ºÚÁÏÉçÇø

Bulletin Description:

This course introduces basic concepts for deep learning and covers the preliminary knowledge of neural networks. Students will learn to implement and train their own neural networks and gain an understanding of research in deep learning. Additionally, ºÚÁÏÉçÇøs will complete a final project to design and train a customized neural network on selected real-world problems. Through multiple hands-on assignments and the final course project, ºÚÁÏÉçÇøs will acquire the skills to perform deep learning tasks and the best practices to train and fine-tune deep neural networks.

Prerequisite:

CSE 374

Learning Outcomes:

1. Describe fundamental concepts of deep learning, including: a) the machine learning workflow, from dataset to training and using a model; and b) the analysis of different neural network models and architectures based on the type of tasks, such as regression or classification.
2. Explain common deep learning algorithms and create a basic deep learning model without relying on third-party libraries. This includes: a) drawing the computational graph for a basic deep neural network model and manually carrying out the backpropagation; b) implementing the basic deep neural network model using Python code, without existing deep learning libraries such as Pytorch and TensorFlow.
3. Explain advanced neural network structures and evaluate the appropriate structure for a given task, specifically: (a) describe different architectures of neural network models such as Convolution Neural Networks (CNNs) and the Long Short-Term Memory (LSTM) networks, (b) identify the tasks on which a given architecture would be appropriate, and (c) apply the right architecture for a given task via the Pytorch library.
4. Apply different training strategies, including by: a) analyzing the training output, and b) using the correct training strategies to improve the performance.
5. (Graduate ºÚÁÏÉçÇøs only) Adapt methods and metrics from recent research articles (beyond what is covered in class) to develop robust, fine-tuned, deep learning models.
6. (Graduate ºÚÁÏÉçÇøs only) Demonstrate an ability to do independent research in identifying and applying state-of-the-art metrics to assess and evaluate deep learning models.