R>> Keras_ODENet>> 返回
项目作者： jason71995

项目描述 ：
Implementation of (2018) Neural Ordinary Differential Equations on Keras
高级语言： Python
项目地址: git://github.com/jason71995/Keras_ODENet.git
创建时间： 2019-01-17T03:33:01Z
项目社区：https://github.com/jason71995/Keras_ODENet
开源协议：
下载


Neural Ordinary Differential Equations in Keras

Introduction

Implementation of (2018) Neural Ordinary Differential Equations.

Attention

ODE solver are use tf.contrib.integrate.odeint which only supported “dopri5” method now.

Environment

GPU: Nvidia GTX 670

  1. python==3.6
  2. tensorflow==1.4.0
  3. keras==2.1.0

Result

Result on 10 epochs

MNIST ODENet

training time: 730s

train_loss: 0.0112 - train_acc: 0.9962 - val_loss: 0.0234 - val_acc: 0.9929

MNIST ResNet

training time: 120s

train_loss: 0.0096 - train_acc: 0.9968 - val_loss: 0.0307 - val_acc: 0.9908