Graduate Students Seminar

Speaker 1: Biswajit Basak

Title

Statistical Privacy Protection

Abstract

Statistical privacy protection focuses on techniques that preserve individual privacy while enabling meaningful data analysis. Key methods include data perturbation, where controlled noise is introduced to obscure sensitive information, and synthetic data generation, which creates artificial datasets to represent the original data. These approaches allow for the analysis of data without compromising the privacy of individuals. In this seminar, I will examine several widely used data perturbation techniques, highlighting their effectiveness in safeguarding privacy while still enabling valid inferences to be drawn. I will also explore methods for assessing privacy measures, ensuring that these techniques provide a balanced approach between privacy protection and data utility.

Speaker 2: Jalil  Ahmad

Title

A Local Nudging-Based Optimization Approach for Parameter Estimation in Dynamical Systems

Abstract

Differential equations serve as the mathematical foundation for models that describe physical, chemical, biological, and even financial systems. These models can involve both constant and time-varying parameters, which are usually estimated from observed data. In this study, we introduce a local nudging-based optimization approach to estimate these parameters. This method is capable of estimating both constant and time-varying parameters in both chaotic and non-chaotic dynamical systems. Numerical simulations conducted on the Lorenz system demonstrate that the parameter estimates achieve machine precision when the data is perfect, and the error remains below or equal to the noise level when the data is noisy. Moreover, this approach is relatively inexpensive compared to other methods available in the literature.