BriefGPT.xyz
Jul, 2021
双重加速快速实现边缘最大化
Fast Margin Maximization via Dual Acceleration
HTML
PDF
Ziwei Ji, Nathan Srebro, Matus Telgarsky
TL;DR
介绍了一种基于动量的梯度方法，使用最大化分类间隔相结合的凸对偶来训练线性分类器，表现出优异的性能
Abstract
We present and analyze a
momentum-based gradient method
for training
linear classifiers
with an
exponentially-tailed loss
(e.g., the expon
→