Q1. Please list the four machine learning paradigms.

Q2. Please define supervised learning and give a simple example.

Q3. Please define unsupervised learning and give a simple example.
Q4. Please describe the difference between classification and clustering.

A1. Supervised learning, Unsupervised learning, semi-supervised learning, and reinforcement learning.
A2. Machine Learning based on a labelled training dataset. Based on its training data, the program can make accurate decisions when given new data. It is comparable / analogous
to learning with a teacher. Classification and regression are variants of supervised learning. Example: you receive bunch of flowers with labels and your program can identify the
species after having been trained on a labelled dataset.
A3. Unsupervised learning occurs when there is no teaching information to train on, or when your program automatically finds patterns and relationships in the data without using
any form of labeling. Clustering and association learning belong to the unsupervised learning paradigm. Example: you receive flowers without labeling so the algorithm has to
group like flowers according to similarity.
A4. Classification: Classifying data according to pre-defined categories
Clustering: Partitioning data into groups
▪ A paradigm of machine learning where the given dataset contains both inputs
(independent variables, x) and desired outputs (dependent variable, y ).
▪ The objective of supervised learning is to develop an association model that
relates the inputs to the outputs, and can predict the output for future inputs.

where n is the total number of the variables (i.e.,

characteristics) in the data samples.
▪ In classification, the output belongs to a set of finite and discrete values that define
the predicted classes.
▪ In regression, the output belongs to a range of infinite continuous values
that define the numerical outcome(s).
The model utilizes the training set to
compute the best fit.
Given: a basket with some unlabeled objects
Required: arrange the objects by their similarity

picking a random object and select any

interesting physical characteristic of it,
such as its surface shape.

pick all other objects that look similar

in shape to the initially picked object,
and group them together.

repeat until all are clustered

▪ problems with unlabeled datasets.
▪ The provided dataset consists of inputs (independent variables, x) while the output
(dependent variable, y) is not known.
▪ This setting is common in practice since the
acquisition of labels is typically expensive in
big data applications.
▪ The aim of unsupervised learning is to
discover the salient patterns within the given
inputs.
▪ Unsupervised learning models learn to represent the inputs by extracting
statistical structure that is intrinsic to the overall collection of data instances.
▪ Unsupervised learning can also be used to find the best independent variables
that can be useful for either visualizing or clustering the dataset.
▪ Unsupervised learning also encompasses the identification of data points with high
difference to the typical data points in a given dataset (i.e., outliers).
▪ The cost function in an unsupervised learning model can be the minimum
quantization error, minimum distance between similar data instances, or the
maximum likelihood estimation of the correct cluster.
Manufacturing process dataset has the following features:
• Temperature (°C)
• Pressure (psi)
• Flow rate (mL/min)
• Time (min)
• pH level
• Voltage (V)
• Current (A)
▪ Datasets where the output (dependent variable, y) is given for only a few instances
of the inputs (independent variables, xi). Therefore, semi-supervised learning is a
mix of the unsupervised learning and supervised learning paradigms, and
combines the properties of both.
▪ A basic classification model is designed on the few labeled data instances, which is
called the semi-supervised classification step,
▪ A semi-supervised clustering step is then performed, where the model is tuned up
to operate without supervision on the remaining large unlabeled data instances,
and assigns them to the classes from the first step.
▪ Thus, the second step of semi-supervised learning
is to transfer labels to all samples through their
similarity to originally labeled data points
▪ A semi-supervised clustering step is then performed, where the model is tuned up
to operate without supervision on the remaining large unlabeled data instances,
and assigns them to the classes from the first step. Thus, the second step of semi-
supervised learning is to
A popular transfer labels
application to all samples
for semi-supervised through
learning is their similarity to
originally labeled
in thedata
field points,
of medical imaging (Chebli et al., 2018). An
expert can label a small number of radiography scans
of a specific disease and forego the time-intensive task
to label all the scans. The developed semi-supervised
learning model will automatically group the rest of the
scans.
▪ If you need to cross a dark field without hitting any obstacles, a naïve approach
would be to start walking from your starting position until you hit an obstacle, then
you return to your initial place and restart walking while trying to detour just
before the obstacle you had previously hit.
▪ Reinforcement learning is a learning approach for sequential decision making.
▪ The model is simply given a goal to achieve, and it learns how to achieve that goal
by trial-and-error interactions with its environment..
▪ Reinforcement learning accomplishes machine learning in a different way from
both the supervised and unsupervised learning paradigms.
▪ It follows the “cause and effect” method because the developed reinforcement
model learns by performing an action in the environment that result in a maximum
reward over a time horizon.
▪ The actions are adjusted until a maximum reward is achieved.
▪ Both supervised and reinforcement learning map from the inputs to the outputs in
their respective models, but in reinforcement learning there is a reward function
that acts as a feedback to the agent as opposed to the set-up in supervised
learning.