Dynamics of Deep Learning with Gaussian Mixture Inputs: Revealing Unexpected Universality

Despite the complex and varied nature of Gaussian mixture distributions, neural networks exhibit asymptotic behaviors in line with predictions under simple Gaussian frameworks, particularly when inputs are standardized.