PIDOptimizer
PIDOptimizer copied to clipboard
tensorboy
Code for this CVPR 2018 paper: "A PID Controller Approach for Stochastic Optimization of Deep Networks", Wangpeng An, Haoqian Wang, Qingyun Sun, Jun Xu, Qionghai Dai, Lei Zhang.
Hi, I have a question about the I & D values. Obviously they correspond to Ki and Kd respectively. But according to your paper, Kp and Ki should be set...
the E(s)
hello,i have some questions in your paper.1 why does the Laplace Transform of θ∗ is θ∗/s. 2 Why does E(s) = θ∗s −theta(s).Doesn't the e(t) is ∂Lt/∂θt
What is the learning rate of training cifar10 and cifa100 data sets in this article?
The documentation for SGD left by pytorch author should be changed to a PID version.
I wanted to understand the role of I, and D terms in PIDOptimizer. Do these signify the coefficients of integral and derivative term described in the paper? Is I coefficient...
Eq. (14) in the paper may be wrong, the last term should be multiplied by 1/4
