6 ms·Every Model Learned By Gradient Descent Is Approximately A Kernel Machine (2020)2 points by optimalsolver 4y ago