Dr. Aaron J. Gutknecht

I develop mathematical foundations for information decomposition, including a part-whole account of how information is decomposed, a Boolean-function framework for unifying decomposition approaches, and a differentiable redundancy measure for neural-network training.

I work on statistical methods for directed information flow, including Granger causality and significant subgraph mining for network inference with multiple-comparisons control.

I develop information-theoretic methods for studying artificial neural networks, including Shannon invariants: scalable summaries that track how information is represented and transformed across layers and through training.