Questions

1. An important operation in evolutionary algorithms is recombination. Which of the following recombination (i.e. crossover) operators we can use for integers?
   1. single arithmetic
   2. subtree swapping
   3. uniform crossover
   4. whole arithmetic

• answers

1. What is (in a nutshell) backpropagation in the context of neural networks?
   1. Gradient descent + chain rule
   2. Propagating through a network back
   3. Neuroevolutionary algorithm
   4. Belief propagation

• answer

1. In convolutional neural networks, using a pooling layer results in:
   1. increasing a size of a feature map
   2. adding zeros to an input
   3. processing input in a recurrent manner
   4. reducing a size of a feature map

• answer

1. Can we train a neural network if it is non-differentiable?
   1. No.
   2. Yes, by using, e.g., an evolutionary algorithm.
   3. Yes, by using, e.g., convex optimization.
   4. Yes, by using, e.g., backpropagation.

• answer

1. Let us assume a neural network with multiple layers and a sigmoid function as an activation function. Such neural network could suffer from a problem called:
   1. vanishing gradient problem
   2. credit assignment problem
   3. vanishing assignment problem
   4. credit gradient problem

• answer

1. Let us assume a random variable $x$ that is distributed by $p(x\mid w)$, where w denotes parameters. We observe $N$ datapoints $\{x_1, ..., x_N \}$, and we assume that all observations are iid. We want to find parameters $w$. How do we call the optimization function in this optimization problem?
   1. the fitness function
   2. the indicator function
   3. the likelihood function
   4. the survival function

• answer

1. Which of the below can describe best the process of visiting the regions in a search space that are not previously visited?
   1. Exploration
   2. Genetic algorithm
   3. Exploitation
   4. Metaheuristic