Page 168 - AI Computer 10

the basis of your previous interactions with them and come up with improved
performance. For example, if you have requested Alexa to play some lively or
fun music at a time period after you come back from work, it will learn this over
time and play similar music when requested around the time you are supposed
to be back from work.
The working principle of ML models is similar to the IPO cycle. In any ML system,
the model is fed relevant data (past or historical), called training data, which the
algorithms use to train the system and improve its learning. The system uses
this algorithm to on the test input and generates the output.
The block representation of a Deep Learning model is shown below:

Training
Input past Machine Learning Building Logical Output
data Algorithm Models

Learn from
data New data

Machine learning can be applied in a range of applications. Some of these are:
u Object Classification: Object classification is the process of separating input data, values or objects into
separate discrete categories. In the process, input data is categorized under different labels, according to the
defined parameters. The model than accepts new data, matches it with the trained dataset, and classifies
the input data under the correct category. For example, consider the Loan Approval system in a bank. The
system accepts the application from a customer, identifies the parameters such as loan amount requested,
income of the applicant, credit rating of the applicant, etc. and classifies the application as either Approved
or Rejected. The ML model for this system is trained on the data provided by previous loan applications over
a period of several years.

Thus, we can say that object classification works on a set of parameters to segregate the input data into
discrete categories or labels.
u Anomaly Detection: An Anomaly is any behaviour which is in contrast to the accepted or expected behaviour.
Anomaly Detection involves identifying items, events, or observations differ significantly from the majority
of the observed data. Such systems accept the input data, compare it with the training dataset, analyse it
for significant differences, and raise a flag in case of an anomaly detected. Anomaly Detection is critical in
areas such as fraud detection, network intrusions, disease detection in healthcare, and defect detection in
manufacturing processes.

For example, consider the image below. The scooter in the image is an anomaly amidst the fleet of cars. A machine
learning system will scan the images and detect the anomaly based on parameters related to identifying cars.

34
34