Expert talk on AI compilers with Pytorch

The speaker, Sasank Chilamkurthy, took the participants through an in-depth exploration of the fundamental principles and advanced methodologies underpinning deep learning. The key role of Graphics Processing Units (GPUs) in facilitating the computational demands inherent in training complex neural network models was discussed. Through a comprehensive examination, participants gained insight into the architectural design of GPUs and their optimization for  parallel processing tasks, particularly in the context of deep learning. The session provided a comprehensive overview of CUDA programming, a parallel computing platform and application programming interface (API) model developed by NVIDIA This entailed a detailed exploration of the hierarchical structure inherent in CUDA, comprising threads, blocks, and kernels, elucidating their respective roles in optimizing computation and memory management. The session covered comparative analysis, juxtaposing PyTorch against alternative deep learning frameworks such as OpenCV and TensorFlow. Through rigorous examination and empirical evidence, participants understood the unique advantages offered by PyTorch, including its flexibility, dynamic computational graph construction, and intuitive programming interface. The guest talk on AI Compilers with Pytorch by Sasank Chilamkurthy, co-founder and CTO of Qure.ai, a leading radiology AI company was organized by IEEE CS PESU on March 6, 2024 at RR campus.