Welcome to the GitHub repo of the course "Deep Learning & AI Ethics" course taught taught at Belmont University in Fall 2023! This course includes an "embedded ethics" pedagogical approach in which we will build DL systems, probe them to try to understand how they work, and investigate their ethical implications.

This course presents an overview of current machine learning techniques and applications, with particular attention to deep neural network models. Applications will include computer vision, audio signal processing, and natural language processing, for tasks such as classification and regression, and generative models. We will learn how these systems work, and will write code to train them and to probe their outputs and behavior. As an integrated part of the course, we will give attention to issues of ethics & society in training and deploying such models, issues including bias, transparency and accountability.

• Transformers (e.g., BERT, ChatGPT)
• EU AI Act
• Diffusion Models
• Licensing/Copyright & (Training) Generative Models
• Emphasis on fine-tuning, LoRAs

Scott H. Hawley

Some Python coding proficiency. (cf. "Lesson 2: Python/Jupyter Review")

We will mostly be using web links, videos, and lecture notes. The following will serve as textbooks:

• Hello World: Being Human in the Age of Algorithms by Hannah Fry, (or equivalently, "...Age of the Machine" is the exact same book.) This is an accessible and engaging survey of AI technology and ethics. It is non-technical though. For technical references (again, beyond lecture notes and web links) we may refer to these supplemental textbooks:

As technical references, a few excellent FREE online textbooks are:

• Neural Networks and Deep Learning by Michael Nielsen.
• Deep Learning by Ian Goodfellow, Yoshua Bengio, and Aaron Courville.
• The FastAI Book, aka "fastbook": "Deep Learning for Coders with Fastai and Pytorch: AI Applications Without a PhD" by Jeremy Howard & Sylvain Gugger. This is written entirely as executable Jupyter notebooks.
• The Little Book of Deep Learning by François Fleuret (2023)

• Schedule: This draft version of the course will be all on your own & asynchronous, but planned according to the synchronous MWF (Monday-Wednesday-Friday) schedule of the final version of the course. M&W will tend to focus on math & coding lessons, and Fridays being for ethics topics. (In the final course, Lessons will be "on your own"/asynchronous readings, videos, quizzes, and assignements, and Ethics Discussion meetings will be synchronous via Zoom.)

• Deep Learning framework: We'll start with raw NumPy, then for most of the course we'll use PyTorch with some fastai on top (for convenience).

• Assignments: You'll write coding assignments using online "notebook" hosting platforms Colab or Gradient, which give GPU access for free so you won't need your own GPU. The assignments contain instructions, boilerplate code, and "fill in the blank" areas, along with sample test outputs so you can check your work as you go. When you are finished, you'll use a Google Form to submit the "sharing URL" for your notebook. Then periodically, a server app will download any updated notebooks and run the assignment's "test suite" against them. You will then receive an email of your score, i.e., how many tests passed.

• Quizzes: For the draft version of this course, grades will not be issued for quizzes. In the final version, quizzes will be on Blackboard.

See the Lessons area. These go with companion videos on YouTube.

See the Assignments area.

These materials are governed by Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International. You may use them, with attribution, for non-commercial purposes. As Francois Fleuret says, "More simply: I am okay with this material being used for regular academic teaching, but definitely not for a book / youtube loaded with ads / whatever monetization model I am not aware of."

See Acknowledgements

(c) Scott H. Hawley, 2021-2023