Sigmoid function doesn't work with MNIST dataset.

10 次查看（过去 30 天）
latios 93
latios 93 2017-8-5
评论： Srikanth Sedimbi 2017-8-8
Hi everybody, my net have to recognise if an image from the MNIST database represents the digit I choose.
An image is a vector of 784 values between 0 and 1. I can't perform the forward propagation because the sigmoid function returns "1" everytime.
All net dimensions are dynamic, but for debugging purposes I'm using a 2 layer net with 2 nodes on the first layer and 1 output node (I only need to see the output changing, but it doesn't!); I already tried with more layers and more nodes but I still have this problen. In the hidden layer the activation function is the sigmoid; in the output layer it is the identity function.
I tried to preprocess the input (scale from a 28x28 to a 14x14 img) but it didn't work.
Thank you in advance for your help!
  1 个评论
Srikanth Sedimbi
Srikanth Sedimbi 2017-8-8
Are you using any of the MATLAB inbuilt functions for sigmoid or did you write your own sigmoid function? Just post the code for sigmoid function and the corresponding inputs and outputs

请先登录，再进行评论。

请先登录，再回答此问题。

回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Deep Learning Toolbox 的更多信息

标签

产品

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by