why we use co-variance in machine learning

Question

I cannot understand why co-variance is use in machine learning.is it to know what is the relationship between two features

Answer 1

Covariance is a measure of how much two variables are related. Machine learning is based on detecting patterns between your input variables and output variables. If you have variables that have a very low covariance it will imply that you will most likely need many of the variables to represent the mapping from the input to the output variables. If you normalize the covariance matrix you will obtain the correlation matrix. If all your variables correlate perfectly this would imply that your inputs all measure the same. Hence, you could only use one input variable (data reduction). At the same time if your input does not show any covariance with the output variables it is very unlikely that there is a stochastic dependence between your input variables and output variables. With this kind of data, it will be very hard to get a meaningful mapping from the inputs to the outputs because they are not related or to put it, in other words, there is no pattern to be discovered by the machine learning algorithms.