Fundamentals of Neural Networks - Lab 3 - Deep CNN

The video tutorial covers the construction and evaluation of a deep convolutional neural network (CNN) using TensorFlow. It begins with an introduction to the CNN architecture and the Fashion MNIST dataset, which consists of 70,000 grayscale images across 10 categories. The tutorial then explains data preprocessing, including rescaling and visualization. The CNN model is built using TensorFlow's sequential API, and the importance of using the functional API for flexibility is highlighted. The model is trained and evaluated, achieving a 92% accuracy on the training set and 90% on the test set. The tutorial concludes with a discussion on prediction accuracy and potential research questions.

What is the primary goal when using a convolutional neural network with the Fashion MNIST dataset?

• To generate high-resolution images

• To classify images into one of ten categories

• To reduce the size of the dataset

• To convert images to color

How many categories are present in the Fashion MNIST dataset?

• 5

• 10

• 15

• 20

What is the resolution of each image in the Fashion MNIST dataset?

• 128 by 128

• 32 by 32

• 64 by 64

• 28 by 28

Why is it important to rescale the numerical inputs of the dataset?

• To increase the image size

• To prevent exploding gradients

• To add color to the images

• To reduce the number of images

What is the advantage of using the Functional API over the Sequential API in TensorFlow?

• It is easier to write

• It allows for more flexible model architectures

• It requires less computational power

• It automatically optimizes the model

Which optimizer is used in the CNN model discussed in the video?

• RMSprop

• SGD

• Adam

• Adagrad

What is the achieved training accuracy after 5 epochs?

• 92%

• 90%

• 85%

• 95%

What method is used to determine the predicted class from the model's output?

• Mean

• ArgMin

• ArgMax

• Sum

What does a blue bar indicate in the prediction visualization?

• The prediction is incorrect

• The prediction is correct

• The image is unclear

• The model is overfitting

What is the typical relationship between training and test accuracy?

• Test accuracy is always higher

• Test accuracy is usually slightly lower

• Training accuracy is always lower

• They are always equal