tag:blogger.com,1999:blog-36083511.post8833946557066808850..comments2023-04-10T22:27:17.268+08:00Comments on Josh's Blog: Models that Account for the Same DataJosh Gohhttp://www.blogger.com/profile/06726842686303817130noreply@blogger.comBlogger1125tag:blogger.com,1999:blog-36083511.post-3049953335787142512008-05-29T13:06:00.000+08:002008-05-29T13:06:00.000+08:00I'm not sure i understand the question at the end ...I'm not sure i understand the question at the end of this post.<BR/><BR/>Methods like PCA, ICA, LLE don't redefine the data in term of other dimensions but are instead used to reduce the dimensionality of the data or to determine either the statistically reliable or task relevant sub-manifold of the data.<BR/><BR/>This approach is perfectly scientific as science is concerned with the predictable (and the predictably useful). Moreover, these methods have well defined generative models (except possibly LLE). As a result, the model selection (or dimension reduction) process has normative solution within the Bayesian framework, and even has the benefit of automatically implementing Occam's Razor.Anonymousnoreply@blogger.com