Abstract
Convolutional neural networks (CNNs) have been a core element in the recent great advances of deep learning. This chapter introduces the basic concepts related to CNNs. The first part of the chapter focuses on the CNN structure and describes common layer types with the main focus on the convolutional layers. Some of the most famous CNN architectures are also briefly presented. The second part of the chapter focuses on training of CNNs. This part first describes how backpropagation is performed on CNNs and introduces common loss functions and different optimizers. Finally, it discusses typical challenges in CNN training and presents some of the most widely spread solutions to these challenges.
Original language | English |
---|---|
Title of host publication | Deep Learning for Robot Perception and Cognition |
Publisher | Elsevier |
Pages | 35-69 |
Number of pages | 35 |
ISBN (Electronic) | 9780323857871 |
ISBN (Print) | 9780323885720 |
DOIs | |
Publication status | Published - 2022 |
Publication type | A3 Book chapter |
Keywords
- Backpropagation
- CNN structure
- CNN training
- Convolutional layers
- Convolutional neural networks
- Training challenges
Publication forum classification
- Publication forum level 2
ASJC Scopus subject areas
- General Computer Science