Page 290 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 10 Deep neural network in medical image processing 281

2.7 Unlabeled data set
Unlabeled data set is a sample of natural or human-made
artifacts; such type of data does not contain any labels identifying
the characteristics. It could be audio data, video data, Tweets, or
just readings of some experiment or observational reading of any
environmental phenomenon, for example, weather data, e-mail
data, images from your gallery, and traffic cam video logs.

2.8 Labeled data set
For a labeled data set, we take a set of unlabeled data that
attach meaningful labels on the basis of some classification so
that a system can be trained using those labels to identify any
particular type of artifact, for example, an e-mail data set with
tags for spam or images from your gallery with tags for faces of
different people.

2.9 Supervised learning

Supervised learning is the most popular and earliest devel-
oped algorithm. In layman terms, such algorithms can be
explained as teaching a child about different animals by showing
him the pictures of different animals; initially, the child gets it
wrong between confusing animals like a sheep and a goat, but af-
ter showing enough pictures, the accuracy improves consider-
ably. Similarly the machine also gets most of the results wrong
initially, but after going through a considerable number of test re-
cords, the machine starts to form a relation between the input at-
tributes and the resultant label. Information in the form of
labeled examples allows one to feed these labeled pairs one by
one to a machine learning algorithm that enables the algorithm
to predict a label for each example and to provide feedback on
whether the response has been correct. These algorithms try to
model relationships between labels (i.e., target prediction) and
input parameters. In these algorithms, a model is trained on a
labeled data set, and then unlabeled data are used to test the
model and subsequently make predictions. When it is fully
trained, a supervised learning machine can take a never-before-
seen example and predict a good label for the same. Different
types of tasks that can be performed using supervised learning
are as follows:
• Spam classification
• Facial recognition
• Other prediction-type tasks