Chapter 5

Question 1

Before Machine Learning, we program ______ to recognize an apple given images of apple.

Answer 1

logic/algorithm

Question 2

With Machine Learning, given new images of apple, ______ make the prediction.

Answer 2

Machine Learning Models

Question 3

In Supervised Learning, we build a machine that can extract ______ between variables.

Answer 3

relationships

Question 4

Lazy learning algorithms generalize only when ______, rather than during the training phase.

Answer 4

queried

Question 5

During the training phase of lazy learning, the algorithm accepts the data as input but refrains from ______ on it.

Answer 5

actively training

Question 6

In lazy learning, the actual model training occurs during the ______.

Answer 6

prediction phase

Question 7

An example of a lazy learning algorithm is the ______ algorithm.

Answer 7

K-nearest neighbors (KNN)

Question 8

Eager learning algorithms process data during the ______ phase.

Answer 8

training

Question 9

Examples of eager learning algorithms include Linear Regression, Logistic Regression, Support Vector Machines, Decision Trees, and ______.

Answer 9

Artificial Neural Networks

Question 10

Instance-based learning involves using the ______ to make predictions.

Answer 10

entire dataset

Question 11

In instance-based learning, the machine compares the new data to the instances it has seen before and uses the ______ to make a prediction.

Answer 11

closest match

Question 12

In instance-based learning, ______ is created; instead, the machine stores all of the training data.

Answer 12

no model

Question 13

Instance-based learning is often used in pattern recognition, ______, and anomaly detection.

Answer 13

clustering

Question 14

Model-based learning involves creating a ______ that can predict outcomes based on input data.

Answer 14

mathematical model

Question 15

In model-based learning, the model can be thought of as a set of ______ that the machine uses to make predictions.

Answer 15

rules

Question 16

Parametric models make specific ______ about the relationship between input and output data.

Answer 16

hypotheses

Question 17

In parametric models, assumptions concern a ______ of parameters and variables that impact the model’s result.

Answer 17

fixed number

Question 18

Parametric models are easier to understand as they build a certain ______ about the data input that has to be followed.

Answer 18

hypothesis

Question 19

Because of the assumptions they make, parametric models often need ______ to reach a certain level of accuracy.

Answer 19

less data

Question 20

If the hypothesis is met, parametric models can be more efficient and perform better than ______.

Answer 20

non-parametric

Question 21

A limitation of parametric models is that the assumption often ______ the problem.

Answer 21

simplifies

Question 22

Parametric models have proven to be sensitive to outliers, show limited performance to ______ problems, and struggle to adapt to new unseen data.

Answer 22

nonlinearity

Question 23

Non-parametric models don’t need to make ______ about the relations between the input and output to generate an outcome.

Answer 23

assumptions

Question 24

Non-parametric models also don’t require a certain number of ______ to be set and learned.

Answer 24

parameters

Question 25

Studies have shown that non-parametric models perform better on ______ and are more flexible.

Answer 25

large datasets

Question 26

Common non-parametric algorithms are random forests, decision trees, Support Vector Machines with non-linear kernels, k-Nearest Neighbors (k-NN) algorithm and Neural networks with ______ activation functions.

Answer 26

non-parametric

Question 27

The basic benefit of non-parametric models is that they are able to catch ______ and relationships without having to follow a hypothesis.

Answer 27

complex patterns

Question 28

Non-parametric models can handle ______ and noisy data effectively.

Answer 28

outliers

Question 29

A limitation of non-parametric models is that they require ______ in order to generate better predictions.

Answer 29

more data input

Question 30

Non-parametric models are usually harder to understand and analyze as there are no ______ about the behavior of input and output.

Answer 30

functional assumptions

Question 31

A machine learning model maps from ______ to prediction, represented as f(x) -> y.

Answer 31

features

Question 32

Learning has three stages: Training, ______, and Test.

Answer 32

Validation

Question 33

In the training stage of learning, one aims to optimize ______.

Answer 33

model parameters

Question 34

The validation stage of learning involves intermediate evaluations to ______ model.

Answer 34

design/select

Question 35

The test stage of learning is for final ______.

Answer 35

performance evaluation

Question 36

Model design and 'hyper parameter' tuning is performed using a ______.

Answer 36

validation set

Question 37

Sometimes, you need to split 'train' into a train set for learning parameters and a ______ for checking model performance.

Answer 37

val set

Question 38

The key principle of machine learning states: if Xi is similar to Xj, then yi is probably ______ to yj.

Answer 38

similar

Question 39

In K-Nearest Neighbors, an arbitrary instance x is described by the feature vector , where ar(x) denotes the value of the ______ of instance x.

Answer 39

rth attribute

Question 40

The distance between two instances xi and xj in KNN is defined to be d(xi, xj), calculated using the ______ distance formula.

Answer 40

Euclidean

Question 41

The nearest neighbor algorithm assigns the label/target value of the ______ training features for given test features.

Answer 41

most similar

Question 42

For KNN classification, prediction is usually the ______ or most common class of the returned labels.

Answer 42

mode

Question 43

For KNN regression, prediction is usually the ______ of the returned values.

Answer 43

arithmetic mean

Question 44

Concept learning involves acquiring ______ from specific training examples.

Answer 44

general concepts

Question 45

Each concept can be thought of as a ______ function defined over a larger set.

Answer 45

boolean-valued

Question 46

Implementing KNN involves three steps: Calculate Distance, Identify Nearest Neighbors, and ______.

Answer 46

Aggregate Nearest Neighbors

Question 47

In KNN Classification, the algorithm aggregates the class labels of the 'k' nearest neighbors to predict the class of the current data point, choosing the ______.

Answer 47

most common class label

Question 48

In KNN Regression, the algorithm calculates the ______ of the k nearest training examples.

Answer 48

mean value

Question 49

In nearest-neighbor learning, the target function may be either ______ or real-valued.

Answer 49

discrete-valued

Question 50

The KNN classification algorithm returns f_hat(xq) <- argmax_v_in_V Σ_i=1_k ______.

Answer 50

δ(v, f(xi))

Question 51

In the Distance-Weighted Nearest Neighbor algorithm, greater weight is given to ______.

Answer 51

closer neighbors

Question 52

In Distance-Weighted Nearest Neighbor Classification, wi is defined as ______.

Answer 52

1 / d(xq, xi)^2

Question 53

In Distance-Weighted Nearest Neighbor Regression, f_hat(xq) is calculated as (Σ_i=1_k w_i * f(x_i)) / ______.

Answer 53

Σ_i=1_k w_i

Question 54

Considering all examples in global KNN with distance weighting means the classifier will run ______.

Answer 54

more slowly

Question 55

Nearest neighbour classifier is also known as lazy learning or ______ or instance-based learning.

Answer 55

memory-based learning

Question 56

Nearest neighbour classifiers are useful for large ______ datasets, when trained models become obsolete in a short time.

Answer 56

fast-changing

Question 57

In a nearest neighbour classifier, the training phase involves ______.

Answer 57

do nothing

Question 58

A ______ shows the decision boundaries in a nearest neighbour classifier.

Answer 58

Voronoi diagram

Question 59

The distance-weighted k-NEAREST NEIGHBOR algorithm is robust to ______ and is effective for large numbers of training examples.

Answer 59

noisy training data

Question 60

A limitation of KNN is the assumption that classification is most similar to instances nearby in ______.

Answer 60

Euclidean distance

Question 61

Another limitation of KNN is that the distance between instances is calculated based on ______ of the instance.

Answer 61

all attributes

Question 62

If only a few attributes are relevant, the presence of many irrelevant attributes can ______ the distance metric in KNN.

Answer 62

mislead

Question 63

One solution to KNN's attribute limitation is to ______ each attribute differently when calculating distance.

Answer 63

weight

Question 64

Weighting attributes differently in KNN corresponds to ______ the axes in the Euclidean space.

Answer 64

stretching

Question 65

A more drastic alternative to attribute weighting in KNN is to completely eliminate the ______ attributes from the instance space.

Answer 65

least relevant

Question 66

An additional practical issue in applying k-NEAREST NEIGHBOR is efficient ______.

Answer 66

memory indexing

Question 67

One indexing method for KNN is the ______.

Answer 67

kd-tree

Question 68

In statistical pattern recognition terminology, regression means approximating a ______ target function.

Answer 68

real-valued

Question 69

In statistical pattern recognition, residual is the error ______ in approximating the target function.

Answer 69

f_hat(x) - f(x)

Question 70

In statistical pattern recognition, the kernel function is the function of distance used to determine the ______ of each training example.

Answer 70

weight