What is a bi-affine classifier?

Recently, doing research on NLP papers, I came across with affine classifiers and biaffine classifiers. What is meant by these terms? What are affine classifiers and what are not?

Solution 1:

Ok, so I found an answer. What the papers meant with biaffine classifier is as follows. In neural networks, we have the usual transformation like Ax+b, where A is a matrix, or more conventionally Wx+b, where W is a weight matrix, x is an input vector and b is a bias.

So this Wx+b transformation is an affine transformation, while if we apply another transformation to this namely W(Wx+b)+b, then this is an affine transformation applied twice. Hence, biaffine.