Contrast CNNs and fully connected networks

Compare convolutional neural networks (CNNs) and fully connected (dense) networks. Explain:

• How convolutional layers differ structurally from fully connected layers (parameter sharing, locality, connectivity pattern).
• The reasons CNNs are more efficient for certain kinds of data.
• Which data or tasks fit CNNs best, and when a fully connected design is the better choice. Overview: This question tests knowledge of neural network architectures by asking the candidate to contrast CNNs and fully connected (dense) networks, with attention to parameter sharing, locality, and connectivity patterns in machine learning.