CSCI-B 658 Trustworthy Causal Artificial Intelligence
3 credits
- Prerequisite(s): None
- Delivery: On-Campus
Description
This course explores how causal methods enhance AI-model trustworthiness. Students apply causal models to represent cause-effect relationships governing human understanding, improving generalization to novel data and yielding fairer, more interpretable results. Topics include AI robustness, privacy, safety, and accountability; tradeoffs among assumptions; associational, causal, and counterfactual conclusions; and trustworthiness architectures.
Topics
Trust models
- Introduction to trust in AI
- Trustworthiness metrics
Structural causal models graphs
- Causal graph definitions
- Directed acyclic graphs (DAGs)
Structural causal models
- Model assumptions
- Causal relations
The backdoor criterion and confounding bias
- Identifying confounders
- Applying the backdoor criterion
The causal calculus
- Do-calculus
- Inference techniques
Causal bounding
- Bounds on causal effects
- Estimating causal effects
Data-fusion framework
- Combining data from multiple sources
- Handling incomplete data
AI interpretability
- Enhancing model transparency
- Causal approaches to interpretability
AI fairness
- Addressing bias in AI models
- Fairness metrics
AI robustness
- Ensuring stability in AI systems
- Robust causal inference
AI privacy
- Privacy-preserving AI techniques
- Causal analysis for privacy concerns
AI safety and accountability
- Ensuring safety in AI systems
- Causal reasoning for accountability
Causality in healthcare
- Applications of causal AI in healthcare
- Causal modeling for medical decisions
Learning Outcomes
- Analyze causal models to maximize the impact of causality on AI models.CS 4
- Compare various causality models and discuss their connection in various fields.CS 3
- Compare associational, causal, and counterfactual queries.CS 1 CS 4
- Design and implement authentication and authorization mechanisms, considering multifactor authentication, role-based access control, and integration with existing systems.CS 5
- Develop causal graphs using existing tools.CS 1
- Explore data-fusion frameworks to develop causality.CS 4
- Explore the role of causality in AI interpretability, fairness, and robustness, as well as AI privacy, safety, and accountability.CS 4 CS 2
- Develop a control framework for trustworthy AI applications.CS 5 CS 1
- Select a research topic on trustworthy causal AI, conduct in-depth research, produce a well-structured research paper, and effectively communicate the findings through a presentation and discussion.CS 7
Policies and Procedures
Please be aware of the following linked policies and procedures. Note that in individual courses instructors will have stipulations specific to their course.