What is the main difference between supervised and unsupervised learning?
2
What is the main purpose of an autoencoder in the context of unsupervised anomaly detection?
3
How can an autoencoder detect an anomaly when trained only on images of a single digit (e.g., the digit 5) in the MNIST dataset?
4
Which loss function is used to train an autoencoder on MNIST images to quantify reconstruction quality?
5
What is the role of the Sigmoid layer in the decoder of a simple autoencoder used to reconstruct grayscale images normalized between 0 and 1, such as those in the MNIST dataset?
6
In the context of a 'denoising autoencoder,' what is the main difference between the input and output data during training?
7
Why is a convolutional autoencoder used for denoising on MNIST rather than a fully connected autoencoder?
8
What is the impact of the noise level (variance) added to images on the denoising task?
9
What role does the BatchNorm function play in the layers of the convolutional autoencoder used for denoising?
10
What is a potential advantage of testing the U-Net architecture for the denoising task compared to a classic autoencoder?