ECE 655 : Advanced GPU Programming and Deep Learning
| Cross-Listings : | None | |
| Years : | 2023- | |
| Lecture : | 1 class per week (2 hours 40 min) | |
| Lab : | - | |
| Prerequisites : | Students are expected to be proficient in C or Python language. | |
| Environment : | HOPPER GPU cluster, Anaconda Package, PyTorch. | |
| Description : |
This course provides an overview of the traditional Machine Learning (ML) methods such as regression and gradient descent before the emerging deep learning (DL) structures became popular 2009. Built on this foundation, it introduces the PyTorch package to implement DL methods. Foundational methods in DL are introduced such as entropy, tensors, optimization methods, feature space, classification, regularization, gradients, dropout, and learning rate. Existing PyTorch architectural building blocks for constructing Convolutional Neural Networks (CNNs) such as activation functions, gradients, and filters are studied to build custom CNNs. Furthermore, existing popular CNNs by Google, Microsoft, and Facebook are also introduced. Methods to speed up these computations using GPUs are rigorously studied. |
|
| Workload : | 6 projects, 1 final presentation. No exams |