QBoard » Artificial Intelligence & ML » AI and ML - Conceptual » Does changing the order of layers in CNN have any impact?

Does changing the order of layers in CNN have any impact?

  • Can we use the hidden layers in any order in a CNN?
      May 5, 2019 12:25 PM IST
    2
  • Rakesh Racharla said:

    Conventionally, CNN layers downsample over and over, which enables them to capture details at different levels of abstractions. Usually, it is observed that the initial layers do nothing more than detecting edges, or filtering color channels; the combinations of these edges are what we perceive as 'features'.

    If you reverse the order, you essentially are changing sampling modes down the line. CNNs detect by 'downsampling' the inputs and therefore 'extracting' features.

    It may not work as expected!



    Nice explanation
      May 5, 2019 4:09 PM IST
    3
  • Not necessarily always.

    Adding more layers will help you to extract more features. But we can do that upto a certain extent. There is a limit. After that, instead of extracting features, we tend to ‘overfit’ the data. Overfitting can lead to errors in some or the other form like false positives.

    I will give you an example. Suppose we train a model for detecting cats. If all cats features are detected and we add more layers, it can start detecting the bell the cat is wearing as a part of the cat. It may end up classifying a cat without bell as not cat and may sometimes classify another animal with bell as cat. This is because adding more layers beyond a certain threshold leads to finding irregularities in the data. Hope it helps!

      December 22, 2021 1:41 PM IST
    0
  • Conventionally, CNN layers downsample over and over, which enables them to capture details at different levels of abstractions. Usually, it is observed that the initial layers do nothing more than detecting edges, or filtering color channels; the combinations of these edges are what we perceive as 'features'.

    If you reverse the order, you essentially are changing sampling modes down the line. CNNs detect by 'downsampling' the inputs and therefore 'extracting' features.

    It may not work as expected!

      May 5, 2019 12:52 PM IST
    0