In a recent paper, Ling et al. investigated the over-parametrized Deep Equilibrium Model (DEQ) with ReLU activation. They proved that the gradient descent converges to a globally optimal solution for the quadratic loss function at a linear convergence rate. This paper shows that this fact still holds for DEQs with any generally bounded activation with bounded first and second derivatives. Since the new activation function is generally non-homogeneous, bounding the least eigenvalue of the Gram matrix of the equilibrium point is particularly challenging. To accomplish this task, we must create a novel population Gram matrix and develop a new form of dual activation with Hermite polynomial expansion.
翻译:暂无翻译