Responsible AI and Application Development

Course details

In this course, award-winning AI researcher Laurence Moroney guides you through how to ensure responsible and ethical AI development. Learn how to collect and preprocess data in ways that maintain privacy and reduce biases. Discover advanced techniques in data augmentation that enhance your datasets while safeguarding against introducing unfairness. Dive into optimizing neural network architectures and learn how to identify and mitigate potential biases in your loss functions and optimization strategies. Explore transfer learning and how to assess and address inherited biases from pre-trained models. By focusing on ethical AI, you can build systems that not only perform well but also serve all user demographics fairly.

Learning objectives
Understand the end-to-end process for creating an AI application with ML.
Explore how biases can create unintended harm at each step of the way.
Learn the tooling and processes to build better AI applications.

Concepts

0. Introduction

01 - Responsible AI and application development

1. Designing Responsible AI

02 - Responsible design
03 - Being responsible end to end

2. Acquiring Data

04 - Ethical data collection
05 - Tools to understand your data
06 - Demo of model cards, nutrition labels, etc.

3. Preprocessing Data

07 - Why preprocess data
08 - Techniques to preprocess data
09 - Demo of data preprocessing

4. Augmenting Data

10 - What is data augmentation
11 - When data augmentation goes wrong
12 - Demo of image augmentation with PyTorch

5. Creating Model Architectures

13 - Optimization - Responsible neural architecture design
14 - Loss - Responsible neural architecture design
15 - Others - Responsible neural architecture design

6. Transfer Learning

16 - What is transfer learning
17 - The risks of transfer learning - Inheriting bias
18 - The risks of fine-tuning - Inheriting bias
19 - Demonstrate BERT, inherited bias, and fine-tuning to fix

7. Model Training

20 - Responsible practices during training
21 - Training an image classifier

8. Model Deployment and Applications

22 - Considerations in model deployment
23 - Ops and continuous evaluation

Conclusion

24 - Continue your responsible AI journey