04 02 21

Page 1

4/2/2021 2021.02.04 Machine Learning GES https://forms.office.com/Pages/ResponsePage.aspx?id=K3EXCvNtXUKAjjCd8ope66_XbeATVphMnuFlOS05Oj1UNVY3OUZGNVZGTUdKOEtVS… 1/6 2021.02.04 Machine Learning GES  Ciao , quando invierai il modulo, il proprietario potrà vedere il tuo nome e l'indirizzo di posta elettronica. 1 Consider the neural network in the figure which takes as input two variables x1 , x2. Which ones among the following answers is correct? The optimization problem associated to the model attempts to find the values of x1 and x2 in order to minimize the miss-classification error. If the activation function g is the identity function then the model is certainly equivalent to a linear model. If the network is used in a regression problem, activation function g is likely the sigmoid function. The model can not be used in a classification task since the output layer contains a single neuron. None of the other answers are correct. The regularization strength of the model is associated to the value in the top left neuron (in the example +1). 4/2/2021 2021.02.04 Machine Learning GES https://forms.office.com/Pages/ResponsePage.aspx?id=K3EXCvNtXUKAjjCd8ope66_XbeATVphMnuFlOS05Oj1UNVY3OUZGNVZGTUdKOEtVS… 2/6 2 R eferring to regression, which ones among the following answers is correct? Heteroscedasticity implies in any case model rejection. If a model is multicollinear then all the variance inflation factors are greater than 5. None of the other answers are correct. Heteroscedasticity can be corrected by appropriate transformations. Cook's distance is used to establish normality of the residuals. Cook's distance can be used to assess anomalies in regression coefficients. 3 R eferring to classification, which ones among the following answers is correct? The V C dimension depends…