DLIP ROZPIERDALATOR 3000

Callbacks in Keras are used to execute code after every epoch. True. False.

Conv2D layer accepts three dimensional data. True. False.

Adversarial attacks change weights of an attacked network. True. False.

Convolutional Neural Networks (CNN) work on sequences of samples. True. False.

Recurrent Neural Networks (RNNs) are good at text processing tasks. True. False.

When Cohen's Kappa returns zero it means that the result is the same as random. True. False.

Accuracy measure is the number of correctly classified samples to the number of all samples. True. False.

Autoencoder is trained to return the same data as it was given on its input. True. False.

The MSE measure may be used instead of accuracy and typically gives better results. True. False.

When the model works very good for training data and it is poor for test data it means that we have a problem with: overfitting. mode collapse. convergence failure.

The RELU activation function returns a value in range <0,1>. True. False.

Assuming that the list lst[] contains 100 elements, how many elements will be returned by the command: lst[1:4].

Generative Adversarial Networks (GANs) may be used to create images. True. False.

The SIGMOID activation function returns a value in range <0,1>. True. False.

The TANH activation function returns a value in range <-1,1>. True. False.

The output from the GAN generator network is one value. True. False.

Recurrent Neural Networks (RNNs) preserve state between processing samples. True. False.

Generative Adversarial Networks (GANs) consist of two competing networks. True. False.

Augmentation is the algorithm that changes the number of layers in a network. True. False.

Point the layers used in Recurrent Neural Networks: GRU. LSTM. UNET. EDSR.

Which of these classes is NOT a layer in Keras/Tensorflow: Dense. Conv2D. BatchNormalization. Numpy.

The Python package that handles multidimensional arrays is named: pandas. numpy. thorpy. math.

When we use an existing network and retrain it with new samples it is called transfer learning. True. False.

Recurrent Neural Networks (RNNs) consist of two competing networks. True. False.

GradientTape is the specific network layer that calculates gradients. True. False.

Every neural network is convolutional. True. False.

LeakyRELU activation function returns a value in range <-infinity,infinity>. True. False.

Convolutional layers have typically less weights than dense layers. True. False.

Lower learning rate makes the model learn faster. True. False.

Convolutional Neural Networks (CNN) are very good at classification of images. True. False.

CycleGAN consists of four networks. True. False.

For the perfect model the confusion matrix has positive values only on its diagonal. True. False.

Recurrent Neural Networks (RNNs) work on sequences of samples. True. False.

Mark the loss function that may be used for regression tasks. binary cross entropy. mean absolute error. categorical cross entropy.