. Future work includes integrating advanced time-series models and spatial data for better predictive accuracy.