What does a Convolutional Neural Network (CNN) do during the training process?

• A. It adjusts filters to optimize image recognition
• B. It increases the dataset size
• C. It normalizes data for better accuracy
• D. It generates new images

Answer: (A)

A Convolutional Neural Network (CNN) operates through a training process that involves adjusting filters, also known as kernels. These filters are crucial since they allow the CNN to learn and identify patterns, features, and hierarchical representations in images. During training, the CNN takes input images, processes them through various layers that include convolutional layers where these filters are applied, and then adjusts the parameters of the filters based on the errors calculated from the output compared to the actual labels. This optimization is typically done using techniques such as backpropagation and gradient descent, allowing the network to improve its image recognition capabilities over time.

In contrast, increasing the dataset size, normalizing data, and generating new images are related tasks but do not directly characterize the core function of a CNN during the training. While data augmentation techniques can artificially expand the dataset size, normalization can help improve training stability and performance, and generative models can create new images, these processes do not encapsulate the primary role of a CNN in learning and optimizing filters for image recognition tasks.