Course overview
This course explains what is meant by bias in the context of machine learning algorithms, how bias can be present in the training data, and how it can be unintentionally introduced in the learning phase. Learners will gain an appreciation of why this is a concern and why it needs to be addressed.
Who is this course for?
This course is divided into two sections:
- Milestones 1-2: Conceptual background and context
- Milestones 3-5: Practical applications
Milestones 1-2 are suitable for all learners and do not require specific background skills and knowledge.
Milestones 3-5 are best suited for learners with the following background skills and knowledge:
- (Beginner Level) Experience writing code in Python
- Comfortable running Python in Jupyter notebooks
- (Beginner Level) Python Data Science Libraries
- Familiarity with the scikit-learn, Pandas and Numpy Python libraries
- (Beginner Level) Machine learning concepts
- Data: preprocessing/cleaning, importing into python, train/test splits, familiarity with tabular data in csv format (input variables and target variables)
- Modelling: training a model, using a model to generate predictions, computing performance metrics, broad understanding of classification and regression problems
Learning outcomes
By the end of Milestones 1-2, learners will be able to:
- Describe foundational ideas in this space, from philosophy, to law, and how they relate to fairness concerns in AI development and deployment
- Explain the legislative and regulatory contexts for AI and considerations within this context for mitigating bias
- Start mapping out the legislative context relating to AI for your own particular organisation and consider the steps you should take to be legally compliant and to mitigate bias from a compliance perspective
By the end of Milestones 3-5, learners will be able to:
- Describe different techniques and metrics for measuring bias
- Explain trade offs for each metric and know which ones are applicable in different scenarios
- Measure bias in data and models using standard techniques
- Describe the different techniques for addressing bias
- Implement Debiasing as a pre-processing and post-processing step
Materials can also be found on GitHub.
You may also be interested in:
Article: The Best Technical Resources for Bias Mitigation
Course: Assessing and mitigating bias and discrimination in AI: Beyond binary classification