Week 13 Info: Deep Learning and Convolutional Neural Networks
Week 13 has arrived. During this week we will continue our study of neural networks by exploring more complex architectures. We will begin by learning about convolutional neural networks, which are the models that are most successful at recognizing images. Convolutional neural networks have an inductive bias that allows them to exploit the symmetries of image classification problems, i.e. that translations of images are still the same object. This makes CNNs extremely efficient per parameter at solving image classification problems. We will learn about other ways to take advantage of the structure of image recognition problems to improve model accuracy. The study of CNNs will lead us to Recurrent Neural Networks, which exploit a sequential symmetry of text and other sequences and have led to the development of LLMs.
You can read more about the plan for the week and the reading/resources in Module 13.
Lab 7 due this Sunday, May 3rd, at midnight.