Below, Prof. Rama Ramakrishnan describes how he has changed the course since its original conception.
We make a number of small tweaks every year, but perhaps the biggest change is in how we teach the transformer architecture.
Previously, we used to teach the use of convolutional neural networks as the fundamental deep learning approach for solving computer vision problems, and the transformer for natural language processing problems. Given the increasing use of transformers for computer vision problems, we have switched to using transformers for both natural language processing and computer vision. We also show how to use transformers to solve problems with multi-modal input data.
