CS7641 MIDTERM | VERIFIED STUDY GUIDE
inductive bias - Answers - set of assumptions about hypotheses as they relate to the
data
restriction bias - Answers - restrict the set of hypotheses considered
preference bias - Answers - prefer certain hypotheses over others
overfitting - Answers - fitting a model too closely to the training data; not generalizing
joint distribution - Answers - the distribution of probabilities over the cross product of two
or more variables
estimation bias - Answers - bias resulting from testing the accuracy of a model on
training examples
estimation variance - Answers - variance from true accuracy based on the makeup of
test examples
perceptron - Answers - basic unit of an ANN
weak learner - Answers - a learner that does better than chance
concept - Answers - a function that maps input to output
randomized optimization - Answers - optimization methods that can be used on
functions that are not differentiable
version space - Answers - the set of hypotheses that is consistent with the training data
consistent learner - Answers - an algorithm that outputs a hypothesis in the version
space
ensemble learning - Answers - combines simple rules to form a complex rule
early stopping - Answers - a method used by ANNs to avoid overfitting by halting
training when the test error increases
pre-pruning - Answers - stops tree growth before overfitting occurs
post-pruning - Answers - allows overfitting to occur before pruning the tree for
generalization
inductive bias - Answers - set of assumptions about hypotheses as they relate to the
data
restriction bias - Answers - restrict the set of hypotheses considered
preference bias - Answers - prefer certain hypotheses over others
overfitting - Answers - fitting a model too closely to the training data; not generalizing
joint distribution - Answers - the distribution of probabilities over the cross product of two
or more variables
estimation bias - Answers - bias resulting from testing the accuracy of a model on
training examples
estimation variance - Answers - variance from true accuracy based on the makeup of
test examples
perceptron - Answers - basic unit of an ANN
weak learner - Answers - a learner that does better than chance
concept - Answers - a function that maps input to output
randomized optimization - Answers - optimization methods that can be used on
functions that are not differentiable
version space - Answers - the set of hypotheses that is consistent with the training data
consistent learner - Answers - an algorithm that outputs a hypothesis in the version
space
ensemble learning - Answers - combines simple rules to form a complex rule
early stopping - Answers - a method used by ANNs to avoid overfitting by halting
training when the test error increases
pre-pruning - Answers - stops tree growth before overfitting occurs
post-pruning - Answers - allows overfitting to occur before pruning the tree for
generalization
