Prasanga Dhungel

Modern Machine Learning systems demand large amounts of data and computational resources, limiting their accessibility and increasing their environmental footprint. This PhD aims to tackle these challenges by developing novel methodologies for improving the overall efficiency of machine learning. The research will focus on two synergistic areas: data efficiency and algorithmic optimization. On the data front, the project will explore curriculum learning strategies to intelligently select and order training samples, thereby maximizing sample efficiency and accelerating model convergence. On the algorithmic side, the research will design and refine mathematical optimizers that reduce computational overhead and improve learnability. The outcomes of this research will contribute to creating more sustainable, scalable, and resource-efficient ML systems that exceed current state-of-the-art performance.