machida-mn / tensorflow-pid

Tensorflow implementation of PID optimization

PID optimizer for Tensorflow

Tensorflow implementation of PID optimization (CVPR 2018).

See the following paper for detail of algorithm.

• A PID Controller Approach for Stochastic Optimization of Deep Networks

This program was tested on Tensorflow r1.4.

Installation

$ git clone https://github.com/machida-mn/tensorflow-pid
$ cd tensorflow-pid
$ pip install .

Usage

import tensorflow_pid

loss = .......

train_op = tensorflow_pid.PIDOptimizer(learning_rate=0.01, kd=0.001).minimize(loss)

Implementation in other frameworks

• PyTorch
  • PIDOptimizer (GitHub)