CS 412 Week 6

Suppose we are now trying to build a classifier but we only have a small amount of labeled data. Which of the following actions can we take to improve our model?

A) Use a model trained on a similar task and apply transfer learning.

B) Co-train two classifiers and use them to label the data for each other.

C) Use expected label distributions as weak supervision.

D) Use existing resources such as knowledge bases to provide weak supervision.

E) All the above

E) All the above

Which of the following statements are correct about transfer learning?

A) No annotated training data is available for source tasks.

B) No annotated training data is available for target tasks.

C) There can be multiple source tasks.

D) There can be multiple target tasks.

C) There can be multiple source tasks.

D) There can be multiple target tasks.

What is a strength of SVM?

A) The prediction accuracy of SVM is generally high.

B) SVM for low dimensional data can have good generalization, even when the number of support vectors is large.

C) SVM is scalable to the number of data objects in terms of memory usage.

D) The training time a SVM classifier is short

A) The prediction accuracy of SVM is generally high.

Which of the following statements about SVM is correct?

A) SVM can perform a non-linear classification using kernel functions.

B) SVM is a classification method only for linear data.

C) SVM is a generative classifier.

D) SVM is an unsupervised learning algorithm

A) SVM can perform a non-linear classification using kernel functions.

Which of the following statements are correct about active learning?

A) We can have human annotators in active learning.

B) When evaluating active learning algorithms, besides classification accuracy, we need to consider the amount of labeled data that an algorithm uses.

C) After each iteration, the size of the labeled training set becomes larger, but the size of the unlabeled pool does not change.

D) We only have access to limited unlabeled data in active learning

A) We can have human annotators in active learning.

B) When evaluating active learning algorithms, besides classification accuracy, we need to consider the amount of labeled data that an algorithm uses.

Which of the following statements about SVM is correct?

A) SVM is a classification method only for linear data.

B) SVM is a discriminative classifier.

C) SVM is an unsupervised learning algorithm.

D) The goal of SVM is to find the optimal separating hyperplane which minimizes the margin of training data

B) SVM is a discriminative classifier.