Introduction to Machine Learning

MK Pembelajaran Mesin

www.paramadina.ac.id
Learning Agenda
01 Kesepakatan Kelas
Perkenalan dosen dan mahasiswa, aturan yang harus ditaati dalam
perkuliahan ini, serta buku-buku referensi yang digunakan.

02 Introduction
Pengenalan Pembelajaran Mesin/ML (Machine Learning)

03 Preliminaries
Keterkaitan ML dengan cabang ilmu-ilmu lainnya.

04 Applications of ML
Contoh-contoh penerapan ML dalam kehidupan sehari-hari.
▪ Nama : Harry T Yani Achsan
▪ Alamat Surel : harry.achsan@ gmail.com
[email protected]
▪ Nomor HP : 0818 0854 0094
▪ Kejujuran adalah modal utama
▪ Keaktifan Anda sangat dihargai
▪ Bobot Penilaian:
▪ Tugas/Quizz 20%
▪ Keaktifan di kelas 30%
▪ UTS 20%
▪ UAS 30%
▪ Keaktifan di kelas adalah: setiap jawaban dari pertanyaan dosen atau
setiap pertanyaan mahasiswa yang dianggap bagus akan mendapat
1 (satu) point. Dalam sehari hanya bisa mendapat 2 point, 1 dari
pertanyaan dan 1 dari jawaban. Dalam satu semester cukup
mengumpulkan 10 point saja.
1. Machine Learning,
Algorithms and Applications
Mohssen Mohammed,
Muhammad Badruddin Khan, and
Eihab Bashier Mohammed Bashier
© 2017 by Taylor & Francis Group, LLC
2. Introduction to Machine Learning
Ethem Alpaydın
© 2010 Massachusetts Institute of Tech
3. Introduction to Machine Learning
with Python
Andreas C. Muller and Sarah Guido
© 2017 by O’Reilly Media
Isi dari bahan presentasi
ini diambil dari
berbagai sumber, bukan
hasil karangan
penulis/dosen.
Data and Information
▪ Data is everywhere in abundant amounts.
• Surveillance cameras continuously capture video,
• every time you make a phone call your name and location gets recorded,
• often your clicking pattern is recorded when surfing the web,
• most financial transactions are recorded,
• satellites and observatories generate tera-bytes of data every year,
• the FBI maintains a DNA-database of most convicted criminals,
• soon all written text from our libraries is digitized,
• need I go on?

▪ But data in itself is useless.

▪ Hidden inside the data is valuable information.
▪ The objective of machine learning is to pull the relevant information from the data
and make it available to the user.
▪ What do we mean by “relevant information”? When analyzing data we typically
have a specific question in mind such as :“How many types of car can be discerned
in this video” or “what will be weather next week”.
▪ So the answer can take the form of a single number (there are 5 cars), or a
sequence of numbers or (the temperature next week) or a complicated pattern (the
cloud configuration next week).
▪ If the answer to our query is itself complex we like to visualize it using graphs, bar-
plots or even little movies.
▪ But one should keep in mind that the particular analysis depends on the task one
has in mind.
What is Machine Learning?
▪ Learning, like intelligence, covers such a broad range of processes
that it is difficult to define precisely.
▪ A dictionary definition includes phrases such as “to gain knowledge,
or understanding of, or skill in, by study, instruction, or experience,"
and “modification of a behavioral tendency by experience.“
▪ There are several parallels between animal and machine learning.
▪ Certainly, many techniques in machine learning derive from the
efforts of psychologists to make more precise their theories of animal
and human learning through computational models.
▪ It seems likely also that the concepts and techniques being explored
by researchers in machine learning may illuminate certain aspects of
biological learning.
▪ As regards machines, we might say, very broadly, that a machine
learns whenever it changes its structure, program, or data (based on
its inputs or in response to external information) in such a manner
that its expected future performance improves.
▪ For example, when the performance of a speech-recognition
machine improves after hearing several samples of a person's
speech, we feel quite justified in that case to say that the machine has
learned.
One might ask “Why should machines have to learn? Why not design
machines to perform as desired in the first place?"
There are several reasons why machine learning is important.
1. Some tasks cannot be defined well except by example; that is, we
might be able to specify input/output pairs but not a concise
relationship between inputs and desired outputs.
We would like machines to be able to adjust their internal structure
to produce correct outputs for a large number of sample inputs and
thus suitably constrain their input/output function to approximate
the relationship implicit in the examples.
2. It is possible that hidden among large piles of data are important
relationships and correlations.
Machine learning methods can often be used to extract these
relationships (data mining).
3. Human designers often produce machines that do not work as well
as desired in the environments in which they are used.
In fact, certain characteristics of the working environment might
not be completely known at design time.
Machine learning methods can be used for on-the-job
improvement of existing machine designs.
4. The amount of knowledge available about certain tasks might be
too large for explicit encoding by humans.
Machines that learn this knowledge gradually might be able to
capture more of it than humans would want to write down.
5. Environments change over time.
Machines that can adapt to a changing environment would reduce
the need for constant redesign.
6. New knowledge about tasks is constantly being discovered by
humans. Vocabulary changes.
There is a constant stream of new events in the world.
Continuing redesign of AI systems to conform to new knowledge is
impractical, but machine learning methods might be able to track
much of it.
Wellsprings of Machine Learning
Work in machine learning is now converging from several sources.
These different traditions each bring differentrent methods and
different vocabulary which are now being assimilated into a more
unified discipline.
Here is a brief listing of some of the separate disciplines that have
contributed to machine learning:
Wellsprings of Machine Learning…
1. Statistics: A long-standing problem in statistics is how best to use
samples drawn from unknown probability distributions to help
decide from which distribution some new sample is drawn.
A related problem is how to estimate the value of an unknown
function at a new point given the values of this function at a set of
sample points.
Statistical methods for dealing with these problems can be
considered instances of machine learning because the decision
and estimation rules depend on a corpus of samples drawn from
the problem environment.
Wellsprings of Machine Learning…
2. Brain Models: Non-linear elements with weighted inputs have
been suggested as simple models of biological neurons.
Brain modelers are interested in how closely these networks
approximate the learning phenomena of living brains.
We shall see that several important machine learning techniques
are based on networks of nonlinear elements - often called neural
networks.
Work inspired by this school is sometimes called connectionism,
brain-style computation, or sub-symbolic processing.
Wellsprings of Machine Learning…
3. Adaptive Control Theory: Control theorists study the problem of
controlling a process having unknown parameters which must be
estimated during operation.
Often, the parameters change during operation, and the control
process must track these changes.
Some aspects of controlling a robot based on sensory inputs
represent instances of this sort of problem.
4. Psychological Models: Psychologists have studied the
performance of humans in various learning tasks.
Wellsprings of Machine Learning…
5. Artificial Intelligence: From the beginning, AI research has been
concerned with machine learning.
Samuel developed a prominent early program that learned
parameters of a function for evaluating board positions in the game
of checkers [Samuel, 1959].
AI researchers have also explored the role of analogies in learning
[Carbonell, 1983] and how future actions and decisions can be
based on previous exemplary cases [Kolodner, 1993].
Another theme has been saving and generalizing the results of
problem solving using explanation-based learning [DeJong &
Mooney, 1986, Laird, et al., 1986, Minton, 1988, Etzioni, 1993].
Wellsprings of Machine Learning…
6. Evolutionary Models: In nature, not only do individual animals
learn to perform better, but species evolve to be better fit in their
individual niches.
Since the distinction between evolving and learning can be blurred
in computer systems, techniques that model certain aspects of
biological evolution have been proposed as learning methods to
improve the performance of computer programs.
Genetic algorithms [Holland, 1975] and genetic programming
[Koza, 1992, Koza, 1994] are the most prominent computational
techniques for evolution.
▪ Machines are by nature not intelligent.
▪ Initially, machines were designed to perform specific tasks, such as
running on the railway, controlling the traffic flow, digging deep
holes, traveling into the space, and shooting at moving objects.
▪ Machines do their tasks much faster with a higher level of precision
compared to humans.
▪ The fundamental difference between humans and machines in
performing their work is intelligence.
▪ The human brain receives data gathered by the five senses: vision,
hearing, smell, taste, and tactility.
▪ These gathered data are sent to the human brain via the neural
system for perception and taking action.
▪ In the perception process, the data is organized, recognized by
comparing it to previous experiences that were stored in the
memory, and interpreted.
▪ Accordingly, the brain takes the decision and directs the body parts
to react against that action. At the end of the experience, it might be
stored in the memory for future benefits.
▪ A machine cannot deal with the gathered data in an intelligent way.
▪ It does not have the ability to analyze data for classification, benefit
from previous experiences, and store the new experiences to the
memory units; that is, machines do not learn from experience.
▪ Because machines have been originally created to help humans in
their daily lives, it is necessary for the machines to think, understand
to solve problems, and take suitable decisions akin to humans.
▪ In other words, we need smart machines.
▪ In fact, the term smart machine is symbolic to machine learning
success stories and its future targets.
1.2.1 Machine Learning: Where Several Disciplines Meet
▪ Machine learning is a branch of artificial intelligence that aims at
enabling machines to perform their jobs skillfully by using intelligent
software.
▪ The statistical learning methods constitute the backbone of
intelligent software that is used to develop machine intelligence.
▪ Because machine learning algorithms require data to learn, the
discipline must have connection with the discipline of database.
▪ Similarly, there are familiar terms such as Knowledge Discovery from
Data (KDD), data mining, and pattern recognition.
▪ One wonders how to view the big picture in which such connection is
illustrated.
1.2.1 Machine Learning: Where Several Disciplines Meet …
▪ The idea is very simple. We are not targeting to understand the underlying
processes that help us learn.
▪ We write computer programs that will make machines learn and enable them to
perform tasks, such as prediction.
▪ The goal of learning is to construct a model that takes the input and produces the
desired result.
▪ Sometimes, we can understand the model, whereas, at other times, it can also be
like a black box for us, the working of which cannot be intuitively explained.
▪ The model can be considered as an approximation of the process we want
machines to mimic. In such a situation, it is possible that we obtain errors for some
input, but most of the time, the model provides correct answers.
▪ Hence, another measure of performance (besides performance of metrics of speed
and memory usage) of a machine learning algorithm will be the accuracy of results.
1.2.2 Supervised Learning
▪ In supervised learning, the target is to infer a function or mapping from training
data that is labeled. The training data consist of input vector X and output vector Y
of labels or tags. A label or tag from vector Y is the explanation of its respective
input example from input vector X. Together they form a training example. In other
words, training data comprises training examples. If the labeling does not exist for
input vector X, then X is unlabeled data.
▪ Why such learning is called supervised learning? The output vector Y consists of
labels for each training example present in the training data. These labels for
output vector are provided by the supervisor. Often, these supervisors are humans,
but machines can also be used for such labeling. Human judgments are more
expensive than machines, but the higher error rates in data labeled by machines
suggest superiority of human judgment. The manually labeled data is a precious
and reliable resource for supervised learning. However, in some cases, machines
can be used for reliable labeling.
1.2.2 Supervised Learning…
▪ Two groups or categories of algorithms come under the umbrella of supervised
learning. They are
1. Regression
2. Classification
1.2.2 Supervised Learning…
▪ Two groups or categories of algorithms come under the umbrella of
supervised learning. They are
1. Regression
2. Classification
1.2.3 Unsupervised Learning
▪ In unsupervised learning, we lack supervisors or training data. In
other words, all what we have is unlabeled data.
▪ The idea is to find a hidden structure in this data.
▪ There can be a number of reasons for the data not having a label. It
can be due to unavailability of funds to pay for manual labeling or the
inherent nature of the data itself.
▪ With numerous data collection devices, now data is collected at an
unprecedented rate. The variety, velocity, and the volume are the
dimensions in which Big Data is seen and judged.
▪ To get something from this data without the supervisor is important.
This is the challenge for today’s machine learning practitioner.
1.2.4 Semi-Supervised Learning
▪ In this type of learning, the given data are a mixture of classified and
unclassified data. This combination of labeled and unlabeled data is
used to generate an appropriate model for the classification of data.
▪ In most of the situations, labeled data is scarce and unlabeled data is
in abundance. The target of semi-supervised classification is to learn
a model that will predict classes of future test data better than that
from the model generated by using the labeled data alone.
▪ The way we learn is similar to the process of semi-supervised
learning. A child is supplied with
1. Unlabeled data provided by the environment. The surroundings of a child are
full of unlabeled data in the beginning.
2. Labeled data from the supervisor. For example, a father teaches his children
about the names (labels) of objects by pointing toward them and uttering their
names.
1.2.5 Reinforcement Learning
▪ The reinforcement learning method aims at using
observations gathered from the interaction with
the environment to take actions that would maximize
the reward or minimize the risk.
▪ In order to produce intelligent programs (also called agents),
reinforcement learning goes through the following steps:
1. Input state is observed by the agent.
2. Decision making function is used to make the agent perform an action.
3. After the action is performed, the agent receives reward or reinforcement from the
environment.
4. The state-action pair information about the reward is stored.
▪ Using the stored information, policy for particular state in terms of
action can be fine-tuned, thus helping in optimal decision making for
our agent.
1.2.6 Validation and Evaluation
▪ Assessing whether the model learnt from machine learning algorithm is
good or not, needs both validation and evaluation.
▪ If one claims that for a particular training data the function fits perfectly,
then for the machine learning community, this claim is not enough. They
will immediately ask about the performance of function on testing data.
▪ A function fitting perfectly on training data needs to be examined.
Sometimes, it is the phenomenon of overfitting that will give best
performance on training data, and when yet-unseen labeled data will
be used to test them, they will fail miserably.
1.2.6 Validation and Evaluation…
▪ To avoid overfitting, it is common practice to divide the labeled data
into two parts:
1. Training data
2. Testing data
▪ A training set is used to build the model and testing set is used to
validate the built model.
▪ Larger portion of the data is used for model training purpose, and the
test metrics of the model are tested on holdout data.
1.2.6 Validation and Evaluation…
▪ The technique of cross-validation is
useful when the available training
dataset is quite small and one cannot
afford to hold out part of the data just
for validation purposes.
▪ In k-fold cross-validation, the available dataset is divided into k equal folds.
Each of these k folds are treated as holdout datasets, and the training of the
model is performed on rest of the k − 1 folds.
▪ The performance of the model is judged on the basis of holdout fold. The
average of performance on all k folds is the overall performance of model.
1.3.1 Automatic Recognition of Handwritten Postal Codes
▪ Today, in order to communicate, we use a variety of digital devices.
However, the postal services still exist, helping us send our mails, gifts,
and important documents to the required destination.
▪ The way machine learning has benefited this sector can be understood
by citing the example of the US Postal Service.
▪ The US Postal Service was able to exploit the potentials of machine
learning in the 1960s when they successfully used machines to
automatically read the city/state/ZIP code line of typed addresses to
sort letters. Optical character recognition (OCR) technology was able to
correctly interpret the postal address using machine learning
algorithm.
1.3.1 Automatic Recognition of Handwritten Postal Codes…
▪ In 2006, the US Postal Service sorted and delivered more than 213
billion pieces of mail.
▪ This enormous herculean service is provided by the US Postal Service
with the help of machines.
▪ OCR has been successful in bringing a new revolution in the efficiency
of postal system.
▪ The OCR camera helped in forming the connection between the
physical mail and the information system that directs it to its
destination.
1.3.1 Automatic Recognition of Handwritten Postal Codes…
▪ Now improved OCR technology accompanied with other mail
processing services is able to enhance the efficiency of different
countries’ postal services.
▪ According to the US Postal Service website [7], “the Postal Service is the
world leader in optical character recognition technology with machines
reading nearly 98% of all hand-addressed letter mail and 99.5% of
machine-printed mail.”
▪ Google is now providing free service to convert the text image to text
documents for 200 languages in more than 25 writing systems.
1.3.1 Automatic Recognition of Handwritten Postal Codes…
▪ The OCR technology is one of the applications of pattern recognition, a
branch of machine learning.
▪ The focus of pattern recognition is to recognize pattern and regularities
in data.
▪ The data can be text, speech, and/or image.
▪ The OCR example is the one in which input data is in the form of an
image.
▪ Another example of the application of pattern recognition using image
data is computer-aided diagnosis.
1.3.2 Computer-Aided Diagnosis
▪ Pattern recognition algorithms used in computer-aided diagnosis can
assist doctors in interpreting medical images in a relatively short
period.
▪ Medical images from different medical tests such as X-rays, MRI, and
ultrasound are the sources of data describing a patient’s condition.
▪ The responsibility of a radiologist is to analyze and evaluate the output
of these medical tests that are in the form of a digital image.
▪ The short time constraint requires that the radiologist be assisted by
machine.
1.3.2 Computer-Aided Diagnosis…
▪ Computer-aided diagnosis uses pattern recognition techniques from
machine learning to identify suspicious structures in the image.
▪ How does an algorithm catch suspicious structure? Supervised learning
is done to perform this task.
▪ Few thousand labeled images are given to the machine learning
algorithm, such as Bayesian classifier, artificial neural network, radial
basis function network, and support vector machine.
▪ The resulting classifier is expected to classify new medical images
correctly.
1.3.2 Computer-Aided Diagnosis…
▪ Mistakes in diagnosis by the machine learning algorithm can bring
disaster for a family.
▪ The fault can cause damage to a person in monetary terms and it can
risk his/her life, too.
▪ The following are two such examples:
1. Suppose our classifier detects breast cancer in a patient who actually
had no such disease.
The results obtained by the classifier will create harmful
psychological conditions for the patient. In order to confirm the result
of the classifier, further tests can result in monetary losses for the
patient.
1.3.2 Computer-Aided Diagnosis…
2. Suppose our classifier does not detect breast cancer in patient who
actually has such a disease.
This will lead to wrong medical treatment and can threaten the life of
the patient in near or far future.
▪ In order to avoid such mistakes, the complete substitution of doctor with
technology is not recommended.
▪ The role of technology should be supportive.
▪ It should be the doctor(generally a radiologist) who must take the
responsibility of the final interpretation of medical image.
1.3.2 Computer-Aided Diagnosis…
▪ Computer-aided diagnosis is assisting medical doctors/ radiologists in
the diagnosis of a number of health problems.
▪ Few examples are as follows:
• Pathological brain detection
• Breast cancer
• Lung cancer
• Colon cancer
• Prostate cancer
• Bone metastases
• Coronary artery disease
• Congenital heart defect
• Alzheimer’s disease
1.3.3 Computer Vision
▪ We want our robots to see and act appropriately after understanding
the situation.
▪ The cameras installed in a robot can provide images, but they will not
help the robot recognize or interpret the image.
▪ Using pattern recognition, what type of learning can a robot perform?
We begin with the discussion of the example of the event called
RoboCup.
The autonomous robots are expected to cooperate with their other team members
(that are also robots) in adversarial dynamic environment in order to win the
match. They need to categorize objects and recognize activities. To perform these
tasks, they get input from their cameras. These tasks lie purely in the pattern
recognition domain, a branch of machine learning.
1.3.3 Computer Vision…
1. Driverless Cars
Autonomous cars with no drivers are also one of the applications where car vision is
actually made possible by advancement in the computer vision technology. In the
industry, it is clear that there is ongoing competition to manufacture driverless cars
running on the roads as early as possible.
According to the BBC report titled Toyota promises driverless cars on roads by 2020,
different competitors are on the bandwagon and announcing their targets for
driverless
2. Face Recognition and Security
Images from smart phones and CCTV cameras are now produced at an
unprecedented rate. A problem pertinent to face recognition is to associate the face
image to its respective identity. Building a classifier for this task is not a trivial job,
because there are too many classes involved with multiple image-related problems.
Face recognition can help security agencies to use a large amount of data from
different sources to automatically find what is very difficult for humans to do
manually.
1.3.4 Speech Recognition
▪ The field of speech recognition aims to develop methodologies and
technologies that enable computers to recognize and translate spoken
language into text. Stenography (writing in shorthand) is no longer
required.
▪ Automatic transcription of speech into text has found its way in areas
such as video captioning and court reporting.
▪ This technology can help people with disabilities.
▪ With the passage of time, the accuracy of speech recognition engines is
increasing. There is no doubt that voice-controlled programs such as
Apple’s Siri, Google Now, Amazon’s Alexa, and Microsoft’s Cortana do
not always understand our speech, but things are likely to be improved
in the near future.
1.3.5 Text Mining
▪ The examples that we have studied up until now are basically using
image or voice data for learning.
▪ We have another source of learning, that is, text data. It was observed
that most of the enterprise-related
▪ information is stored in text format. The challenge was how to use this
unstructured data or text.
▪ Another venue where the unstructured data or text is available in
abundance for researchers is social media.
▪ Social media is the place where we can see the production of text data
at an unprecedented level.
1.3.5 Text Mining…
▪ The sharing of personal experiences in the form of text has provided
stakeholders, such as business, the opportunity to analyze and use them
for beneficial purpose.
▪ Text mining is helpful in a number of applications including
• Business intelligence
• National security
• Life sciences
• Those related to sentiment classification
• Automated placement of advertisement
• Automated classification of news articles
• Social media monitoring
• Spam filter
1.3.5 Text Mining…
▪ It is possible that in order to solve a particular problem, both text and
image data are used. For example, the problem of author identification
for a particular written corpus of data can be solved in two ways:
1. Handwriting detection: The known corpus of handwritten data can be
used to make a classifier that can assign a document to an author
based on different features.
2. Writing style detection: This is a text mining problem. We want to find
features that are related to a peculiar author using known documents
attributed to the author. These features can be used to build a
classifier that can identify whether the particular document belongs
to the author or not.
▪ It is possible that the two classifiers are joined together to develop a
new classifier with improved performance for author identification.
1.3.5 Text Mining…
▪ Another area where such data can be helpful in solving the problem is
in the identification of unwanted material in a video. In order to identify
unwanted material, we can approach the problem in two ways:
1. Use video images and apply machine learning techniques on image
data to make a model to identify unwanted material in the video.
2. Use comments from social media related to video to understand the
content of the video by making a model that can predict the presence
or absence of unwanted material in the video.
▪ Once again, the two classifiers can be combined to improve the
performance of the system.
1.4.1 Thinking Machines
▪ The question of whether a machine can think was first asked by the
British mathematician Alan Turing in 1955, which was the start of the
artificial intelligence history. He was the one who proposed a test to
measure the machine’s performance in terms of intelligence.
▪ In 2014, a chatbot was able to pass this Turing test. A chatbot is a
computer program that simulates an intelligent conversation with one
or more human users. This conversation can be performed via audio or
text communication methods.
1.4.1 Thinking Machines…
▪ Researchers at Google have programmed an advanced type of chatbot that is
able to learn from training data comprising of examples from dialogues. The
two sources of training data were IT helpdesk troubleshooting dataset and
movie transcript dataset.
▪ They trained their chatbot with language model based on recurrent neural
network.
▪ It means that these are not just canned answers that are given by chatbots
seeing some patterns in human chats.
▪ Some of the interesting and artistic answers by the chatbot from Google are
available in the research paper titled, “A neural conversational model”. The
researchers admitted the limitation of the work in their research paper that the
chatbot was unable to have a realistic conversation currently and, hence
passed the Turing test; however proper answers to many different types of
questions without rules is a surprising discovery.
1.4.2 Smart Machines
▪ The dream of machines appearing as smart as humans is still far from being realized.
▪ In general, a smart machine is an intelligent system that uses equipment such as
sensors, RFID, a Wi-Fi, or cellular communications link to receive data and interpret it
to make decisions.
▪ They use machine learning algorithms to accomplish tasks usually performed by
humans in an order to enhance efficiency and productivity.
▪ A true smart machine meets two criteria:
1. First, a smart machine does something that no machine was ever thought to be able
to do.
Using that yardstick, a drone delivering a package—a model being contemplated
by Amazon—would qualify as a smart machine
2. Machine is capable of learning.
Using the second criterion for a true smart—the delivery drone fails the test
1.4.3 Deep Blue
▪ In May 1997, IBM’s Deep Blue became the first computer system to
defeat the then-chess world champion Garry Kasparov in a match.
▪ The brute force of the computing power due to specialized hardware
made Deep Blue capable of evaluating 200 million positions per
second.
▪ The 259th most powerful supercomputer of 1997 was able to defeat the
human world champion of chess.
▪ It was a historical achievement for the artificial intelligence community.
1.4.3 Deep Blue
▪ In May 1997, IBM’s Deep Blue became the first computer system to
defeat the then-chess world champion Garry Kasparov in a match.
▪ The brute force of the computing power due to specialized hardware
made Deep Blue capable of evaluating 200 million positions per
second.
▪ The 259th most powerful supercomputer of 1997 was able to defeat the
human world champion of chess.
▪ It was a historical achievement for the artificial intelligence community.
1.4.4 Personal Assistant
▪ Google Now
▪ Apple’s Siri
▪ Microsoft’s Cortana
1.4.5 Predictions for the Future of Machine Learning
▪ Most applications will include machine learning.
In only a few years, machine learning will become part of nearly every
software application. Engineers will even embed these capabilities
directly into our devices. Think of how well your TV streaming service
knows what to recommend. Expect this level of personalization to
become ubiquitous and improve the customer experience everywhere.
▪ Machine learning as a service will become more common.
As machine learning becomes increasingly valuable and the
technology matures, more businesses will start using the cloud to offer
machine learning as a service (MLaaS). This will allow a wider range of
organizations to take advantage of machine learning without making
large hardware investments or training their own algorithms.
1.4.5 Predictions for the Future of Machine Learning…
▪ Computers will get really good at talking like humans.
Before machine learning, computers had a very hard time
understanding even simple human language. Machine learning helps
computers understand the context and meaning of sentences much
better though natural language processing (NLP). As the technology
improves, solutions such as IBM Watson Assistant will learn to
communicate seamlessly without using code.
▪ Algorithms will constantly retrain.
Currently, most machine learning systems train only once. Based on that
initial training, the system will then address any new data or problems.
Over time, the training information often becomes dated or imperfect.
In the near future, more machine learning systems will connect to the
internet and continuously retrain on the most relevant information.
1.4.5 Predictions for the Future of Machine Learning…
▪ Specialized hardware will deliver performance breakthroughs.
Traditional CPUs alone have had limited success running machine
learning systems. GPUs, however, have an advantage in running these
algorithms because they have a large number of simple cores. AI
experts are also using field-programmable gate arrays (FPGAs) for
machine learning. At times, FPGAs can even outperform GPUs.
As specialized hardware continues to improve and become more
affordable, more organizations will gain access to increasingly powerful
machines. These improvements in the underlying hardware will enable
breakthroughs in all areas of AI, including machine learning.
One trend is consistent across all five of these predictions: as this
technology advances, more businesses will embrace the AI revolution.
Competition to make the most effective use of data and machine
learning will tighten. The teams with the strongest AI strategies will
have a major competitive advantage.
A. Supervised Learning Algorithms,
we will discuss following algorithms:
1. Decision trees
2. Rule-based algorithms
3. Naïve Bayesian algorithm
4. Nearest neighbor algorithm
5. Neural networks
6. Linear discriminant analysis
7. Support vector machine
B. Unsupervised Learning Algorithms,
we will discuss following algorithms:
1. K-means
2. Gaussian mixture model
3. Hidden Markov model
4. Principal components analysis in the context of dimensionality
reduction