The course covers the mathematical, statistical, and algorithmic principles that underlie the development of trustworthy machine learning systems, with a focus on fairness and diversity, interpretability and explainability, reliability and robustness, and privacy and federation.
- The need for trustworthy machine learning; motivation and use cases.
- Algorithmic fairness, diversity, and bias mitigation.
- Interpretability and explainability of machine learning models.
- Confidence and uncertainty in machine learning models.
- Resilience in the presence of changing and adversial environments.
- Integrity and federated learning.
After passing the course, the student should be able to
- apply trustworthy machine learning algorithms
- implement baseline algorithms for value-based, transparent, and robust machine learning
- analyze the trustworthiness of machine learning model inferences
- explain fundamental concepts that ensure trustworthiness in machine learning
- derive and prove mathematical statements that guarantee trustworthiness in machine learning
- describe strengths and weaknesses of methods to enhance trustworthiness in machine learning models
in order to understand how to build models that are not only accurate but also reliable in the sense that they are trustworthy, transparent, and align with human values. Without this knowledge, developers of machine learning algorithms risk implementing systems that are biased, opaque, vulnerable to attack, or unfit for high-stakes decision-making situations.
