What is a good definition of machine learning?
Machine learning ( ML ) is an artificial intelligence(AI) application in which computer
programs use algorithms to find patterns in data.
You can do this without relying on humans and without
any special programming.
Machine learning (ML) is the process of using a
mathematical model of data to help a computer learn without direct
instructions.
It is considered a subset of artificial intelligence
(AI).
Machine learning uses algorithms to identify patterns
within data and uses those patterns to create data models that can make
predictions.
Much like how humans improve with practice, machine
learning also improves results when data and environments improve.
Because machine learning is adaptive, it is well
suited for scenarios where data is always changing, the nature of a request or
task is always changing, or where coding a solution is practically impossible.
The link between machine learning and AI
Machine learning is considered a subset of AI.
'Intelligent' computers think like humans and do their
own thing.
One way to
train a computer to mimic the human reasoning process is to use a neural
network, a set of algorithms modeled after the human brain.
The link between machine learning and predictive analytics
Machine learning is a type of predictive analytics,
but the notable difference is that machine learning is much easier to implement
with real-time updates because it can get more data.
Predictive analytics typically use static data sets
and need to be refreshed for updates.
The link between machine learning and deep learning
Deep learning is a specialized form of machine
learning that uses neural networks (NNs) to provide answers.
Deep learning, which can verify its accuracy on its
own, classifies information like the human brain and powers some of the most
human-like AIs.
Benefits of machine learning
The applications of machine learning are diverse and
the possibilities are constantly expanding.
Here are some of the key benefits that businesses can
gain from machine learning projects:
Get insights
Machine learning can help you tell the story your data
tells you by identifying patterns or structures in both structured and
unstructured data.
Improve data integrity
Machine learning is very useful for data mining, and
it can go one step further and improve its capabilities over time.
Improved user experience
Adaptive interfaces, targeted content, chatbots, and
voice-enabled virtual assistants are all examples of how machine learning can
be used to optimize customer experiences.
Reduce risk
As fraud strategies are constantly changing, machine
learning also monitors and identifies new patterns accordingly and catches them
in the trial phase before they succeed.
Anticipating customer behavior
Machine learning mines customer-related data to
uncover patterns and behaviors to optimize product recommendations and deliver
the best customer experience.
Cut down the money
As one machine learning application, process
automation frees up time and resources, freeing your team to focus on what
matters most.
machine learning techniques
There are three main techniques used in machine
learning:
supervised learning
Data that addresses datasets with labels or structures
acts like a teacher, 'training' the machine, enhancing the machine's ability to
make predictions or decisions.
self-study
Address data sets without labels or structures, and
group data into clusters to uncover patterns and relationships.
Reinforcement learning
On behalf of a human operator, an agent, a computer
program on behalf of a person or thing, helps determine a result according to a
feedback loop.
How machine learning works to solve problems
Here's an overview of the machine learning process used
to solve the problem:
Step 1: Data Collection and Preparation
Once the data source is identified, usable data is
compiled. Depending on the type of data you have, you know what machine
learning algorithms are available. When you review data, you can check for
anomalies, develop structure, and troubleshoot data integrity.
Step 2: Train the model
The prepared data is split into two groups: a training
set and a test set. The training set is a fairly large part of the data and is
used to tune the machine learning model to the highest accuracy.
Step 3: Validate the Model
When you are ready to choose your final data model,
you use a set of tests to evaluate its performance and accuracy.
Step 4: Interpret the results
Review results to find insights, draw conclusions, and
predict outcomes.
What can machine learning do?
Regression algorithms, useful for determining cause
and effect between variables, use the values to build models, and these
algorithms are used for a variety of predictive tasks.
Regression studies are useful for forecasting the
future to predict product demand, forecast sales figures, or estimate campaign
outcomes.
Identification of anomalies
Anomaly detection algorithms, often used to identify
potential risks, find data that deviates from expected standards. Equipment
malfunctions, structural defects, text errors, and instances of fraud are
examples of how machine learning can be used to solve problems.
find structure
Clustering algorithms are often the first steps in
machine learning and represent the underlying structure within a data set.
Clustering to categorize common items is commonly used
in market segmentation and provides insights that can aid in price selection
and predicting customer preferences.
Category prediction
Classification algorithms help determine the exact
category of information. Classification is similar to clustering, but differs
in that it applies to supervised learning to which predefined labels are
assigned.
What do machine learning engineers do?
Machine learning engineers transform raw data collected from various data pipelines into data science models that can be applied and scaled as needed. Machine learning engineers connect that structured data to models defined by data scientists they work with.
Machine learning engineers also develop algorithms and
build programs that machines, computers, and robots can use to process incoming
data and identify patterns.
How many industries are using machine learning
Across industries, companies are using machine
learning in many different ways. Some examples of machine learning in key
industries include:
Banking and finance
Risk management and fraud prevention are key areas
where machine learning is adding enormous value to the financial sector.
Medical treatment
There are many examples of how machine learning can
improve patient care, including diagnostic tools, patient monitoring, and
outbreak prediction.
Transit
Identifying anomalous traffic volumes, optimizing
delivery routes, and autonomous vehicles are examples of how machine learning
can positively impact traffic.
Customer service
By answering questions, measuring customer intent, and
providing virtual assistance, machine learning is being leveraged to support
the customer service industry.
Retail
Machine learning helps retailers analyse buying
patterns, optimize products and prices, and use the data to improve the overall
customer experience.
Agriculture
Developing robots to solve job shortages, diagnosing
plant diseases and monitoring the health of agriculture are examples of how
machine learning can improve agriculture
Machine Learning Platform Features
When you choose a machine learning platform, you get a
solution that includes the following features:
Cloud computing
Easy to set up and deploy, the cloud is perfect for
handling workloads of all sizes, and it doesn't require advanced knowledge to
connect to data sources and scale as needed.
Accessible development environment
The ideal platform supports all skill levels with
accessible authoring options. These platforms allow users to reap the benefits
of machine learning, whether they work extensively in code or prefer automated
tools and drag-and-drop interfaces that don't require any coding experience.
Built-in support for familiar machine learning frameworks
Find platforms where you can work with the tools
you're familiar with and preferred: ONNX, Python, PyTorch, scikit-learn, or
TensorFlow.
Enterprise-grade security
Find a platform that provides controls that support
enterprise-level governance, security, and infrastructure protection.
Machine learning algorithms
What is a machine
learning algorithm?
Introduction to the mathematics and logic behind
machine learning
Machine learning
algorithms are pieces of code that help users explore, analyze, and find
meaning in complex data sets. Each algorithm is a finite set of clear,
step-by-step instructions that a machine can follow to achieve
a specific goal. In
machine learning models, the goal is to establish or discover patterns that
users can use to make predictions or classify information.
Machine learning
algorithms use parameters based on training data (a subset of data representing
a larger set). As the training data is scaled to represent the world more
realistically, the algorithm computes more accurate results.
Different algorithms
analyze data in different ways. Algorithms are grouped into commonly used
machine learning techniques (supervised learning, unsupervised learning, and
reinforcement learning).
The most commonly used
algorithms use regression and classification to predict target categories, find
anomalous data points, predict values, and search for similarities.
machine learning techniques
As you dive deeper into
machine learning algorithms, you'll notice that machine learning algorithms
typically fall into one of three machine learning techniques.
supervised learning
In supervised learning,
an algorithm makes predictions based on a set of labelled examples supplied by
the user.
This technique is useful when you know what
the result will look like.
For example, you want to
provide a data set containing city populations by year for the last 100 years,
and you want to know what the population of a particular city will be 4 years
from now.
The labels (population, city, and year)
already in the data set are used in the results.
In unsupervised learning,
data points are not labeled. These algorithms organize data or label it in a
way that describes its structure.
This technique is useful
when you don't know what the result will look like.
For example, you provide
customer data, and you want to create a segment of customers who prefer similar
products.
The data you provide is not labeled, and the
labels in the results are generated based on the similarities found between the
data points.
Reinforcement learning
Reinforcement learning
uses algorithms that learn from the results and decide what action to take
next.
After each action, the algorithm receives
feedback to help it decide whether a choice was correct, neutral, or wrong.
This is a good technique
to use for automated systems that need to make a lot of small decisions without
human guidance.
For example, let's say
you're designing a self-driving car, and you want to make sure it obeys the law
and keeps people safe.
As the car gains
experience and reinforcement over time, it learns to stay in its lane, follow
the speed limit, and brake for pedestrians.
What Machine Learning Algorithms Can Do
Machine learning
algorithms help answer questions that are too complex to answer through manual
analysis.
Although there are many
different types of machine learning algorithms, use cases for machine learning
algorithms typically fall into one of the following categories:
Target Category Prediction
A two-class (binary)
classification algorithm divides data into two categories. This algorithm is
useful for questions that have only two mutually exclusive possible answers
(including yes/no questions).
See the following
example.
Will this tire explode
during the next 1,000 miles?
Which option can I get
more referrals for? (10 USD credit or 15% off).
Multiclass (polynomial) classification algorithms
Divide data into three or
more categories.
This algorithm is useful
for questions with three or more possible answers that are mutually exclusive.
See the following example.
What month do most
travelers buy their tickets?
What emotions does the
person in this picture show?
Find abnormal data points
Anomaly detection
algorithms identify data points that do not fall within the range of parameters
defined for what is “normal”.
For example, it uses an anomaly
detection algorithm to answer questions such as:
Where is the defect in
this batch?
What credit card
purchases can be fraudulent?
value prediction
A regression algorithm
predicts the value of a new data point based on historical data. This algorithm
can answer the following questions.
What will the average
cost of a two-bedroom house in my city be next year?
How many patients visit
the hospital on Tuesday?
See how values change over time
A time series algorithm
shows how a given value changes over time.
With time series analysis
and time series forecasting, you collect data at regular time intervals and use
that data to make forecasts and identify trends, seasonality, periodicity, and
irregularities.
Time series algorithms
are used to answer questions such as:
Will the price of a
particular stock go up or down next year?
How much will it cost
next year?
Similarity Search
Clustering algorithms
divide data into groups by determining the level of similarity between data
points.
The clustering algorithm
works well for questions like:
What kind of viewers like
the same kind of movies?
Which printer model does
the error occur in the same way?
classification
Classification algorithms
use predictive calculations to assign data to preset categories.
A classification
algorithm is trained using input data and used to answer questions such as:
Is this spam?
What is the sentiment
(positive, negative, or neutral) of the given text?
A linear regression algorithm displays or predicts a relationship
between two variables or factors by fitting a continuous straight line to the
data. This line is usually calculated using the squared error cost function.
Linear regression is one of the very popular types of regression analysis.
A logistic regression analysis algorithm fits a
continuous sigmoidal curve to the data. Logistic regression is another popular
type of regression analysis.
The Naïve Bayes algorithm calculates the probability
that an event will occur based on the occurrence of a related event.
A support vector machine draws a hyperplane between
the two nearest data points. This will ignore the class and maximize the
distance between the two data points to make the distinction clearer.
Decision tree algorithms partition data into two or
more sets of the same type. Use the if–then rule to separate data according to
the most important differentiators between data points.
The K- Nearest Algorithm stores all available data
points and classifies each new data point according to the nearest data point
measured as a function of distance.
The random forest algorithm is based on a decision
tree, but rather than creating a single tree, it creates a forest of trees and
then randomizes the trees in that forest. It then aggregates the votes of other
random constructs in the decision tree to determine the final class of the test
object.
Gradient boosting algorithms produce predictive models
that bundle weak predictive models (usually decision trees) through an ensemble
process that improves the overall performance of the model.
The K-means algorithm classifies data into clusters.
where K is equal to the number of clusters. The data points within each cluster
are of the same type and different from the data points in the other clusters.
Comments