19.3 Advantages of using CNNs Over Traditional Neural Networks.

Traditional neural networks process input data in a flattened form, losing spatial structure and requiring an excessive number of parameters for high-dimensional data like images. CNNs overcome these limitations through:

• Local Connectivity: Neurons in convolutional layers connect only to small regions of the input, reducing parameters while preserving spatial relationships.

• Weight Sharing: The same filter is applied across the entire image (instead of having one weight per input feature (as in FFNNs), CNNs use one filter across many locations), drastically cutting down trainable weights and significantly reduces the number of parameters, allowing CNNs to scale to large input sizes like 224x224 images, which would be computationally intractable with fully connected layers.

• Hierarchical Feature Learning: Early layers detect simple features (edges, textures), while deeper layers recognize complex patterns (shapes, objects).

This makes CNNs far more efficient for image-related tasks compared to fully connected networks.