Bio
Chanju is a PhD student in Physics at Swansea University and the Centre for Doctoral Training in Artificial Intelligence, Machine Learning and Advanced Computing (AIMLAC). Previously, he studied Lattice Field Theory, which discretises the target quantum field theory and simulates it on a cluster. He is interested in the Quantum / Statistical Field theoretical description of Stochastic Algorithms and Generative Models.
Research interests
The statistical/field theoretical description of machine learning algorithms allows us to analytically understand the flow of information during the training process and the role of hyperparameters within. Moreover, a qualitative understanding of the dynamics not only lets us explain why and how the models work but also lays a mathematical foundation to explain phenomena such as the Linear Scaling Rule and the Grokking transition. Studying the information flow of machine learning algorithms is also important in the context of application to lattice field theory. Generative machine learning models are good candidates to replace or aid the traditionally used Monte Carlo sampling method, reducing the critical slowing down or mitigating the sign problem. However, naive usage of machine learning models without understanding the symmetry of the system can lead to generating distributions representing incorrect physics. Therefore, an analytical understanding of the generative models is essential in the context of physical applications.