Convolutional Neural Networks and Image Understanding

This phase checks whether the learner truly understands how neural networks “see” images, instead of treating CNNs as black-box layers.

Section A: Why CNNs Exist (Q1–Q6)

Q1. Why Fully Connected Networks Fail for Images

Question Description:

Explain why using only fully connected (dense) layers for image data is inefficient and impractical. Focus on parameter explosion, spatial information loss, and scalability issues.

Sample Input:

32×32 RGB image

Sample Output:

Fully connected layers create too many parameters and ignore spatial structure.

Q2. What Problem CNNs Solve

Question Description:

Explain the core problem that Convolutional Neural Networks are designed to solve when working with images.

Sample Input:

Image pixel relationships

Sample Output:

CNNs efficiently learn spatial patterns like edges and shapes.