Choose the correct statement

Choose the correct statement

The empirical risk is more important than $($population$)$ risk, because it is defined in the sense of finite samples and can be evaluated in practice. In contrast, the risk can never be evaluated in practice, so it is meaningless to consider it in the real$-$world machine learning problems.

Risk $($or generalization error$)$ is the essential performance measure of a machine learning model

A PAC learning algorithm guarantees learning a machine learning model that approximates both deterministic and stochastic underlying concept function.

Bayes error is the best $($smallest$)$ possible risk that we can achieve using a pre$-$defined hypothesis class such as a linear function hypothesis class.

Question at position $9$

$9$

$1$ point

Question at position $9$

Choose the correct statement

Choose the correct statement

Logistic regression and Softmax classifier can be used for binary and multiclass classification problems, respectively.

KL divergence is equal to the summation of entropy and cross entropy

Hinge loss is not a good choice of loss function since it is concave

Maximizing log$-$likelihood function is an approach to building a machine learning model and it is unrelated to the empirical risk minimization paradigm.

Question at position $10$

$10$

$1$ point

Question at position $10$

MLE principle can only be used to learn a model that approximates the Bernoulli distribution.

MLE principle can only be used to learn a model that approximates the Bernoulli distribution.

True

False