Machine learning using large datasets stands as one of the most transformative technologies of the 21st century.   It enables reliable face and speech recognition, internet search and monetization, computer vision, and fully autonomous vehicles.  Machine learning proficiency requires software skills as well as an understanding of the underlying mathematics and theoretical concepts. This class introduces important aspects of deep learning using a computation-first approach .  It emphasizes using frameworks to solve reasonably well-defined machine learning problems.  Two advanced courses provide a deeper study of mathematical concepts:  EE 559 Machine Learning I: Supervised Methods and EE 641 Deep Learning Systems.

Lecture

Monday (section: 30799)

15:00 - 16:50

Attendance and Participation

This class is offered with in-person enrollment ONLY. Attendance is mandatory to all lectures. You are responsible for missed announcements or changes to the course schedule or assignments. Taping or recording lectures or discussions is strictly forbidden.

Piazza

https://piazza.com/usc/fall2022/ee541

Piazza enables fast and efficient help from classmates and instructors.  Use Piazza to post questions about course material, homeworks, and policies instead of emailing questions to the teaching staff.

Canvas

https://canvas.usc-ece.com

Use Canvas to electronically submit your homework and view course grades.  You will receive an email to register during the first week of classes. Contact Dr.  Franzke with any technical issues.

Use Autolab to electronically submit programming portions of homework for  “auto-grading” .  You will receive an email to register during the first weeks of the course.  Contact Dr.  Franzke with technical issues.

Grader: Zeyu Liu

E-mail: [email protected]

Hours: by appointment

Learning objectives

Upon completion of this course a student will be able to:

❼ Understand the fundamentals and implement linear regression, a linear classifier, and logistic re-

gression.

❼ Apply common deep architectures such as multilayer perceptron and convolutional neural networks

and understand where each is most applicable.

❼ Be proficient in Python programming, including loops, conditionals, lists, dictionaries, classes, and

standard programmatic patterns.

❼ Be proficient in numerical Python using NumPy, SciPy, and matplotlib for design and analysis. ❼ Organize, store, and access datasets such as .npz files, .h5 files, pickle, and pandas.

❼ Understand the role of machine learning frameworks such as scikit-learn.

❼ Understand the role and use of deep-learning frameworks such as PyTorch.

❼ Use frameworks to train MLP, convolutional, and recurrent networks to solve machine learning

problems.

❼ Be proficient with relevant computing and cloud computing resources such as:

– linux command line interface and automation with shell scripts

– version control software such as git

– cloud GPUs to train deep networks and understand fundamentals of accelerated computing.

Course materials

❼ “Neural Networks and Deep Learning”, Michael Nielson. (online: http://neuralnetworksanddeeplearning. com).

❼ “Deep Learning with PyTorch”,  Eli Stevens,  Luca Antiga, Thomas Viehmann,  Manning, 2020.

(online: https://pytorch.org/assets/deep-learning/Deep-Learning-with-PyTorch.pdf).

❼ “Deep Learning”, Ian Goodfellow, Yoshua Bengio, Aaron Courville, The MIT Press, 2016. (online:

http://www.deeplearningbook.org).

❼ “Python Programming And Numerical Methods: A Guide For Engineers And Scientists”, Qingkai

Kong, Timmy Siauw, Alexandre Bayen, Elsevier, 2020. (online: https://pythonnumericalmethods. berkeley.edu/notebooks/Index.html).