Evaluate the impact of an AI application used in the real world. (case study) : Working with Flower Images: Case Study -

The video tutorial emphasizes the importance of normalization in image processing, particularly in smoothing the training process. It explains how data points with varying ranges can affect gradients during backpropagation, especially when using activation functions like sigmoid. The tutorial also addresses the issue of outliers and presents different normalization techniques, such as using percentiles and standard deviation. It highlights the need for appropriate normalization methods for different types of images, like natural and X-ray images, to ensure effective training outcomes.

Why is normalization important in image processing?

• It eliminates the need for activation functions.

• It makes the training process smoother.

• It increases the size of the dataset.

• It decreases the computational cost.

What happens to the gradients when using sigmoid activation with outliers?

• Gradients drop to almost zero.

• Gradients remain unaffected.

• Gradients become very large.

• Gradients become negative.

Which of the following is a common practice in normalization?

• Dividing by 255

• Multiplying by 255

• Subtracting 255

• Adding 255

What is a limitation of dividing by 255 in normalization?

• It increases the range of data.

• It is not suitable for images with outliers.

• It works only for grayscale images.

• It is only applicable to color images.

Which technique is suggested for handling outliers in normalization?

• Using the mode value

• Using the median value

• Using percentiles

• Using the mean value

Why might dividing by 255 not be suitable for X-ray images?

• X-ray images have a different range of values.

• X-ray images do not require normalization.

• X-ray images are always in color.

• X-ray images are always grayscale.

What is the benefit of using standard deviation in normalization?

• It reduces the size of the dataset.

• It helps in handling outliers effectively.

• It increases the computational cost.

• It eliminates the need for activation functions.