I was reading an article where they have a set of pages. For every page they extract a feature vector of dimention 300. Then they fuse these vectors into one vector by using a method called average pooling: Quoting : " to encode sets of pages into a compact document representation. The main idea consists of performing average pooling of page representations to obtain a single representation of an entire document"

P is the set of pages P = {p1,p2,p3,...,pn} and pi is the vector

F(p) = 1/|P| * Somme(pi) Such that p E P



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.