VL Hands-on AI I Exam 1 and 2 WS2020

# VL Hands-on AI I Exam 1 and 2 WS2020 ###### tags: `Exam` `Hands-on AI` `2020S` ## EXAM 1 **Q:** Affinity Propagation ... * ... does not need a fixed number of cluster centers as input. * ... is a clustering method. --- **Q:** A polynomial of degree one ... * ... is a straight line. * ... has two parameters to be adjusted. --- **Q:** The logistic function "y = sigma(x)" ... * ... is 0.5 for "x = 0". * ... has output values "y" between 0 and +1. --- **Q:** Principal Component Analysis ... * ... enables visualization of high-dimensional data. * ... is a dimensionality reduction method. --- **Q:** Which of the following is a meaningful data augmentation? * Rotating an image by 270 degrees. * Rotating an image by 90 degrees. * Rotating an image by 180 degrees. --- **Q:** Which of the following statements is/are true about the test set method? * The test set is used to estimate the risk. * The underlying assumption is that samples are identically and independently distributed. * Empirical risk minimization is performed on the training set. --- **Q:** Which of the following statements is/are true about an 8-bit RGB image? * Every pixel is represented by 3 channels. * Every channel can encode 2^8 different values. * Every channel can encode 256 different values. --- **Q:** In which format comes the Iris dataset? * Tabular --- **Q:** What of the following is true about a regression task? * The mean-squared error is a suitable loss function. * The target is a numerical value. --- **Q:** What are strengths of frameworks like PyTorch? * Easy switching of computations between CPU and GPU. * Automatic differentiation. * Straightforward construction of neural networks. --- **Q:** What is typical for a supervised machine learning task? * Learning a mapping from input to target values. * Learning with knowing the input and target values. --- **Q:** Which of the following statements is/are true about under- and overfitting? * If you run into underfitting, the model has most probably problems to fit the training data. * If you run into underfitting, the model complexity is probably too low. * If you run into overfitting, the model has most probably fitted the training data pretty well. * If you run into overfitting, the model complexity is probably too high. --- **Q:** Which of the following statements is/are true about the generalization error? * It is straightforward to calculate, if the distribution of future, unseen data is known. * It is defined as the error on on future, unseen data. --- **Q:** Which of the following statements is/are true about a greyscale 16-bit image? * Every channel can encode 2^16 different values. --- **Q:** A linear regression model ... * ... is e.g. a polynomial of degree 0. * ... is e.g. a polynomial of degree 8. * ... has a closed-form solution --- **Q:** What does a Fourier transform of a sound signal do? * It decomposes the signal into its constituent frequencies. --- **Q:** For a logistic regression classifier adequately trained on the training set, color-inverting the corresponding test set ... * decreases the performance to random accuracy. * decreases the performance as these kind of images were not seen during training. --- **Q:** How many peaks are visible in a Fourier spectrum of a sine wave of 448 Hertz? * Only one. --- **Q:** The bias-variance trade-off is closely related to ... * ... training and test sets. * ... over- and underfitting. * ... empirical risk minimization. --- **Q:** For a logistic regression classifier adequately trained on the MNIST training set, horizontally flipping the corresponding test set ... * ... decreases the performance, while still being able to distinguish some of the classes correctly. * ... decreases the performance as these kind of images were not seen during training. --- **Q:** For a logistic regression classifier adequately trained on the MNIST training set, color-inverting the corresponding test set ... * ... decreases the performance as these kind of images were not seen during training. * ... decreases the performance to random accuracy. --- **Q:** What is a hyperparameter in the k-nearest-neighbor classification algorithm? * The number of nearest neighbors. ___ ## EXAM 2 **Q:** Which of the following modules is normally not found in a fully-connected neural network: * A convolutional network layer. * A max-pooling layer. ___ **Q:** Which of the following statements is/are true about a 2 times 2 max-pooling layer in a convolutional neural network? * It reduces the spatial size of the representation. * It takes the maximum value of 2 times 2 input values. * It is a form of non-linear downsampling. --- **Q:** Dropout ... * may increase the validation performance ___ **Q:** Downsampling of an input image may be achieved by ... * ... a 3 times 1 mean-pooling. ___ **Q:** Which of the following statements is/are true about the ReLU activation function? * ReLU leaves all positive values unchanged. * ReLU sets all negative values to zero. ___ **Q:** Which of the following statements is/are true about a vanishing gradient? * The error signal backpropagated through the network vanishes. ___ **Q:** The Sobel operator ... * ... has two variants, one for horizontal and one for vertical edges. * ... may enable easy recognition of edges in images. * ... is a crude approximation of the gradient of the image. ___ **Q:** A non-convex function ... * ... usually occurs when training neural networks. * ... has at least one global minimum but potentially several local minima. ___ **Q:** Which of the following statements is/are true about convolutions? * A convolutional layer in a neural network involves a kernel. * A convolution is a mathematical operation on two tensors * A convolutional layer is often followed by an activation function and a pooling layer. ___ **Q:** Which of the following statements is/are true about the PyTorch snippet "nn.Linear(4, 4)"? * It defines a linear neural network layer. ___ **Q:** Which of the following statements is/are true about the concept of "strides" in convolutional networks? * The stride specifies the amount of pixels by which a filter is shifted. * The stride influences the output size of a convolutional layer. ___ **Q:** Which of the following statements is/are true about the cross-entropy error? * The cross-entropy is an error measure between model predictions and the corresponding target values. * The cross-entropy error is a suitable loss function for classification tasks. ___ **Q:** Which of the following statements is/are true about the PyTorch snippet "nn.Conv2d(3, 2, 3)"? * It specifies 3 input and 2 output channels. * It defines a convolutional neural network layer. ___ **Q:** Which of the following statements is/are true about the mean-squared error? * The mean-squared error is a suitable loss function for regression tasks. * The mean-squared error is the mean of the squared differences between model predictions and the corresponding target values. ___ **Q:** Which of the following statements is/are true about gradient descent? * Gradient descent is an optimization algorithm. * Gradient descent is not guaranteed to find the global minimum. * Gradient descent is commonly used when training neural networks. * Gradient descent takes steps in the direction of the negative gradient of the function to minimize. ___ **Q:** Which of the following statements is/are true about the PyTorch module "nn.Sequential"? * It may be used to group PyTorch modules in a sequential manner. ___ **Q:** Residual connections ... * allow to create deeper neural networks, while maintaining trainability. * create shortcuts for gradients. ___ **Q:** Which of the following statements is/are true about confusion matrices? * Confusion matrices show which classes the underlying model has confused with each other. * Confusion matrices ease the comparison of different classifiers. ___ **Q:** Which of the following statements is/are true? * Dropout during training may reduce overfitting. * Data augmentation during training may reduce overfitting. ___ **Q:** The PyTorch snippet "torch.tensor([[1., -1.], [-1., 1.]])" ... * ... returns a tensor of size 2 times 2. * ... may be an appropriate input to a neural network in PyTorch. ___