From Information to Generative Exponent: Learning Rate Induces Phase Transitions in SGD

Open in new window