Embedded Files
Gade, N. D. and J. Rodu. (2023+). Change Point Detection with Conceptors. preprint arXiv:2308.06213.
For the at most one change point problem with arbitrary temporal dependence, echo state networks serve as advantageous functional transformations to a higher-dimensional domain. A conceptor matrix captures the dynamic relationships of the network states and quantifies a distance away from a labelled "baseline" state.
Gade, N. D. and J. Rodu. (2023+). Nonlinear Permuted Granger Causality. preprint arXiv:2308.06220.
Granger causal inference with in-sample tests, like penalized optimization techniques, can inflate dependence on covariates of interest and lead to spurious relationships. We advocate for a switch to permutation testing in nonlinear Granger causality, and illustrate using simple, single-layer MLPs as functional transformations to capture the nonlinear dependence structure.
Page updated
Google Sites
Report abuse