What's a difference (in terms of architecture) between the neoperceptron and CNN?
Both ANNs have hidden layers and scanners, as I understood, but many sources subdivide them in two classes.
Data Science Stack Exchange is a question and answer site for Data science professionals, Machine Learning specialists, and those interested in learning more about the field. It only takes a minute to sign up.
Sign up to join this communityWhat's a difference (in terms of architecture) between the neoperceptron and CNN?
Both ANNs have hidden layers and scanners, as I understood, but many sources subdivide them in two classes.
According to the research paper, neoperceptrons are a class of CNN that are not sensitive to rotations.
One of the issues with traditional kernels (that was the case before CNN and it is still true with them) is that the rotation of the input image would lead to different results, because the neurons in the dense layer would have different levels of activations.
With these new neurons, you don't get an issue with orientation. So in theory, if you have a gradient in your image, no matter what the orientation is, you would get the same value.
For a traditional CNN, you would get maximum activation with the original orientation, inverse with a 180° rotated image, and no activation with 90° or 270°.