This document provides an introduction to machine learning, including definitions, applications, and types of problems. It discusses how machine learning can be applied to data from sensors in industrial settings for tasks like risk prevention in real-time, searching historical data for patterns related to failures, estimating machine lifespan and product quality, and enabling more automated intelligent systems. Large amounts of data are generated every day from various sources, making it impossible for humans to analyze, but machine learning can find hidden patterns in this data.
2. What is machine learning?
Some examples of applications
Microarrays analysis:
Braking assistance system:
Engine failure detection:
Predictions of the films punctuations:
Identify, classify or count cells in microscopy images:
Types of machine learning problems
Classification:
Numerical prediction:
Clustering:
Association detection
Detection of anomalies
Text mining
Time series
Main paradigms in Machine learning
Supervised learning
Unsupervised learning
Machine Learning and Data Sensors
Risks prevention in real time (monitoring)
Searching patterns in historical data in order to know the cause of previous failures
Machines usage life estimation
Product quality estimation
Automatization: intelligent systems
Lots of data is generated every day and everywhere. Facebook generates 500 TBytes every day, Twitter produces
12 TBytes of comments every day, NASA generates more than 200 TBytes per day and this is increasing
exponentially. Usually all of this data have “hidden” information, these information could be crucial for determinate
phenomena analysis, predictive maintenance, decision taking and much more, giving machines the capacity of
learn by themselves from experience (data). So it is impossible for humans to find most of the info
hidden(patterns) on these huge amounts of data.
What is machine learning?
● “Field of study that gives computers the ability to learn without being explicitly programmed.”-Arthur
Samuel 1959.
● “A computer program is said to learn from experience E with respect to some class of tasks T and
performance measure P, if its performance at tasks in T, as measured by P, improves with experience E.”
-Tom Michel 1998
● “Machine learning is a scientific discipline that explores the construction and study of algorithms that can
learn from data.
Such algorithms operate by building a model from example inputs and using that to make
predictions or decisions, rather than following strictly static program instructions. Machine learning is
closely related to and often overlaps with computational statistics; a discipline which also specializes in
prediction-making-applications.”-Wikipedia 2015.
3. Some examples of applications
Machine learning is a subfield of the artificial intelligence with a lot of applications, like smart robots, text mining
(web search,sentimental analysis, anti spam), artificial vision, audio, medical informatics, data mining and many
other areas.
Microarrays analysis:
Microarrays technique are usedinterpreting the
data generated from experiments on DNA which allow
researchers to investigate the expression state of genes. It is
large used to discover what genes influence a certain disease.
A microarray can contains information of the expression levels
from more than 40.000 genes, it impossible for humans
calculate al combination in deep.
Braking assistance system:
Calculating the braking intensity depending on the car
weight, velocity and the distance to the next car. An algorithm
can learn automatically a model in function of the experience
and use for calculating the new situations.
Engine failure detection:
Detecting which pieces have more failure in short
time taking measurements about the temperature, intensity of
the vibrations, pressure in a set of determined pieces.
4. Predictions of the films punctuations:
Netflix has an algorithm called “CineMatch” that is able to predict what punctuations users
will give to the films in the 95% of the cases.
Identify, classify or count cells in microscopy images:
The algorithm gets a set of images of cells and what
kind of cell are (experience). Then, from that information,
automatically learns the characteristics patterns of each type of
cell.
Types of machine learning problems
There are different types of machine learning problems, Some of the most common are described in the next
points.
Classification:
Consist of finding models able to determinate the
class of any determinate object belongs. There are several
techniques: Decision trees, regression,rules systems, bayes
nets, neural networks, etc
Numerical prediction:
The algorithm estimates a output value (e.g braking distance). Most of the same techniques than
in classification are used for numerical predictions. (eg. regresión)
5. Clustering:
The clustering objective is to group objects
determined by similar characteristics or patterns.(e.g
news) At the end, each cluster determined a class, but
these classes are not previously defined.
Association detection
Aims in finding associations or correlations of interest in a set of transactions, usually represented
by rules. e.g recommendation system in only shopping
Detection of anomalies
An anomaly or outlier is an input that not matches with the general behavior of the data. E.g
engine failure detection, intruders detections, etc.
Text mining
Is the process of extract useful information from texts. Text mining usually involves the process
of structuring the input text,deriving patterns within the structured data, and finally evaluation and
interpretation of the output.
Eg: classify an opinion in negative or positive.
Time series
In this type of problem the algorithms treat with multiple variables with values along of the time.(e.g,data
from temperature sensors in a engine)
The aim is to find patterns in the historical data. (e.g, if the temperature influences the pressure in a
critical moment)
This kind of problems also are frequently use in business for market predictions.
6. Main paradigms in Machine learning
Supervised learning
This kind of learning is carrying out on
classification or numeral prediction tasks. On
these tasks, each case is divided in input
values and output values that are what the
algorithm determines.
E.g:
➢ input: cell image
➢ output: kind of cell.
In supervised learning we have training set
for each input, an it output. The algorithm
uses this information with corrected
classifications to construct a model.When a
new input that we don't know its class,the
model predicts what class this input belongs.
Unsupervised learning
In this case, we don't have previous
knowledge. The algorithm algorithm does not
have info about the outputs.It has its onw
metrics to determinte patterns. Clustering is a
clear example of unsupervised learning.
7. Machine Learning and Data Sensors
Time series is a widely machine learning problem in the industrial sector,when a lot of sensors( temperature,
pressure,etc) are collecting data along production time. This data can be analyzed in different ways due to Time
Series involves other problems like clasificación, numeral prediction, clustering etc. A set of application examples
coming up next.
Risks prevention in real time (monitoring)
We could predict the future values of a sensor. For example we can predict the temperature value in any
determined time,if the system detects with enough time that temperature will be above the any threshold(e.g so
high temperature), the machinery could react against this problem or notify the workers avoiding several problems
in production.
Searching patterns in historical data in order to know the cause of previous
failures
Finding the cause of historical issues is a common machine learning task , it is possible to
deep in the relationships between thousand of variables from data sensors and discovering hidden
crucial information. For example if could know that in a determined situation which variables have
influence in others and relate this with an historical issue like a failure..
Machines usage life estimation
It is possible to estimate the live of the machinery using Machine Learning, we could use a
machine more than its normal use if the system detects that it is possible, or replace a determined
machine if the system detects a coming failure.
8. Product quality estimation
We can estimate the product quality before the product will be completely manufactured
avoiding unnecessary quality tests. This is possible if the relations within the variables and the quality
are discovered by Machine Learning models generated.
Automatization: intelligent systems
The final objective of Machine Learning this fields is to provide a higher level of automatization to the
system. This means that the system would be able to react itself against possible failures and
recovering itself. An example could be if the system detects the an engine is going to break down in
the next few hours, it could stop the production and proceed to solve this problem itself.