Every Model Learned by Gradient Descent Is Approximately a Kernel Machine

13 points| mathematically | 4 years ago |medium.com