Stochastic Chebyshev Gradient Descent for Spectral Optimization

Unfortunately, computing the gradient of a spectral function is generally of cubic complexity, as such gradient descent methods are rather expensive for optimizing objectives involving the spectral function.