Dr. Linu Pinto

The research focuses on various transformations and advanced optimization techniques used in developing mathematical models for algorithms in deep learning. It specifically examines the development of optimization algorithms designed to minimize the objective functions by iteratively moving in the direction of the steepest descent or negative gradient. The research also involves analyzing the curvature of the objective function and how well the concept of Hessian can contribute to the development of advanced optimization algorithms in deep learning with less computational complexity. Advanced concepts from linear algebra, such as eigen decomposition, singular value decomposition (SVD), and the pseudoinverse, are utilized to explore the optimization in deep models. Additionally, the research explores the advantages and the limitations of various adaptive activation functions and development of new adaptive activations for better convergence of iteration algorithms and to identify an optimal differentiable functions for prediction and classification models in deep neural networks. Future research will concentrate on studying various non-differentiable objective functions to develop better optimization strategies in deep learning.