Open AccessCONSTRUCTION OF NEURAL ARCHITECTURES WITH DESIRED BEHAVIOUR UNDER GEOMETRIC TRANSFORMATIONS OF THE INPUT
Author(s) -
Viacheslav Dudar,
В. В. Семенов
Publication year - 2020
Publication title -
žurnal občislûvalʹnoï ta prikladnoï matematiki
Language(s) - English
Resource type - Journals
eISSN - 2706-9699
pISSN - 2706-9680
DOI - 10.17721/2706-9699.2020.1.03
Subject(s) - convolutional neural network , geometric transformation , transformation (genetics) , transformation geometry , computer science , pixel , geometric pattern , algorithm , geometric modeling , simple (philosophy) , geometric shape , artificial intelligence , mathematics , image (mathematics) , geometry , biochemistry , gene , philosophy , epistemology , chemistry
We present a general method for analysis of convolutional layers under geometric transformations of the input that are linear with respect to pixel values. We also describe the algorithm for finding all possible types of behaviours of the output of convolutional layers under geometric transformations of the input. We also present a general method for construction of convolutional architectures with desired behaviour under geometric transformations of the input.