Week 1: Introduction to Pattern Recognition

Question 1

Pattern Recognition

Answer 1

It’s concerned with creating algorithms that can assign names to observations, or make decisions based on data. Pattern recognitions is suitable for tasks that aren’t purely analytical, such as classifying numbers as prime or composite, or using Newtonian equations to calculate the trajectory of a projectile.

Question 2

Exemplar

Answer 2

A particular datapoint which is represented by a feature vector (also called item, sample, instance,…)

Question 3

Dataset

Answer 3

The collection of feature vectors for all exemplars.

Question 4

Generalisation

Answer 4

How well a model performs on new data.

Question 5

Overfitting

Answer 5

Making the model so specific to the training data that it fails to generalise to new data.

Question 6

Decision Theory

Answer 6

Methods for making decisions that reduce cost rather than misclassification rate.

Question 7

Feature Space

Answer 7

The (multidimensional) space defined by the feature vectors in the dataset.

Question 8

Linearly Separable

Answer 8

Exemplars from two classes can be separated by a hyperplane in the feature space.

Question 9

Dichotomiser

Answer 9

A classifier that places exemplars in one of two classes (also called a binary classifier).

Question 10

Hyper-Parameter

Answer 10

A value used by the learning algorithm in its search for the optimal parameters of the classifier.

Question 11

Grid search

Answer 11

A method of trying to find suitable hyper-parameters that searches all possible combinations of values within defined ranges.

Question 12

Training data

Answer 12

The collection of feature vectors used by the learning algorithm to tune the parameters of the classifier.

Question 13

Test data

Answer 13

The collection of feature vectors used by the model to evaluate the performance of the trained classifier (this dataset should be distinct from the training data to ensure generalisation).

Question 14

Classification

Answer 14

A method that learns to predict a class label associated with each exemplar.

Question 15

Regression

Answer 15

A method that learns to predict a continuous value for each exemplar.

Question 16

Semi-supervised

Answer 16

A method that learns using both labelled and unlabelled training exemplars.

Question 17

Transfer

Answer 17

A method that pre-trains a classifier on another task before training it on the main task in the hope that the pre-training will help improve performance on the main task.

Question 18

Supervised Learning

Answer 18

When the dataset is labelled with the desired output being known for each instance, and the learning algorithm is able to evaluate the performance as the model is trained.

Question 19

Unsupervised Learning

Answer 19

When the dataset is unlabelled with the desired output being unknown for each instance, and the learning algorithm must find ways to find similar groupings of instances.

Question 20

Weakly-Supervised Learning

Answer 20

Supervised learning with inexact or inaccurate labels

Question 21

Machine Learning

Answer 21

It’s a broader discipline than Pattern Recognition that also includes reasoning about data in A.I.

Question 22

Artificial Neural Networks

Answer 22

It’s a narrower discipline than Pattern Recognition, focuses more in neuroscience, inspired neural network architectures.

Question 23

Deep Learning

Answer 23

A subset of Artificial Neural Networks, focusing on networks with many hidden layers with more complexity and less explainability.

Question 24

Error Rate

Answer 24

(False Positive + False Negative)/(False Positive + False Negative + True Positive + True Negative)

Question 25

Confusion Matrix

Answer 25

A table that measured the number of False Positives, False Negatives, True Positive, and True Negatives

Question 26

True Positive

Answer 26

When a positive instance is predicted as positive.

Question 27

True Negative

Answer 27

When a negative instances is predicted as negative.

Question 28

False Positive

Answer 28

When a negative instance is predicted as positive.

Question 29

False Negative

Answer 29

When a positive instance is predicted as negative.

Question 30

Accuracy

Answer 30

(True Positive + True Negative)/(False Positive + False Negative + True Positive + True Negative)

Question 31

Recall

Answer 31

(True Positive)/(True Positive + False Negative)

Question 32

Precision

Answer 32

(True Positive)/(True Positive + False Positive)

Question 33

F-1 Score

Answer 33

(2 * recall * precision)/(recall + precision) = (2 * True Positive) / (2 * True Positive + False Positive + False Negative)