Exploring the likelihood surface and confidence regions in some phylogenetic Gaussian processes
||Exploring the likelihood surface and confidence regions in some phylogenetic Gaussian processes|
||NAISS Small Compute|
||Hao Chi Kiang <email@example.com>|
||2023-01-12 – 2024-02-01|
In phylogenetic comparative methods, multivariate Gaussian branching processes is a class of useful models to infer about the evolution of traits of species.
Confidence regions that result from approximating the likelihood with Gaussian appear to work only when the phylogeny is big enough. When the phylogeny is small, evolutionary pressure is low, and especially when there are no fossil record, the log-likelihood surface seems to deviate very significantly from quadratic and maximum likelihood is sometimes very far away from the truth.
Therefore, I propose a computation project to further explore the likelihood surface of these models, especially on some realistic phylogenetic trees.