Get the latest tech news

Machine Learning Model Homotopy

For a long time I have been thinking on how can one apply the ideal of homotopy to statistics and machine learning. The idea is: for many modeling situations, a single model fit characterize enough…

For a long time I have been thinking on how can one apply the ideal of homotopy to statistics and machine learning. Examples of this in practice include the Lasso showing regularization trade-offs, and also fitting a classification model with different data prevalences. In this example I show in even a linear regression we can expect a fit coefficient to change signs as many times as the number of variables minus 1!

Get the Android app