1 paper
Samy Badreddine, Emile van Krieken, Luciano Serafini
Neural networks often map low-dimensional embeddings to high-dimensional output spaces. Usually, the output layer is linear, which can create a "rank bottleneck" that limits the fu…