1 paper · 1 filter
Enes Arda, Semih Cayci, Atilla Eryilmaz
Understanding why Transformers perform so well remains challenging due to their non-convex optimization landscape. In this work, we analyze a shallow Transformer with m independe…