Why Stochastic Gradient Descent (SGD) is Outdated and How to Fix it with Kronecker Sequences
中文摘要
使用 Kronecker 序列替代 SGD 并通过优化 PyTorch 数据读取方式，可将深度学习训练成本降低一半。
English Summary
Replacing SGD with Kronecker Sequences can halve deep learning training costs by mathematically shifting how PyTorch reads data.
原文节选
How a simple mathematical shift in the way PyTorch reads data can cut your Deep Learning training costs in half. Continue reading on Medium »