18-753 Information Theory Measures for Artificial and Natural Intelligence Systems
12 units, Carnegie Mellon University
We will discuss how, through examples, counterexamples, thought experiments, and desirable properties, one can systematically arrive at new measures of information relevant in Fairness, Accountability, Transparency, and Explainability FATE of machine learning. We will use data and examples from the real world to explore and understand these issues. The focus will be on defining measures classic and novel as well as the interplay between causality and information that is increasingly important when deciding whether and how to intervene on a decisionmaking system. The course introduces the measures of entropy, mutual information, partial information decomposition, causal inference, and their operational meaning. We will also discuss examples of how causal inference and existing measures come together. The course will discuss FATE of ML, information flow and interventions in neuroengineering and neuroscience, and making inferences in computationaldynamical systems e.g. those in AI and neuroscience, and will engage with handson data analyses and real world examples.
Prerequisites: 36-219
Courseweave needs JavaScript to show sections, workloads and planning tools.