QBoard » Artificial Intelligence & ML » AI and ML - Conceptual » Are there any methods to find the kernel size to use in a neural network?
User Dashboard

Are there any methods to find the kernel size to use in a neural network?

Back To Topics
Tags : neural_network kernal_size
  • 32 1
    Should we follow a trial and error approach, or are there any methods or thumb rules we can use to find the right kernel size
      January 9, 2021 4:33 PM IST
    0
  • 84

    I have checked a lot of questions here and in other websites. What I concluded is that there is no rules for choosing the right number of hyper-parameters in CNN, all what can we do is just trying many cases, and select the best one. Thus, I still wondering if we can choose a preliminary number of those hyper-parameters (point to start) based on the type of input images (histological medical images in my case), the size and the number of input data, the architecture of the network, etc.

      August 21, 2021 6:58 PM IST
    0
  • 13
    Commonly 3x3 kernel is used, Kernel size doese not effect accuracy much, so we can decide the kernel size by considering the input size, if the input is very large kernel size with larger size can be used inorder to reduce the computation complexity
      August 5, 2021 9:54 PM IST
    0
  • 126
    A common choice is to keep the kernel size at 3x3 or 5x5. The first convolutional layer is often kept larger. Its size is less important as there is only one first layer, and it has fewer input channels: 3, 1 by color
      August 7, 2021 7:00 PM IST
    0
  • 102 9
    This problem is similar to choosing the other hyperparameters of a model architecture like the number of layers, activation functions, etc. 

    There are no methods/rules to find them easily. One of the approaches is to check the kernel sizes used in popular models like VGG16, Resnet, etc.. and fine-tune them to improve the performance.
      January 9, 2021 4:54 PM IST
    0