Fetching the paper…

Eigenvalue Decay Implies Polynomial-Time Learnability for Neural Networks · Around