SlideShare ist ein Scribd-Unternehmen logo
1 von 126
Basics of Machine Learning
By. Assistant Professor – Hasib Masud
Shaikh
KCCEMSR, Thane
What Will I Learn?
How Machine learning works.
What are some simple applications of Machine
learning.
What are the ethics of Machine learning.
How big is the future of Machine learning.
Who is the target audience?
People who are progressing their journey towards
machine learning
Where there is data and it needs to be analyzed,
Machine learning is the best way to do so.
Benefits of ML
Data Science sector is increasing rapidly, so is the
demand of people who can write algorithms to
analyze that data.
With increasing amount of data, the accuracy of
the result has to be increased.
Introduction
Machine Learning, probably one of the most
intriguing topics of this decade.
Also known as ML, this field is considered one of
the pillars of the 21st century.
Everyone is talking about it, so there has to be a
reason why!
Introduction
Machine Learning is a field of Computer Science.
It tries to teach computers how to learn, so they
can help humans solve difficult tasks.
The nice thing is that computers should learn
without being explicitly programmed to do that.
...because they are dumb
Computers just see 0s and 1s, and nothing
more.
The app you use daily? 0s and 1s. Games we
play? 0s and 1s.
Everything, from the easiest to the most
complex task, is just 0s and 1s (called bits) for
Lot of research
For this reason ML researchers had to struggle for
many years, before being able to find a way to solve
this problem.
Believe me, a lot of nasty mathematics has been
used, scary stuff.
But we won’t go in depth about it, understanding
the big picture is what’s important now!
Artificial Intelligence
Machine Learning is a subfield of Artificial Intelligence, usually
called AI.
What AI does is to teach Computers how to simulate intelligence,
in order to behave like a human.
Many different aspects of intelligence are studied, each one
expanding in a particular subfield.
Artificial Intelligence branches
To give you an example, following are some of the
branches that AI is composed of:
Handwriting recognition
Speech recognition
Translation
Spam filtering
Automation
Game AI
Planning
Artificial Intelligence branches
•Handwriting recognition
Where a machine tries to understand handwriting
and its properties.
Artificial Intelligence branches
•Translation
Allowing machines to create a bridge between two
different languages
Artificial Intelligence branches
•Speech recognition
Similar to handwriting recognition, but with the intent
of understanding meaning and sentiment of
sentences.
Artificial Intelligence branches
•Spam filtering
Used mostly with emails (but extendable to other
domains), where something unwanted is filtered.
Artificial Intelligence branches
•Automation
Reducing the tasks where a human is needed, in
order to allow him to concentrate on other aspects of
his work.
Artificial Intelligence branches
•Game AI
To be used for non playable characters in games, in
order to show more intelligent behavior or better
performance.
Where is Machine Learning?
Machine Learning can be seen as the part of AI
that allow computers to take as input some data and
makes decisions by learning from it.
Every time there is something to learn from
data, ML comes to the rescue!
It can be something easy like deciding if it’s raining
based on the humidity outside.
Or it can be something difficult, like telling if a
review on a website is positive or negative.
What about Deep Learning?
Then what is Deep Learning (abbreviated DL) in this figure?
It’s a subfield of Machine Learning, that
uses Artificial Neural Networks (ANN) to make decisions, a lot
more powerful stuff!
Don’t worry if you don’t know what those words mean, we will
encounter them later in the course.
Fact Check
Machine Learning teaches computers to learn by
themselves, and this isn’t particularly difficult
because computers already are quite intelligent.
•True
•False
Machine Learning is a subfield of…?
•Robotics
•Deep Learning
•Artificial Intelligence
•None of the above
Fact Check
Machine Learning teaches computers to learn by
themselves, and this isn’t particularly difficult
because computers already are quite intelligent.
•True
•False
Machine Learning is a subfield of…?
•Robotics
•Deep Learning
•Artificial Intelligence
•None of the above
How are we able to teach machines
to reason on something?
The answer is with numbers.
Machine Learning can be seen as an elegant way
of formulating real world problems into
their mathematical equivalent.
In this way a computer, relying only on numbers,
can be used to solve those mathematical problems.
Sentiment analysis
For example, consider the task of saying whether
someone is happy based on what he says.
Sure, this is a particularly difficult problem, but trust
me, scientists found a way of converting it to just
numbers.
And allowing a computer to try and predict it.
Sentiment analysis
•I know, you are just waiting me to tell you how to do
it.
•But it’s nothing simple, this is in fact something that
machines can do only in recent years.
•Maybe I will give you some details in a future ML
course ;).
Numbers everywhere
•The important thing to remember is that at the core
of every algorithm in Machine Learning, there
are numbers.
•And if it seems too strange to be true, take some
simple problems, and try to think about a way to
translate it into math.
Chess
•For example, think about chess.
•Just the game, and not how to win, so we keep
everything simple.
•Normal people may see just a checker pattern, with
some wooden pieces on top.
Chess
•But marvelous miracles of technology(ML)
sees math, formulas, bits and bytes in it.
•Every piece can be seen as a number, for example
1 for pawn, 3 for bishop, 5 for the rook etc.
•Every square can be seen as a coordinate.
•Every move could be just a number for the piece
and two coordinates for the move.
Fact Check
Choose correct option:
•You need to guess someone’s age based on his
gender and height, what is something that doesn’t
help the computer in solving the problem?
Convert the gender into numbers.
Transform the age into words.
Convert heights if measured using different units.
Fact Check
Correct answer:
•You need to guess someone’s age based on his
gender and height, what is something that doesn’t
help the computer in solving the problem?
Convert the gender into numbers.
Transform the age into words.
Convert heights if measured using different units.
Patterns
•“Ok, Now i know how a computer sees the world,
but WHAT does it look for?”
•Computers look for patterns in order to solve the
majority of their problems.
A pattern is something that repeats itself over and
over.
Which patterns?
•For example, consider the past question.
•How would a computer do that?
•It has to find patterns in the data.
•Like the fact that women are usually shorter than men.
•Or even that most of the times when people get old they start to bend in
front thus reducing their vertical height.
From patterns to functions
•All this is translated in some kind of function from the ML program (in this
field it is called model)
•Then it uses the learned functions to solve the given task.
•And this is how ML intuitively works :).
Fact Check
Choose correct option:
•If you don’t see a clear pattern in your problem ML
won’t be able to help you.
True
False
•Patterns can help you better understand the
problem you are solving.
True
False
Fact Check
Choose correct option:
•If you don’t see a clear pattern in your problem ML
won’t be able to help you.
True
False
•Patterns can help you better understand the
problem you are solving.
True
False
Fact Check
Choose correct option:
•If you don’t see a clear pattern in your problem ML
won’t be able to help you.
True
False
•Patterns can help you better understand the
problem you are solving.
True
False
Fact Check
•What can be considered a pattern?
The height of people during their lives
The number of children born for each month
The fluctuations of stock markets
Series of moves in a chess game
Fact Check
•What can be considered a pattern?
The height of people during their lives
The number of children born for each month
The fluctuations of stock markets
Series of moves in a chess game
Correct answer: All of the above
Is ML useful?
•The main purpose of ML is to solve difficult
problems without the help of humans (well, without
much help).
•It has been developed in order to overcome the fact
that many tasks require knowledge that few people
have.
•For example, engineers have built chess engines
without ML for years but...
Is ML useful?
•ML allowed the creation of stronger engines.
•That don’t need explicit rules from experts to work.
•And recently don’t even need to know what game
they are playing to beat the world champions!
ML Power
•This is the power of Machine Learning, and in the
last years its role got more prominent.
•Tasks are more specialized, requiring more expert
knowledge.
•The problems are harder to solve.
•Thus requiring more powerful methods than the
typical handcrafted ones.
ML needs data
•Everything seems beautiful, but there is a catch.
•ML needs a lot of data to be able to work.
•Like, thousands and thousands samples, frequently
millions.
•This was a problem in the past, as not much digital
data was available.
ML needs data
•ML wasn’t considered a very useful tool at the start,
because data wasn’t simply enough.
•This got the impression to the researchers that
handcrafted models were a much more valid
alternative.
Big Data
•Contrary to the past, nowadays we
have abundance of data.
•More specifically, we are living in the Big Data era.
•Big data is a term that indicates the high availability
of data of some sort, in our case in digital format.
•Images, videos, songs, books etc., everything is
largely available to us.
•This, along with more powerful computers, led to a
new blossom of AI and ML.
Big Data - Examples
For example, a model that is able to recognize a
written digit needs hundred of thousands images
of written digits.
A model that predicts future sales of a specific
product needs years worth of sale records.
And so on.
In summary
•In summary, ML is a very powerful tool, that
required years of research to be perfected.
•It also needs a lot of data to understand and
complete its tasks.
•And the world we are living in right now is the
perfect environment for it.
•That’s why everybody is now talking about it!
Spam filtering
•Suppose you are working for a company that
receives dozen of emails every day.
•The boss is tired of checking every email in the
inbox, as many of them are just for marketing from
rival companies.
•He marked every email as wanted or unwanted,
and you have to do the same with every new email.
Spam filtering – The hard way
•You start checking the first email marked as spam,
it tells you about this offer of a wonderful new
vacuum cleaner.
•Ok, no vacuum cleaners.
•You check the second email marked as useful, a
report of last quarter’s earnings. Important.
•The third one is marked as spam, another offer.
•You understand that basically all offers are not
needed for the company. Got it.
•The forth one is from a rich prince that wants to give
you all his money. Spam.
•Of course strange and shady money offers are not
wanted.
Spam filtering – The hard way
OK! No offers and shady money, spam filter created!
But then you find out that the boss also doesn’t want emails from
some customers.
You continue for months, finding another kind of email that is not
wanted and has to be filtered.
And you would stay there continuing to add rules to your filter
indefinitely.
Spam filtering – The easy way -
ML
A task with an underlying pattern that can be discovered, but that
doesn’t need to be “easy” to see. Check.
Abundance of data to be used to train our ML model. Check.
Powerful computer (although for this task just a normal computer would
do). Check.
Well, it’s the process in which we take the training data and feed it to the
model, chunks at a time.
With it, the model will slowly get better and improve its ability to make
correct predictions!
Spam filtering – The easy way -
ML
•Of course, we also need to know how to create
our model.
•With the power of Machine Learning our model will
be able to understand which kind of emails to keep
and which to trash.
•The model will find a way to interpret what we give
him, and will soon be able to create a fantastic spam
filter!
•This is why ML is so hyped nowadays, you don’t
need to know what to look for in your data.
•You just need to know that it’s there.
The easy way – ML - Downside
•Of course this comes with a downside, that with
complicated tasks we don’t know why and how the
model is working.
•We know it works, but not exactly what it’s looking
for.
•This is a current topic of research, in order to create
“explainable” models and methods.
Power of ML - Tanks and forests
•This story narrates how the US government wanted
“something” that was able to recognize enemy tanks
from images.
•Well, this was a perfect moment to show the power
of ML!
•So a model that takes images as input was built,
predicting if the image represented a tank or a
forest.
•The model was trained on a small set of images,
half tanks and half forests.
Power of ML - Tanks and forests
•It was able to always recognize the tanks!
•Scientists were very proud and the government
amazed.
•But it didn’t last long, as it was found out soon after
that the model wasn’t recognizing tanks.
•It was checking the brightness of the image!
•When new images were used, under other light
conditions, the model wasn’t able to get a single
prediction right.
Power of ML - Tanks and forests
•This was an example of how people were
disappointed by this new technology.
•They expected wonders, and ML didn’t give them to
the people.
•So they lost trust in it.
A comeback
•BUT ML DIDN’T STOP.
•A few scientists kept believing in it, and results
came a decade after.
•1985, an Artificial Neural Network (this time created
as a computer program) learns to pronounce new
words.
•In a week, the Network was able to pronounce
20,000 words!
Deep learning
•2006, the term Deep Learning (DL) is coined,
indicating the subfield of Machine Learning that
works with Neural Networks.
•It uses much more data, and “deep Networks”.
•Basically it allows to create more powerful models
inspired by the human brain (Neural Networks).
ML in detail
Are you ready to learn what the
concrete aspects of ML are?
Datasets
•First of all, let me explain concretely what is fed to
a ML model in practice.
•When you have a task to be done and have the
data to use, that data is called a dataset.
•Everything the ML model uses is one or more
datasets.
•An example of a dataset is MNIST, containing tens
of thousands images of black handwritten digits on
white background.
•This dataset is almost always used when learning
to create ML models.
Datasets
•The MNIST database (Modified National Institute of
Standards and Technology database) is a
large database of handwritten digits that is
commonly used for training various image
processing systems.
•The database is also widely used for training and
testing in the field of machine learning.
When feeding a dataset to a
model, each sample usually
needs to be in a particular
format, otherwise math would
be difficult.
Samples
•Each sample is usually stored as an array of
values, usually normalized between a certain range.
•Normalization shrinks the input values, think of it
like if you would proportionally shrink all the people
to be 0 cm to 100 cm tall.
•This helps ML a lot.
Tasks
•Having defined what datasets and samples are, we
can understand what a task is.
•A task is an instruction that we want to give the ML
model, for example, recognize what digit has been
written.
•This has to also be expressed in a mathematical
way to allow the model to work with it.
Tasks
•In the MNIST example, the task of recognizing the
digits could be translated into minimize the
prediction error.
•Our model will predict what number is indicated
each time, and we will train it by telling if he is
correct or not.
•At the end of the training, our model will hopefully
be able to recognize new digits!
Features
•How will the model be able to recognize the digits?
•It will learn what features are important for it to
solve the task.
•For example, height and eye color are features of a
person.
Similarly, the fact that a digit is all
loopy (like an 8) or straight (like a
1) can be considered digit
features.
The model will try to memorize a
way to recognize those features,
and use this knowledge for every
new input.
When looking at a new digit, he
will check which features are
present, and decide accordingly.
Features
•Of course, the problem is getting the model to
actively look for the correct features.
•For this reason, many possible types of model exist,
each one dedicated to looking at a particular type of
features.
•In the case of MNIST, a good choice would be to
use a model that is designed to work with images.
Separate points - Example
•You have a dataset consisting of points in a 2D
space (meaning that they have an x coordinate and
a y coordinate).
•Each sample in the dataset is encoded as an array
of two elements (features), indicating the x and y
value.
Separate points - Example
•We have a task that requires us to tell whether the
input lies above a certain line, y=1 in our case.
•How would the model work?
•A simple way to solve this problem is to check the
second value of the sample (y coordinate), and
answer yes if it’s greater than 1.
•But what if we don’t know where the line is?
Separate points - Example
•Now our samples have 3 features, x value, y value,
and a bit (0 or 1) indicating whether it lies above or
below the unknown line.
•Now we need to solve the problem in another way.
Are we even sure that the line is
horizontal? What if it is inclined?
What if it’s not even straight?
That’s where the power of ML
comes in, it is a function
approximator.
This means that if the model is
big enough, it will be able to
approximate many different
functions.
Separate points - Example
•In our case the function to approximate is the
unknown line dividing the two sets of points.
•Training the model will then mean letting it learn the
function in an automatic fashion.
•Initially the model thinks of a completely random
line, and then, with trial and error, it fixes it more and
more, getting better predictions.
How is the function stored in the
model?
•Without going too much into details, every model
has an internal representation of the solution.
•Can any problem be converted into a function
approximation task? Yes, almost every time.
•Even our handwritten digit classification task can be
expressed as a function approximation.
The function(s) will tell
the model when a digit is 3
or 7 for example.
But real world problems
like this require to
approximate much harder
functions.
Hard functions!
Functions with millions of parameters, impossible to solve
by hand.
With Machine Learning, we can create models that will be
able to approximate them automatically!
But in order to approximate functions with millions of
parameters, we need millions of variables to tune.
Black box
The model will then be so complex, that
understanding how it works will be a serious problem, it
works but nobody really knows how.
This is where the term black box comes from.
The inside of a model is like a black box, dark and without
any details visible.
Revision
•You should now have a clearer idea on what ML is,
and how it deals with problems.
•The input of a ML model is a dataset, each dataset
consists of samples, each one with features.
•The model then tries to solve the task by modeling
an internal function tuning its parameters by learning
from its mistakes.
Revision – Fact Check
•When looking at a sample, for a ML model every
feature is equally important.
True
False
Revision – Fact Check
•When looking at a sample, for a ML model every
feature is equally important.
True
False
tasks
As I explained to you in the past topic,
various ML models exists, in order to use an approach that
better suits the task given.
In this topic we will look at some of the basic models that ML
can offer, with a small example for each.
This will help you understand when they are used to deal
with real problems.
Trees
Suppose you want to make decisions, for example if you
should take an umbrella with you today.
Easy! If it’s going to rain I’m going to take it, no ML needed!
Well not exactly. What if you are going to take your car?
Trees
And what if you have to take the bus, but there
is a covered path to the bus stop so you won’t get
wet.
Maybe you are not going to stay outside, or
maybe you will!
Of course, you could take a piece of paper and
start writing down all the possible cases,
Elements of a decision tree
And what if you have to take the bus, but there
is a covered path to the bus stop so you won’t get
wet.
Maybe you are not going to stay outside, or
maybe you will!
Of course, you could take a piece of paper and
start writing down all the possible cases,
Elements of a decision tree
•There is a particular ML model
called decision tree, the name comes from the fact
that, when created, it resembles a tree.
•Its objective is to look at the input features (I hope
you remember what those are), and take decisions
based on them.
•Think of an upside down tree: you start from the
top, and at every branch split, you take a decision.
Elements of a decision tree
The top of the decision tree is called root, this is
where your samples start from, going from top to
bottom.
In order to make a decision (final or not) the tree
must have a split, that is a place where
two branches start from.
At the split, a decision is made based on the features
of the input, and depending on the result a different
Elements of a decision tree
•The decision tree continues to move the input down
a certain path, deciding what to do at each split.
•When no splits are available anymore, the input is
said to have reached a leaf.
•At that point, the decision tree will make a decision
based on which leaf the input ended up at.
Elements of a decision tree
•For our example, a simple path would be
Will it rain? If yes…
...are you going to take the car? If yes…
...are you going to stay outside? If yes…
...take the umbrella!
Learning Trees
•This can be adapted to many other problems, like
deciding a loan amount for example.
•The nice property of ML, as with other models, is
the fact that it doesn’t require human supervision!
•It just needs a bunch of data, for example
containing what you did on a particular day and if
you took the umbrella.
Trees and forests
•A decision tree is only that good into predicting
something. It is called a weak classifier, because it
does simple decisions.
•If you combine multiple trees together, all
performing the same task, you instead get
a strong classifier.
•This collection of trees is called a decision forest,
and usually performs much better than a single tree.
Fact Check
•When are decision trees usually used?
When predicting a value, like stock market
When you want to reduce the number of features of
your input
When making decisions based on certain features
None of them
Fact Check
•When are decision trees usually used?
When predicting a value, like stock market
When you want to reduce the number of features of
your input
When making decisions based on certain features
None of them
Although they can be used for other tasks, their
main purpose is to make decisions
Regression – Fitting Curves
What if you don’t want to make decisions, but instead
simplify your data or find trends? Use regression!
Usually this technique is applied when we have data that is
composed by values in the first place.
For example, monetary values, measurements, years etc.,
this is data that uses values naturally.
Regression – Fitting Curves
So how does regression work? It tries to model a function in
such a way that it is able to explain data as best as possible in a
simpler way.
For example, a height chart is a function that explains the
height of a person in an easier way.
Using this function we will then be able to predict values for
new data, and understand why we got those values.
Height chart
For example, we want to predict the height of a
person given its age and gender, to build a height
chart.
Each sample will contain a person’s height,
gender and age, and the regression model will fit a
function to them.
Clustering
Our next model is *drum roll* clustering!
Clustering is the process in which ML model will
try to match together similar samples.
Think about dividing colors into bright and dark
ones, this is a form of clustering.
Clustering
•A clustering algorithm usually knows in advance the
number of clusters that the data will be divided into.
•In the last example we need two clusters, one for
the bright colors and one for the dark ones.
•Each cluster has a center, like the center of a circle.
•When deciding which cluster to assign a sample to,
the algorithm uses the cluster with the closest
center.
•When training, the model moves the centers
around, such that each cluster takes different
Clustering students
•Our model will pick random positions for the cluster's
centers, and will update them based on the distances
with the samples.
•A student will be assigned to the closest center, and
centers will move around to be in the middle of their
respective group of students.
•The training will continue until no more improvements
can be made, and at that point the clusters will be fixed.
Now you have three
groups and three locations
where students can meet!
There are many ways to
do clustering, but the main
idea is always the same.
Fact Check
•Multiple decision trees together are more powerful than
a single decision tree, and they can be considered as:
A weak classifier
A clustering model
A strong classifier
A regression model
Fact Check
•Multiple decision trees together are more powerful than
a single decision tree, and they can be considered as:
A weak classifier
A clustering model
A strong classifier
A regression model
Remember that strong classifiers like decision
forests use a lot of weak classifiers to perform
better!
Deep Learning (DL)
We took a look about decision trees,
clustering and regression, showing how
to solve tasks with them.
Do you remember this Venn diagram,
showing AI encapsulating ML
encapsulating DL?
Deep Learning (DL)
•Deep Learning is the sub-branch of Machine Learning
that studies how to create a new kind of models.
•Those models should try to learn gradually given some
data, similarly on how humans do.
•In this way, researchers believed, new models would be
able to be more robust and capable than existing
approaches!
•It started with the creation
of Neural Networks.
•With this new models researchers
were able to obtain amazing results.
Deep Learning (DL)
•Deep Learning is the sub-branch of Machine Learning
that studies how to create a new kind of models.
•Those models should try to learn gradually given some
data, similarly on how humans do.
•In this way, researchers believed, new models would be
able to be more robust and capable than existing
approaches!
•It started with the creation
of Neural Networks.
•With this new models researchers
were able to obtain amazing results.
Deep Learning (DL)
•The important thing to remember is that DL gives much
more power to Machine Learning.
•It opened many new application opportunities, with new
tasks to solve discovered every day!
•Due to which, Deep Learning is really cool these days in
the AI sector.
Neural Network
•A Neural Network is a model that has been created in a
way to mimic how your human brains work.
Neural Network – The Human Brain
Our brain is composed by millions of neurons.
Each neuron is connected to other neurons through
some filaments called synapses.
Information in the brain goes from one neuron to the
other, travelling as electricity using synapses.
Neural Network – The Human Brain
•When electricity arrives at a neuron, usually nothing
happens.
•This is because every neuron transmits a signal to other
neurons only if enough electricity has been sent at them.
•As soon as the amount of electricity (potential)
overcomes a certain threshold, electricity is released to
nearby neurons.
From neurons to NNs
•Researchers thought that, if our brain is so special,
maybe it’s because of its structure.
•For this reason they tried to implement
a Machine Learning model that would mimic your brain.
NNs (Neural Networks)
Let’s suppose for now that a NN is representing one
neuron only.
The inputs are fed to the model, one at a time, you can
think of them as signals from other neurons.
Each input is splitted into its features and processed by
the model, that will output a value accordingly. How is the
value computed?
NNs (Neural Networks)
Every feature is manipulated mathematically, combined
with other manipulated features and summed into one final
number.
You can think of the manipulation as transferring the signal
through the synapses.
The final number will be the potential (total stored energy)
of the neuron.
We set a threshold! If the number is greater than x output
Fact Check
Select the correct option
•Do you remember what happens when energy is built
into a neuron?
It is immediately released into other synapses.
It is simply blocked.
It is stored until it overcomes the neuron’s potential.
None of the above.
Fact Check
Correct answer is:
•Do you remember what happens when energy is built
into a neuron?
It is immediately released into other synapses.
It is simply blocked.
It is stored until it overcomes the neuron’s
potential.
None of the above.
Fact Check
Correct answer is:
•Do you remember what happens when energy is built
into a neuron?
It is immediately released into other synapses.
It is simply blocked.
It is stored until it overcomes the neuron’s
potential.
None of the above.
More neurons!
•One neuron works for really simple tasks, something
that may be faster to do with pen and paper.
•But what if we apply this principle but with two neurons,
or more?
•We can duplicate this process, and do everything in
parallel, two neurons equals two outputs.
•We can now think of this outputs as an input for another
neuron (or two, or multiple).
This results in a structure
where the first column of
neurons is
called hidden layer and
the second is called
output layer.
More neurons!
•You see, even if we have a lot of hidden neurons (that
are the powerhouse of the NN) still we can’t do many
tasks.
•A way to overcome this problem is to increase the
number of hidden layers.
•In this way a shallow network becomes a deep
network.
•So what can they do? Neural Networks can play
videogames, speak, compose music, recognize objects.
They can tell what your
feelings are from a photo,
transform photos into
others, translate!
Fact Check
Select correct option:
•The part of a Neural Network that gives it the power to
do complex part is
The output layer
The input layer
The hidden layer
Fact Check
Correct answer is:
•The part of a Neural Network that gives it the power to
do complex part is
The output layer
The input layer
The hidden layer
Ethics in ML – Good or bad?
•Now we have self driving cars, everything
is computerized, and ML is everywhere.
•A robot called Sophia got citizenship in Saudi Arabia!
•This raises doubts on the fact that we may be going too
fast with this technology, and it may backslash.
Ethics in ML – Good or bad?
•Let’s talk about how ML can be bad.
Consider a self-driving car, it has to drive and securely
transport his human inside.
But what if a faulty sensor tells it to jump down a
bridge?
Many people are afraid that if you allow ML to enter in
our your lives, trouble would arise.
What about robots used to do other automated tasks
like medical surgery operations? What if an hacker gets
control?
Benefits in using Machine
Learning!
•Take your phone as an example.
•Your translating app uses ML, your keyboard uses ML,
even your camera probably uses ML!
•The advancement of ML and AI also brought new jobs
on the scene.
•Machine Learning consultant, researcher and expert are
now well paid positions on the market.
•And the fact that automation is increasing, the request
for more qualified jobs increases as well.
Benefits in using Machine
Learning!
•With ML learning impaired people can also try to
improve their life quality, with speech and visual aids.
•New entertainment ideas can be brought to life, in new
and improved ways.
Evil ML
Think about an algorithm that allows a robot to learn
how to make tea.
You show the procedure (take the teabag, boil the water
and put them together in a cup) and the robot learns how
to make tea.
But of course, you don’t have to tell him that he doesn’t
have to hurt humans, it’s just a simple tea-making robot!
Evil ML
What if you are standing in front of the kitchen drawer that contains
the cup, and you told the robot to make tea.
How is it going to take the cup, it needs to serve you!
Well, he could simply ask you, or if you didn’t tell it how, push you
on the ground potentially hurting you!
Of course, this is just a simple and funny example, but still, there
are serious implications especially in robotics!
Fact Check
•True or False?
Even a simple task could bring bad unexpected results
if every aspect of it isn’t taken into consideration.
Fact Check
•Correct answer is:
Even a simple task could bring bad unexpected results
if every aspect of it isn’t taken into consideration.
True - Especially when ML needs to interact with
humans
ML - nearby future
•Researchers believe that at this rate, smartphones will
have more and more importance on your lives (even
more than now).
•This translates in more problems to be solved, tasks to
be automated and new fields of research.
•The issue is, smartphones are not as powerful as
laptops and computers, so researchers need to adapt
their models.
Lightweight ML
•Take into account object recognition, something that
ML models can achieve already pretty well.
•It requires a lot of memory to load everything needed,
and a big chunk of processing power is needed
for predictions.
•In order to bring even more ML technology into smart
devices, current models need to be optimized.
Lightweight ML
•Reducing the complexity of a model reduces its
performance, and of course they want the best results.
•This results in compression, pruning and re-design
research, that is already giving neat results.
Neural network compression, for example, allows to
reduce the memory footprint a model, still achieving the
same results.
Pruning removes unnecessary and redundant
parameters, merging them with others.
Re-design can be achieved by learning how models
work, and improving them consequently.
Results, right now!
•For example, there are models that can generate a
description of an input image, in natural language.
•“There is a red car parked in front of some trees”, this is
something that you may get as an output.
•How amazing would be to have a computer that can tell you
in real-time what happens?
•Speeding up models started to be a more involved
competition among researchers, and improvements are made
every day!
•Another field that would benefit from speed improvements
are games, autopilots, art creation and more!
Quantum computing
•Quantum computing is basically a computer that,
instead of using just 0s or 1s, can also use both at the
same time for the same variable.
•Quantum computers have really neat properties that
help certain particular tasks that require randomness.
•Future applications for ML will consist of what is
called Quantum Machine Learning.
Quantum Machine Learning (QML)
•Why Quantum Machine Learning? Because quantum
computers could give a solid base to develop new
approaches.
•And on the other hand, QML could be used to study how
quantum systems work.
•This is the newest and still emerging.
ML – Advance concepts
•There are much more amazing concepts to discover in
ML listed below:
Want to know how computers play video
games? Reinforcement Learning!
A totally different approach on optimization based on
evolution? Genetic Algorithms!
Ways to generate data without any knowledge on what
it looks like? Generative Adversarial Networks!
Thank You
LETS DO PRACTICAL NOW !!!

Weitere ähnliche Inhalte

Was ist angesagt?

Ai history to-m-learning
Ai history to-m-learningAi history to-m-learning
Ai history to-m-learningKyung Eun Park
 
Web 2.0 Collective Intelligence - How to use collective intelligence techniqu...
Web 2.0 Collective Intelligence - How to use collective intelligence techniqu...Web 2.0 Collective Intelligence - How to use collective intelligence techniqu...
Web 2.0 Collective Intelligence - How to use collective intelligence techniqu...Paul Gilbreath
 
A paper presentation abstract
A paper presentation abstractA paper presentation abstract
A paper presentation abstractJagadeesh Kumar
 
ARTIFICIAL INTELLIGENCE Presentation
ARTIFICIAL INTELLIGENCE PresentationARTIFICIAL INTELLIGENCE Presentation
ARTIFICIAL INTELLIGENCE PresentationMuhammad Ahmed
 
Artificial intelligence introduction
Artificial intelligence introductionArtificial intelligence introduction
Artificial intelligence introductionRujalShrestha2
 
artificial Intelligence
artificial Intelligence artificial Intelligence
artificial Intelligence Ramya SK
 
Artificial Intelligence by Jayant
Artificial Intelligence by JayantArtificial Intelligence by Jayant
Artificial Intelligence by JayantJayant Jain
 
DWX 2018 Session about Artificial Intelligence, Machine and Deep Learning
DWX 2018 Session about Artificial Intelligence, Machine and Deep LearningDWX 2018 Session about Artificial Intelligence, Machine and Deep Learning
DWX 2018 Session about Artificial Intelligence, Machine and Deep LearningMykola Dobrochynskyy
 
Artificial Intelligence Notes Unit 1
Artificial Intelligence Notes Unit 1 Artificial Intelligence Notes Unit 1
Artificial Intelligence Notes Unit 1 DigiGurukul
 
Artificial Intelligence for Business
Artificial Intelligence for BusinessArtificial Intelligence for Business
Artificial Intelligence for BusinessNicola Mattina
 
CS 561a: Introduction to Artificial Intelligence
CS 561a: Introduction to Artificial IntelligenceCS 561a: Introduction to Artificial Intelligence
CS 561a: Introduction to Artificial Intelligencebutest
 
Artificial Intelligence power point presentation
Artificial Intelligence power point presentationArtificial Intelligence power point presentation
Artificial Intelligence power point presentationDavid Raj Kanthi
 
Machine Learning: Inteligencia Artificial no es sólo un tema de Ciencia Ficci...
Machine Learning: Inteligencia Artificial no es sólo un tema de Ciencia Ficci...Machine Learning: Inteligencia Artificial no es sólo un tema de Ciencia Ficci...
Machine Learning: Inteligencia Artificial no es sólo un tema de Ciencia Ficci....NET Conf UY
 
Machine learning seminar presentation
Machine learning seminar presentationMachine learning seminar presentation
Machine learning seminar presentationsweety seth
 
Artificial Intelligence Vs Machine Learning Vs Deep Learning
Artificial Intelligence Vs Machine Learning Vs Deep LearningArtificial Intelligence Vs Machine Learning Vs Deep Learning
Artificial Intelligence Vs Machine Learning Vs Deep Learningvenkatvajradhar1
 
OA Introduction to AI from Object Automation
OA Introduction to AI from Object Automation OA Introduction to AI from Object Automation
OA Introduction to AI from Object Automation Object Automation
 

Was ist angesagt? (20)

Ai history to-m-learning
Ai history to-m-learningAi history to-m-learning
Ai history to-m-learning
 
Artificial intelligence
Artificial intelligenceArtificial intelligence
Artificial intelligence
 
Web 2.0 Collective Intelligence - How to use collective intelligence techniqu...
Web 2.0 Collective Intelligence - How to use collective intelligence techniqu...Web 2.0 Collective Intelligence - How to use collective intelligence techniqu...
Web 2.0 Collective Intelligence - How to use collective intelligence techniqu...
 
A paper presentation abstract
A paper presentation abstractA paper presentation abstract
A paper presentation abstract
 
ARTIFICIAL INTELLIGENCE Presentation
ARTIFICIAL INTELLIGENCE PresentationARTIFICIAL INTELLIGENCE Presentation
ARTIFICIAL INTELLIGENCE Presentation
 
Artificial intelligence introduction
Artificial intelligence introductionArtificial intelligence introduction
Artificial intelligence introduction
 
artificial Intelligence
artificial Intelligence artificial Intelligence
artificial Intelligence
 
Artificial Intelligence by Jayant
Artificial Intelligence by JayantArtificial Intelligence by Jayant
Artificial Intelligence by Jayant
 
DWX 2018 Session about Artificial Intelligence, Machine and Deep Learning
DWX 2018 Session about Artificial Intelligence, Machine and Deep LearningDWX 2018 Session about Artificial Intelligence, Machine and Deep Learning
DWX 2018 Session about Artificial Intelligence, Machine and Deep Learning
 
Artificial Intelligence Notes Unit 1
Artificial Intelligence Notes Unit 1 Artificial Intelligence Notes Unit 1
Artificial Intelligence Notes Unit 1
 
Beekman5 std ppt_14
Beekman5 std ppt_14Beekman5 std ppt_14
Beekman5 std ppt_14
 
Artificial Intelligence for Business
Artificial Intelligence for BusinessArtificial Intelligence for Business
Artificial Intelligence for Business
 
CS 561a: Introduction to Artificial Intelligence
CS 561a: Introduction to Artificial IntelligenceCS 561a: Introduction to Artificial Intelligence
CS 561a: Introduction to Artificial Intelligence
 
Chapter 1 (final)
Chapter 1 (final)Chapter 1 (final)
Chapter 1 (final)
 
Artificial Intelligence power point presentation
Artificial Intelligence power point presentationArtificial Intelligence power point presentation
Artificial Intelligence power point presentation
 
Machine Learning: Inteligencia Artificial no es sólo un tema de Ciencia Ficci...
Machine Learning: Inteligencia Artificial no es sólo un tema de Ciencia Ficci...Machine Learning: Inteligencia Artificial no es sólo un tema de Ciencia Ficci...
Machine Learning: Inteligencia Artificial no es sólo un tema de Ciencia Ficci...
 
Machine learning seminar presentation
Machine learning seminar presentationMachine learning seminar presentation
Machine learning seminar presentation
 
Artificial Intelligence Vs Machine Learning Vs Deep Learning
Artificial Intelligence Vs Machine Learning Vs Deep LearningArtificial Intelligence Vs Machine Learning Vs Deep Learning
Artificial Intelligence Vs Machine Learning Vs Deep Learning
 
An overview on ai
An overview on aiAn overview on ai
An overview on ai
 
OA Introduction to AI from Object Automation
OA Introduction to AI from Object Automation OA Introduction to AI from Object Automation
OA Introduction to AI from Object Automation
 

Ähnlich wie Basics of machine_learning

NLP & Machine Learning - An Introductory Talk
NLP & Machine Learning - An Introductory Talk NLP & Machine Learning - An Introductory Talk
NLP & Machine Learning - An Introductory Talk Vijay Ganti
 
NLP & Machine Learning - An Introductory Talk
NLP & Machine Learning - An Introductory Talk NLP & Machine Learning - An Introductory Talk
NLP & Machine Learning - An Introductory Talk Vijay Ganti
 
Hacking Predictive Modeling - RoadSec 2018
Hacking Predictive Modeling - RoadSec 2018Hacking Predictive Modeling - RoadSec 2018
Hacking Predictive Modeling - RoadSec 2018HJ van Veen
 
Artificial Intteligence-unit 1.pptx
Artificial Intteligence-unit 1.pptxArtificial Intteligence-unit 1.pptx
Artificial Intteligence-unit 1.pptxhoneydv1979
 
Machine Learning for Non-technical People
Machine Learning for Non-technical PeopleMachine Learning for Non-technical People
Machine Learning for Non-technical Peopleindico data
 
An Introduction to Machine Learning
An Introduction to Machine LearningAn Introduction to Machine Learning
An Introduction to Machine LearningAngelo Simone Scotto
 
Introduction to Machine Learning
Introduction to Machine LearningIntroduction to Machine Learning
Introduction to Machine LearningCloudxLab
 
Deep learning with tensorflow
Deep learning with tensorflowDeep learning with tensorflow
Deep learning with tensorflowCharmi Chokshi
 
Engineering Intelligent Systems using Machine Learning
Engineering Intelligent Systems using Machine Learning Engineering Intelligent Systems using Machine Learning
Engineering Intelligent Systems using Machine Learning Saurabh Kaushik
 
Artificial intelligence and Legal Profession
Artificial intelligence and Legal ProfessionArtificial intelligence and Legal Profession
Artificial intelligence and Legal ProfessionDharmendra919733
 
Machine Learning Basics - By Animesh Sinha
Machine Learning Basics - By Animesh Sinha Machine Learning Basics - By Animesh Sinha
Machine Learning Basics - By Animesh Sinha Animesh Sinha
 
Unit 1-ML (1) (1).pptx
Unit 1-ML (1) (1).pptxUnit 1-ML (1) (1).pptx
Unit 1-ML (1) (1).pptxChitrachitrap
 
Deep Learning Overview
Deep Learning OverviewDeep Learning Overview
Deep Learning OverviewCloudxLab
 
INTRODUCTION TO ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING
INTRODUCTION TO ARTIFICIAL INTELLIGENCE AND MACHINE LEARNINGINTRODUCTION TO ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING
INTRODUCTION TO ARTIFICIAL INTELLIGENCE AND MACHINE LEARNINGsowmyamPSGRKCW
 
scratch course-part2-2023.pdf
scratch course-part2-2023.pdfscratch course-part2-2023.pdf
scratch course-part2-2023.pdfDoaa Mohey Eldin
 
UNIT 1 IX (1) (1).pptx
UNIT 1 IX (1) (1).pptxUNIT 1 IX (1) (1).pptx
UNIT 1 IX (1) (1).pptxsiddhichaddha2
 

Ähnlich wie Basics of machine_learning (20)

Deep learning
Deep learningDeep learning
Deep learning
 
NLP & Machine Learning - An Introductory Talk
NLP & Machine Learning - An Introductory Talk NLP & Machine Learning - An Introductory Talk
NLP & Machine Learning - An Introductory Talk
 
NLP & Machine Learning - An Introductory Talk
NLP & Machine Learning - An Introductory Talk NLP & Machine Learning - An Introductory Talk
NLP & Machine Learning - An Introductory Talk
 
Hacking Predictive Modeling - RoadSec 2018
Hacking Predictive Modeling - RoadSec 2018Hacking Predictive Modeling - RoadSec 2018
Hacking Predictive Modeling - RoadSec 2018
 
Artificial Intteligence-unit 1.pptx
Artificial Intteligence-unit 1.pptxArtificial Intteligence-unit 1.pptx
Artificial Intteligence-unit 1.pptx
 
Machine Learning for Non-technical People
Machine Learning for Non-technical PeopleMachine Learning for Non-technical People
Machine Learning for Non-technical People
 
tensorflow.pptx
tensorflow.pptxtensorflow.pptx
tensorflow.pptx
 
An Introduction to Machine Learning
An Introduction to Machine LearningAn Introduction to Machine Learning
An Introduction to Machine Learning
 
Introduction to Machine Learning
Introduction to Machine LearningIntroduction to Machine Learning
Introduction to Machine Learning
 
Deep learning with tensorflow
Deep learning with tensorflowDeep learning with tensorflow
Deep learning with tensorflow
 
Engineering Intelligent Systems using Machine Learning
Engineering Intelligent Systems using Machine Learning Engineering Intelligent Systems using Machine Learning
Engineering Intelligent Systems using Machine Learning
 
Artificial intelligence and Legal Profession
Artificial intelligence and Legal ProfessionArtificial intelligence and Legal Profession
Artificial intelligence and Legal Profession
 
Machine Learning Basics - By Animesh Sinha
Machine Learning Basics - By Animesh Sinha Machine Learning Basics - By Animesh Sinha
Machine Learning Basics - By Animesh Sinha
 
Ai lecture1 final
Ai lecture1 finalAi lecture1 final
Ai lecture1 final
 
Unit 1-ML (1) (1).pptx
Unit 1-ML (1) (1).pptxUnit 1-ML (1) (1).pptx
Unit 1-ML (1) (1).pptx
 
Deep Learning Overview
Deep Learning OverviewDeep Learning Overview
Deep Learning Overview
 
Darshana'AI .pptx
Darshana'AI .pptxDarshana'AI .pptx
Darshana'AI .pptx
 
INTRODUCTION TO ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING
INTRODUCTION TO ARTIFICIAL INTELLIGENCE AND MACHINE LEARNINGINTRODUCTION TO ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING
INTRODUCTION TO ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING
 
scratch course-part2-2023.pdf
scratch course-part2-2023.pdfscratch course-part2-2023.pdf
scratch course-part2-2023.pdf
 
UNIT 1 IX (1) (1).pptx
UNIT 1 IX (1) (1).pptxUNIT 1 IX (1) (1).pptx
UNIT 1 IX (1) (1).pptx
 

Kürzlich hochgeladen

"Exploring the Essential Functions and Design Considerations of Spillways in ...
"Exploring the Essential Functions and Design Considerations of Spillways in ..."Exploring the Essential Functions and Design Considerations of Spillways in ...
"Exploring the Essential Functions and Design Considerations of Spillways in ...Erbil Polytechnic University
 
Main Memory Management in Operating System
Main Memory Management in Operating SystemMain Memory Management in Operating System
Main Memory Management in Operating SystemRashmi Bhat
 
Comprehensive energy systems.pdf Comprehensive energy systems.pdf
Comprehensive energy systems.pdf Comprehensive energy systems.pdfComprehensive energy systems.pdf Comprehensive energy systems.pdf
Comprehensive energy systems.pdf Comprehensive energy systems.pdfalene1
 
Prach: A Feature-Rich Platform Empowering the Autism Community
Prach: A Feature-Rich Platform Empowering the Autism CommunityPrach: A Feature-Rich Platform Empowering the Autism Community
Prach: A Feature-Rich Platform Empowering the Autism Communityprachaibot
 
Input Output Management in Operating System
Input Output Management in Operating SystemInput Output Management in Operating System
Input Output Management in Operating SystemRashmi Bhat
 
Module-1-(Building Acoustics) Noise Control (Unit-3). pdf
Module-1-(Building Acoustics) Noise Control (Unit-3). pdfModule-1-(Building Acoustics) Noise Control (Unit-3). pdf
Module-1-(Building Acoustics) Noise Control (Unit-3). pdfManish Kumar
 
70 POWER PLANT IAE V2500 technical training
70 POWER PLANT IAE V2500 technical training70 POWER PLANT IAE V2500 technical training
70 POWER PLANT IAE V2500 technical trainingGladiatorsKasper
 
CS 3251 Programming in c all unit notes pdf
CS 3251 Programming in c all unit notes pdfCS 3251 Programming in c all unit notes pdf
CS 3251 Programming in c all unit notes pdfBalamuruganV28
 
TEST CASE GENERATION GENERATION BLOCK BOX APPROACH
TEST CASE GENERATION GENERATION BLOCK BOX APPROACHTEST CASE GENERATION GENERATION BLOCK BOX APPROACH
TEST CASE GENERATION GENERATION BLOCK BOX APPROACHSneha Padhiar
 
Cost estimation approach: FP to COCOMO scenario based question
Cost estimation approach: FP to COCOMO scenario based questionCost estimation approach: FP to COCOMO scenario based question
Cost estimation approach: FP to COCOMO scenario based questionSneha Padhiar
 
Gravity concentration_MI20612MI_________
Gravity concentration_MI20612MI_________Gravity concentration_MI20612MI_________
Gravity concentration_MI20612MI_________Romil Mishra
 
Secure Key Crypto - Tech Paper JET Tech Labs
Secure Key Crypto - Tech Paper JET Tech LabsSecure Key Crypto - Tech Paper JET Tech Labs
Secure Key Crypto - Tech Paper JET Tech Labsamber724300
 
ROBOETHICS-CCS345 ETHICS AND ARTIFICIAL INTELLIGENCE.ppt
ROBOETHICS-CCS345 ETHICS AND ARTIFICIAL INTELLIGENCE.pptROBOETHICS-CCS345 ETHICS AND ARTIFICIAL INTELLIGENCE.ppt
ROBOETHICS-CCS345 ETHICS AND ARTIFICIAL INTELLIGENCE.pptJohnWilliam111370
 
Forming section troubleshooting checklist for improving wire life (1).ppt
Forming section troubleshooting checklist for improving wire life (1).pptForming section troubleshooting checklist for improving wire life (1).ppt
Forming section troubleshooting checklist for improving wire life (1).pptNoman khan
 
Computer Graphics Introduction, Open GL, Line and Circle drawing algorithm
Computer Graphics Introduction, Open GL, Line and Circle drawing algorithmComputer Graphics Introduction, Open GL, Line and Circle drawing algorithm
Computer Graphics Introduction, Open GL, Line and Circle drawing algorithmDeepika Walanjkar
 
Theory of Machine Notes / Lecture Material .pdf
Theory of Machine Notes / Lecture Material .pdfTheory of Machine Notes / Lecture Material .pdf
Theory of Machine Notes / Lecture Material .pdfShreyas Pandit
 
Stork Webinar | APM Transformational planning, Tool Selection & Performance T...
Stork Webinar | APM Transformational planning, Tool Selection & Performance T...Stork Webinar | APM Transformational planning, Tool Selection & Performance T...
Stork Webinar | APM Transformational planning, Tool Selection & Performance T...Stork
 
Javier_Fernandez_CARS_workshop_presentation.pptx
Javier_Fernandez_CARS_workshop_presentation.pptxJavier_Fernandez_CARS_workshop_presentation.pptx
Javier_Fernandez_CARS_workshop_presentation.pptxJavier Fernández Muñoz
 
A brief look at visionOS - How to develop app on Apple's Vision Pro
A brief look at visionOS - How to develop app on Apple's Vision ProA brief look at visionOS - How to develop app on Apple's Vision Pro
A brief look at visionOS - How to develop app on Apple's Vision ProRay Yuan Liu
 
2022 AWS DNA Hackathon 장애 대응 솔루션 jarvis.
2022 AWS DNA Hackathon 장애 대응 솔루션 jarvis.2022 AWS DNA Hackathon 장애 대응 솔루션 jarvis.
2022 AWS DNA Hackathon 장애 대응 솔루션 jarvis.elesangwon
 

Kürzlich hochgeladen (20)

"Exploring the Essential Functions and Design Considerations of Spillways in ...
"Exploring the Essential Functions and Design Considerations of Spillways in ..."Exploring the Essential Functions and Design Considerations of Spillways in ...
"Exploring the Essential Functions and Design Considerations of Spillways in ...
 
Main Memory Management in Operating System
Main Memory Management in Operating SystemMain Memory Management in Operating System
Main Memory Management in Operating System
 
Comprehensive energy systems.pdf Comprehensive energy systems.pdf
Comprehensive energy systems.pdf Comprehensive energy systems.pdfComprehensive energy systems.pdf Comprehensive energy systems.pdf
Comprehensive energy systems.pdf Comprehensive energy systems.pdf
 
Prach: A Feature-Rich Platform Empowering the Autism Community
Prach: A Feature-Rich Platform Empowering the Autism CommunityPrach: A Feature-Rich Platform Empowering the Autism Community
Prach: A Feature-Rich Platform Empowering the Autism Community
 
Input Output Management in Operating System
Input Output Management in Operating SystemInput Output Management in Operating System
Input Output Management in Operating System
 
Module-1-(Building Acoustics) Noise Control (Unit-3). pdf
Module-1-(Building Acoustics) Noise Control (Unit-3). pdfModule-1-(Building Acoustics) Noise Control (Unit-3). pdf
Module-1-(Building Acoustics) Noise Control (Unit-3). pdf
 
70 POWER PLANT IAE V2500 technical training
70 POWER PLANT IAE V2500 technical training70 POWER PLANT IAE V2500 technical training
70 POWER PLANT IAE V2500 technical training
 
CS 3251 Programming in c all unit notes pdf
CS 3251 Programming in c all unit notes pdfCS 3251 Programming in c all unit notes pdf
CS 3251 Programming in c all unit notes pdf
 
TEST CASE GENERATION GENERATION BLOCK BOX APPROACH
TEST CASE GENERATION GENERATION BLOCK BOX APPROACHTEST CASE GENERATION GENERATION BLOCK BOX APPROACH
TEST CASE GENERATION GENERATION BLOCK BOX APPROACH
 
Cost estimation approach: FP to COCOMO scenario based question
Cost estimation approach: FP to COCOMO scenario based questionCost estimation approach: FP to COCOMO scenario based question
Cost estimation approach: FP to COCOMO scenario based question
 
Gravity concentration_MI20612MI_________
Gravity concentration_MI20612MI_________Gravity concentration_MI20612MI_________
Gravity concentration_MI20612MI_________
 
Secure Key Crypto - Tech Paper JET Tech Labs
Secure Key Crypto - Tech Paper JET Tech LabsSecure Key Crypto - Tech Paper JET Tech Labs
Secure Key Crypto - Tech Paper JET Tech Labs
 
ROBOETHICS-CCS345 ETHICS AND ARTIFICIAL INTELLIGENCE.ppt
ROBOETHICS-CCS345 ETHICS AND ARTIFICIAL INTELLIGENCE.pptROBOETHICS-CCS345 ETHICS AND ARTIFICIAL INTELLIGENCE.ppt
ROBOETHICS-CCS345 ETHICS AND ARTIFICIAL INTELLIGENCE.ppt
 
Forming section troubleshooting checklist for improving wire life (1).ppt
Forming section troubleshooting checklist for improving wire life (1).pptForming section troubleshooting checklist for improving wire life (1).ppt
Forming section troubleshooting checklist for improving wire life (1).ppt
 
Computer Graphics Introduction, Open GL, Line and Circle drawing algorithm
Computer Graphics Introduction, Open GL, Line and Circle drawing algorithmComputer Graphics Introduction, Open GL, Line and Circle drawing algorithm
Computer Graphics Introduction, Open GL, Line and Circle drawing algorithm
 
Theory of Machine Notes / Lecture Material .pdf
Theory of Machine Notes / Lecture Material .pdfTheory of Machine Notes / Lecture Material .pdf
Theory of Machine Notes / Lecture Material .pdf
 
Stork Webinar | APM Transformational planning, Tool Selection & Performance T...
Stork Webinar | APM Transformational planning, Tool Selection & Performance T...Stork Webinar | APM Transformational planning, Tool Selection & Performance T...
Stork Webinar | APM Transformational planning, Tool Selection & Performance T...
 
Javier_Fernandez_CARS_workshop_presentation.pptx
Javier_Fernandez_CARS_workshop_presentation.pptxJavier_Fernandez_CARS_workshop_presentation.pptx
Javier_Fernandez_CARS_workshop_presentation.pptx
 
A brief look at visionOS - How to develop app on Apple's Vision Pro
A brief look at visionOS - How to develop app on Apple's Vision ProA brief look at visionOS - How to develop app on Apple's Vision Pro
A brief look at visionOS - How to develop app on Apple's Vision Pro
 
2022 AWS DNA Hackathon 장애 대응 솔루션 jarvis.
2022 AWS DNA Hackathon 장애 대응 솔루션 jarvis.2022 AWS DNA Hackathon 장애 대응 솔루션 jarvis.
2022 AWS DNA Hackathon 장애 대응 솔루션 jarvis.
 

Basics of machine_learning

  • 1. Basics of Machine Learning By. Assistant Professor – Hasib Masud Shaikh KCCEMSR, Thane
  • 2. What Will I Learn? How Machine learning works. What are some simple applications of Machine learning. What are the ethics of Machine learning. How big is the future of Machine learning.
  • 3. Who is the target audience? People who are progressing their journey towards machine learning Where there is data and it needs to be analyzed, Machine learning is the best way to do so.
  • 4. Benefits of ML Data Science sector is increasing rapidly, so is the demand of people who can write algorithms to analyze that data. With increasing amount of data, the accuracy of the result has to be increased.
  • 5. Introduction Machine Learning, probably one of the most intriguing topics of this decade. Also known as ML, this field is considered one of the pillars of the 21st century. Everyone is talking about it, so there has to be a reason why!
  • 6. Introduction Machine Learning is a field of Computer Science. It tries to teach computers how to learn, so they can help humans solve difficult tasks. The nice thing is that computers should learn without being explicitly programmed to do that.
  • 7. ...because they are dumb Computers just see 0s and 1s, and nothing more. The app you use daily? 0s and 1s. Games we play? 0s and 1s. Everything, from the easiest to the most complex task, is just 0s and 1s (called bits) for
  • 8. Lot of research For this reason ML researchers had to struggle for many years, before being able to find a way to solve this problem. Believe me, a lot of nasty mathematics has been used, scary stuff. But we won’t go in depth about it, understanding the big picture is what’s important now!
  • 9. Artificial Intelligence Machine Learning is a subfield of Artificial Intelligence, usually called AI. What AI does is to teach Computers how to simulate intelligence, in order to behave like a human. Many different aspects of intelligence are studied, each one expanding in a particular subfield.
  • 10. Artificial Intelligence branches To give you an example, following are some of the branches that AI is composed of: Handwriting recognition Speech recognition Translation Spam filtering Automation Game AI Planning
  • 11. Artificial Intelligence branches •Handwriting recognition Where a machine tries to understand handwriting and its properties.
  • 12. Artificial Intelligence branches •Translation Allowing machines to create a bridge between two different languages
  • 13. Artificial Intelligence branches •Speech recognition Similar to handwriting recognition, but with the intent of understanding meaning and sentiment of sentences.
  • 14. Artificial Intelligence branches •Spam filtering Used mostly with emails (but extendable to other domains), where something unwanted is filtered.
  • 15. Artificial Intelligence branches •Automation Reducing the tasks where a human is needed, in order to allow him to concentrate on other aspects of his work.
  • 16. Artificial Intelligence branches •Game AI To be used for non playable characters in games, in order to show more intelligent behavior or better performance.
  • 17. Where is Machine Learning? Machine Learning can be seen as the part of AI that allow computers to take as input some data and makes decisions by learning from it. Every time there is something to learn from data, ML comes to the rescue! It can be something easy like deciding if it’s raining based on the humidity outside. Or it can be something difficult, like telling if a review on a website is positive or negative.
  • 18. What about Deep Learning? Then what is Deep Learning (abbreviated DL) in this figure? It’s a subfield of Machine Learning, that uses Artificial Neural Networks (ANN) to make decisions, a lot more powerful stuff! Don’t worry if you don’t know what those words mean, we will encounter them later in the course.
  • 19. Fact Check Machine Learning teaches computers to learn by themselves, and this isn’t particularly difficult because computers already are quite intelligent. •True •False Machine Learning is a subfield of…? •Robotics •Deep Learning •Artificial Intelligence •None of the above
  • 20. Fact Check Machine Learning teaches computers to learn by themselves, and this isn’t particularly difficult because computers already are quite intelligent. •True •False Machine Learning is a subfield of…? •Robotics •Deep Learning •Artificial Intelligence •None of the above
  • 21. How are we able to teach machines to reason on something? The answer is with numbers. Machine Learning can be seen as an elegant way of formulating real world problems into their mathematical equivalent. In this way a computer, relying only on numbers, can be used to solve those mathematical problems.
  • 22. Sentiment analysis For example, consider the task of saying whether someone is happy based on what he says. Sure, this is a particularly difficult problem, but trust me, scientists found a way of converting it to just numbers. And allowing a computer to try and predict it.
  • 23. Sentiment analysis •I know, you are just waiting me to tell you how to do it. •But it’s nothing simple, this is in fact something that machines can do only in recent years. •Maybe I will give you some details in a future ML course ;).
  • 24. Numbers everywhere •The important thing to remember is that at the core of every algorithm in Machine Learning, there are numbers. •And if it seems too strange to be true, take some simple problems, and try to think about a way to translate it into math.
  • 25. Chess •For example, think about chess. •Just the game, and not how to win, so we keep everything simple. •Normal people may see just a checker pattern, with some wooden pieces on top.
  • 26. Chess •But marvelous miracles of technology(ML) sees math, formulas, bits and bytes in it. •Every piece can be seen as a number, for example 1 for pawn, 3 for bishop, 5 for the rook etc. •Every square can be seen as a coordinate. •Every move could be just a number for the piece and two coordinates for the move.
  • 27. Fact Check Choose correct option: •You need to guess someone’s age based on his gender and height, what is something that doesn’t help the computer in solving the problem? Convert the gender into numbers. Transform the age into words. Convert heights if measured using different units.
  • 28. Fact Check Correct answer: •You need to guess someone’s age based on his gender and height, what is something that doesn’t help the computer in solving the problem? Convert the gender into numbers. Transform the age into words. Convert heights if measured using different units.
  • 29. Patterns •“Ok, Now i know how a computer sees the world, but WHAT does it look for?” •Computers look for patterns in order to solve the majority of their problems. A pattern is something that repeats itself over and over.
  • 30. Which patterns? •For example, consider the past question. •How would a computer do that? •It has to find patterns in the data. •Like the fact that women are usually shorter than men. •Or even that most of the times when people get old they start to bend in front thus reducing their vertical height.
  • 31. From patterns to functions •All this is translated in some kind of function from the ML program (in this field it is called model) •Then it uses the learned functions to solve the given task. •And this is how ML intuitively works :).
  • 32. Fact Check Choose correct option: •If you don’t see a clear pattern in your problem ML won’t be able to help you. True False •Patterns can help you better understand the problem you are solving. True False
  • 33. Fact Check Choose correct option: •If you don’t see a clear pattern in your problem ML won’t be able to help you. True False •Patterns can help you better understand the problem you are solving. True False
  • 34. Fact Check Choose correct option: •If you don’t see a clear pattern in your problem ML won’t be able to help you. True False •Patterns can help you better understand the problem you are solving. True False
  • 35. Fact Check •What can be considered a pattern? The height of people during their lives The number of children born for each month The fluctuations of stock markets Series of moves in a chess game
  • 36. Fact Check •What can be considered a pattern? The height of people during their lives The number of children born for each month The fluctuations of stock markets Series of moves in a chess game Correct answer: All of the above
  • 37. Is ML useful? •The main purpose of ML is to solve difficult problems without the help of humans (well, without much help). •It has been developed in order to overcome the fact that many tasks require knowledge that few people have. •For example, engineers have built chess engines without ML for years but...
  • 38. Is ML useful? •ML allowed the creation of stronger engines. •That don’t need explicit rules from experts to work. •And recently don’t even need to know what game they are playing to beat the world champions!
  • 39. ML Power •This is the power of Machine Learning, and in the last years its role got more prominent. •Tasks are more specialized, requiring more expert knowledge. •The problems are harder to solve. •Thus requiring more powerful methods than the typical handcrafted ones.
  • 40. ML needs data •Everything seems beautiful, but there is a catch. •ML needs a lot of data to be able to work. •Like, thousands and thousands samples, frequently millions. •This was a problem in the past, as not much digital data was available.
  • 41. ML needs data •ML wasn’t considered a very useful tool at the start, because data wasn’t simply enough. •This got the impression to the researchers that handcrafted models were a much more valid alternative.
  • 42. Big Data •Contrary to the past, nowadays we have abundance of data. •More specifically, we are living in the Big Data era. •Big data is a term that indicates the high availability of data of some sort, in our case in digital format. •Images, videos, songs, books etc., everything is largely available to us. •This, along with more powerful computers, led to a new blossom of AI and ML.
  • 43. Big Data - Examples For example, a model that is able to recognize a written digit needs hundred of thousands images of written digits. A model that predicts future sales of a specific product needs years worth of sale records. And so on.
  • 44. In summary •In summary, ML is a very powerful tool, that required years of research to be perfected. •It also needs a lot of data to understand and complete its tasks. •And the world we are living in right now is the perfect environment for it. •That’s why everybody is now talking about it!
  • 45. Spam filtering •Suppose you are working for a company that receives dozen of emails every day. •The boss is tired of checking every email in the inbox, as many of them are just for marketing from rival companies. •He marked every email as wanted or unwanted, and you have to do the same with every new email.
  • 46. Spam filtering – The hard way •You start checking the first email marked as spam, it tells you about this offer of a wonderful new vacuum cleaner. •Ok, no vacuum cleaners. •You check the second email marked as useful, a report of last quarter’s earnings. Important. •The third one is marked as spam, another offer. •You understand that basically all offers are not needed for the company. Got it. •The forth one is from a rich prince that wants to give you all his money. Spam. •Of course strange and shady money offers are not wanted.
  • 47. Spam filtering – The hard way OK! No offers and shady money, spam filter created! But then you find out that the boss also doesn’t want emails from some customers. You continue for months, finding another kind of email that is not wanted and has to be filtered. And you would stay there continuing to add rules to your filter indefinitely.
  • 48. Spam filtering – The easy way - ML A task with an underlying pattern that can be discovered, but that doesn’t need to be “easy” to see. Check. Abundance of data to be used to train our ML model. Check. Powerful computer (although for this task just a normal computer would do). Check. Well, it’s the process in which we take the training data and feed it to the model, chunks at a time. With it, the model will slowly get better and improve its ability to make correct predictions!
  • 49. Spam filtering – The easy way - ML •Of course, we also need to know how to create our model. •With the power of Machine Learning our model will be able to understand which kind of emails to keep and which to trash. •The model will find a way to interpret what we give him, and will soon be able to create a fantastic spam filter! •This is why ML is so hyped nowadays, you don’t need to know what to look for in your data. •You just need to know that it’s there.
  • 50. The easy way – ML - Downside •Of course this comes with a downside, that with complicated tasks we don’t know why and how the model is working. •We know it works, but not exactly what it’s looking for. •This is a current topic of research, in order to create “explainable” models and methods.
  • 51. Power of ML - Tanks and forests •This story narrates how the US government wanted “something” that was able to recognize enemy tanks from images. •Well, this was a perfect moment to show the power of ML! •So a model that takes images as input was built, predicting if the image represented a tank or a forest. •The model was trained on a small set of images, half tanks and half forests.
  • 52. Power of ML - Tanks and forests •It was able to always recognize the tanks! •Scientists were very proud and the government amazed. •But it didn’t last long, as it was found out soon after that the model wasn’t recognizing tanks. •It was checking the brightness of the image! •When new images were used, under other light conditions, the model wasn’t able to get a single prediction right.
  • 53. Power of ML - Tanks and forests •This was an example of how people were disappointed by this new technology. •They expected wonders, and ML didn’t give them to the people. •So they lost trust in it.
  • 54. A comeback •BUT ML DIDN’T STOP. •A few scientists kept believing in it, and results came a decade after. •1985, an Artificial Neural Network (this time created as a computer program) learns to pronounce new words. •In a week, the Network was able to pronounce 20,000 words!
  • 55. Deep learning •2006, the term Deep Learning (DL) is coined, indicating the subfield of Machine Learning that works with Neural Networks. •It uses much more data, and “deep Networks”. •Basically it allows to create more powerful models inspired by the human brain (Neural Networks).
  • 56. ML in detail Are you ready to learn what the concrete aspects of ML are?
  • 57. Datasets •First of all, let me explain concretely what is fed to a ML model in practice. •When you have a task to be done and have the data to use, that data is called a dataset. •Everything the ML model uses is one or more datasets. •An example of a dataset is MNIST, containing tens of thousands images of black handwritten digits on white background. •This dataset is almost always used when learning to create ML models.
  • 58. Datasets •The MNIST database (Modified National Institute of Standards and Technology database) is a large database of handwritten digits that is commonly used for training various image processing systems. •The database is also widely used for training and testing in the field of machine learning. When feeding a dataset to a model, each sample usually needs to be in a particular format, otherwise math would be difficult.
  • 59. Samples •Each sample is usually stored as an array of values, usually normalized between a certain range. •Normalization shrinks the input values, think of it like if you would proportionally shrink all the people to be 0 cm to 100 cm tall. •This helps ML a lot.
  • 60. Tasks •Having defined what datasets and samples are, we can understand what a task is. •A task is an instruction that we want to give the ML model, for example, recognize what digit has been written. •This has to also be expressed in a mathematical way to allow the model to work with it.
  • 61. Tasks •In the MNIST example, the task of recognizing the digits could be translated into minimize the prediction error. •Our model will predict what number is indicated each time, and we will train it by telling if he is correct or not. •At the end of the training, our model will hopefully be able to recognize new digits!
  • 62. Features •How will the model be able to recognize the digits? •It will learn what features are important for it to solve the task. •For example, height and eye color are features of a person. Similarly, the fact that a digit is all loopy (like an 8) or straight (like a 1) can be considered digit features. The model will try to memorize a way to recognize those features, and use this knowledge for every new input. When looking at a new digit, he will check which features are present, and decide accordingly.
  • 63. Features •Of course, the problem is getting the model to actively look for the correct features. •For this reason, many possible types of model exist, each one dedicated to looking at a particular type of features. •In the case of MNIST, a good choice would be to use a model that is designed to work with images.
  • 64. Separate points - Example •You have a dataset consisting of points in a 2D space (meaning that they have an x coordinate and a y coordinate). •Each sample in the dataset is encoded as an array of two elements (features), indicating the x and y value.
  • 65. Separate points - Example •We have a task that requires us to tell whether the input lies above a certain line, y=1 in our case. •How would the model work? •A simple way to solve this problem is to check the second value of the sample (y coordinate), and answer yes if it’s greater than 1. •But what if we don’t know where the line is?
  • 66. Separate points - Example •Now our samples have 3 features, x value, y value, and a bit (0 or 1) indicating whether it lies above or below the unknown line. •Now we need to solve the problem in another way. Are we even sure that the line is horizontal? What if it is inclined? What if it’s not even straight? That’s where the power of ML comes in, it is a function approximator. This means that if the model is big enough, it will be able to approximate many different functions.
  • 67. Separate points - Example •In our case the function to approximate is the unknown line dividing the two sets of points. •Training the model will then mean letting it learn the function in an automatic fashion. •Initially the model thinks of a completely random line, and then, with trial and error, it fixes it more and more, getting better predictions.
  • 68. How is the function stored in the model? •Without going too much into details, every model has an internal representation of the solution. •Can any problem be converted into a function approximation task? Yes, almost every time. •Even our handwritten digit classification task can be expressed as a function approximation. The function(s) will tell the model when a digit is 3 or 7 for example. But real world problems like this require to approximate much harder functions.
  • 69. Hard functions! Functions with millions of parameters, impossible to solve by hand. With Machine Learning, we can create models that will be able to approximate them automatically! But in order to approximate functions with millions of parameters, we need millions of variables to tune.
  • 70. Black box The model will then be so complex, that understanding how it works will be a serious problem, it works but nobody really knows how. This is where the term black box comes from. The inside of a model is like a black box, dark and without any details visible.
  • 71. Revision •You should now have a clearer idea on what ML is, and how it deals with problems. •The input of a ML model is a dataset, each dataset consists of samples, each one with features. •The model then tries to solve the task by modeling an internal function tuning its parameters by learning from its mistakes.
  • 72. Revision – Fact Check •When looking at a sample, for a ML model every feature is equally important. True False
  • 73. Revision – Fact Check •When looking at a sample, for a ML model every feature is equally important. True False
  • 74. tasks As I explained to you in the past topic, various ML models exists, in order to use an approach that better suits the task given. In this topic we will look at some of the basic models that ML can offer, with a small example for each. This will help you understand when they are used to deal with real problems.
  • 75. Trees Suppose you want to make decisions, for example if you should take an umbrella with you today. Easy! If it’s going to rain I’m going to take it, no ML needed! Well not exactly. What if you are going to take your car?
  • 76. Trees And what if you have to take the bus, but there is a covered path to the bus stop so you won’t get wet. Maybe you are not going to stay outside, or maybe you will! Of course, you could take a piece of paper and start writing down all the possible cases,
  • 77. Elements of a decision tree And what if you have to take the bus, but there is a covered path to the bus stop so you won’t get wet. Maybe you are not going to stay outside, or maybe you will! Of course, you could take a piece of paper and start writing down all the possible cases,
  • 78. Elements of a decision tree •There is a particular ML model called decision tree, the name comes from the fact that, when created, it resembles a tree. •Its objective is to look at the input features (I hope you remember what those are), and take decisions based on them. •Think of an upside down tree: you start from the top, and at every branch split, you take a decision.
  • 79. Elements of a decision tree The top of the decision tree is called root, this is where your samples start from, going from top to bottom. In order to make a decision (final or not) the tree must have a split, that is a place where two branches start from. At the split, a decision is made based on the features of the input, and depending on the result a different
  • 80. Elements of a decision tree •The decision tree continues to move the input down a certain path, deciding what to do at each split. •When no splits are available anymore, the input is said to have reached a leaf. •At that point, the decision tree will make a decision based on which leaf the input ended up at.
  • 81. Elements of a decision tree •For our example, a simple path would be Will it rain? If yes… ...are you going to take the car? If yes… ...are you going to stay outside? If yes… ...take the umbrella!
  • 82. Learning Trees •This can be adapted to many other problems, like deciding a loan amount for example. •The nice property of ML, as with other models, is the fact that it doesn’t require human supervision! •It just needs a bunch of data, for example containing what you did on a particular day and if you took the umbrella.
  • 83. Trees and forests •A decision tree is only that good into predicting something. It is called a weak classifier, because it does simple decisions. •If you combine multiple trees together, all performing the same task, you instead get a strong classifier. •This collection of trees is called a decision forest, and usually performs much better than a single tree.
  • 84. Fact Check •When are decision trees usually used? When predicting a value, like stock market When you want to reduce the number of features of your input When making decisions based on certain features None of them
  • 85. Fact Check •When are decision trees usually used? When predicting a value, like stock market When you want to reduce the number of features of your input When making decisions based on certain features None of them Although they can be used for other tasks, their main purpose is to make decisions
  • 86. Regression – Fitting Curves What if you don’t want to make decisions, but instead simplify your data or find trends? Use regression! Usually this technique is applied when we have data that is composed by values in the first place. For example, monetary values, measurements, years etc., this is data that uses values naturally.
  • 87. Regression – Fitting Curves So how does regression work? It tries to model a function in such a way that it is able to explain data as best as possible in a simpler way. For example, a height chart is a function that explains the height of a person in an easier way. Using this function we will then be able to predict values for new data, and understand why we got those values.
  • 88. Height chart For example, we want to predict the height of a person given its age and gender, to build a height chart. Each sample will contain a person’s height, gender and age, and the regression model will fit a function to them.
  • 89. Clustering Our next model is *drum roll* clustering! Clustering is the process in which ML model will try to match together similar samples. Think about dividing colors into bright and dark ones, this is a form of clustering.
  • 90. Clustering •A clustering algorithm usually knows in advance the number of clusters that the data will be divided into. •In the last example we need two clusters, one for the bright colors and one for the dark ones. •Each cluster has a center, like the center of a circle. •When deciding which cluster to assign a sample to, the algorithm uses the cluster with the closest center. •When training, the model moves the centers around, such that each cluster takes different
  • 91. Clustering students •Our model will pick random positions for the cluster's centers, and will update them based on the distances with the samples. •A student will be assigned to the closest center, and centers will move around to be in the middle of their respective group of students. •The training will continue until no more improvements can be made, and at that point the clusters will be fixed. Now you have three groups and three locations where students can meet! There are many ways to do clustering, but the main idea is always the same.
  • 92. Fact Check •Multiple decision trees together are more powerful than a single decision tree, and they can be considered as: A weak classifier A clustering model A strong classifier A regression model
  • 93. Fact Check •Multiple decision trees together are more powerful than a single decision tree, and they can be considered as: A weak classifier A clustering model A strong classifier A regression model Remember that strong classifiers like decision forests use a lot of weak classifiers to perform better!
  • 94. Deep Learning (DL) We took a look about decision trees, clustering and regression, showing how to solve tasks with them. Do you remember this Venn diagram, showing AI encapsulating ML encapsulating DL?
  • 95. Deep Learning (DL) •Deep Learning is the sub-branch of Machine Learning that studies how to create a new kind of models. •Those models should try to learn gradually given some data, similarly on how humans do. •In this way, researchers believed, new models would be able to be more robust and capable than existing approaches! •It started with the creation of Neural Networks. •With this new models researchers were able to obtain amazing results.
  • 96. Deep Learning (DL) •Deep Learning is the sub-branch of Machine Learning that studies how to create a new kind of models. •Those models should try to learn gradually given some data, similarly on how humans do. •In this way, researchers believed, new models would be able to be more robust and capable than existing approaches! •It started with the creation of Neural Networks. •With this new models researchers were able to obtain amazing results.
  • 97. Deep Learning (DL) •The important thing to remember is that DL gives much more power to Machine Learning. •It opened many new application opportunities, with new tasks to solve discovered every day! •Due to which, Deep Learning is really cool these days in the AI sector.
  • 98. Neural Network •A Neural Network is a model that has been created in a way to mimic how your human brains work.
  • 99. Neural Network – The Human Brain Our brain is composed by millions of neurons. Each neuron is connected to other neurons through some filaments called synapses. Information in the brain goes from one neuron to the other, travelling as electricity using synapses.
  • 100. Neural Network – The Human Brain •When electricity arrives at a neuron, usually nothing happens. •This is because every neuron transmits a signal to other neurons only if enough electricity has been sent at them. •As soon as the amount of electricity (potential) overcomes a certain threshold, electricity is released to nearby neurons.
  • 101. From neurons to NNs •Researchers thought that, if our brain is so special, maybe it’s because of its structure. •For this reason they tried to implement a Machine Learning model that would mimic your brain.
  • 102. NNs (Neural Networks) Let’s suppose for now that a NN is representing one neuron only. The inputs are fed to the model, one at a time, you can think of them as signals from other neurons. Each input is splitted into its features and processed by the model, that will output a value accordingly. How is the value computed?
  • 103. NNs (Neural Networks) Every feature is manipulated mathematically, combined with other manipulated features and summed into one final number. You can think of the manipulation as transferring the signal through the synapses. The final number will be the potential (total stored energy) of the neuron. We set a threshold! If the number is greater than x output
  • 104. Fact Check Select the correct option •Do you remember what happens when energy is built into a neuron? It is immediately released into other synapses. It is simply blocked. It is stored until it overcomes the neuron’s potential. None of the above.
  • 105. Fact Check Correct answer is: •Do you remember what happens when energy is built into a neuron? It is immediately released into other synapses. It is simply blocked. It is stored until it overcomes the neuron’s potential. None of the above.
  • 106. Fact Check Correct answer is: •Do you remember what happens when energy is built into a neuron? It is immediately released into other synapses. It is simply blocked. It is stored until it overcomes the neuron’s potential. None of the above.
  • 107. More neurons! •One neuron works for really simple tasks, something that may be faster to do with pen and paper. •But what if we apply this principle but with two neurons, or more? •We can duplicate this process, and do everything in parallel, two neurons equals two outputs. •We can now think of this outputs as an input for another neuron (or two, or multiple). This results in a structure where the first column of neurons is called hidden layer and the second is called output layer.
  • 108. More neurons! •You see, even if we have a lot of hidden neurons (that are the powerhouse of the NN) still we can’t do many tasks. •A way to overcome this problem is to increase the number of hidden layers. •In this way a shallow network becomes a deep network. •So what can they do? Neural Networks can play videogames, speak, compose music, recognize objects. They can tell what your feelings are from a photo, transform photos into others, translate!
  • 109. Fact Check Select correct option: •The part of a Neural Network that gives it the power to do complex part is The output layer The input layer The hidden layer
  • 110. Fact Check Correct answer is: •The part of a Neural Network that gives it the power to do complex part is The output layer The input layer The hidden layer
  • 111. Ethics in ML – Good or bad? •Now we have self driving cars, everything is computerized, and ML is everywhere. •A robot called Sophia got citizenship in Saudi Arabia! •This raises doubts on the fact that we may be going too fast with this technology, and it may backslash.
  • 112. Ethics in ML – Good or bad? •Let’s talk about how ML can be bad. Consider a self-driving car, it has to drive and securely transport his human inside. But what if a faulty sensor tells it to jump down a bridge? Many people are afraid that if you allow ML to enter in our your lives, trouble would arise. What about robots used to do other automated tasks like medical surgery operations? What if an hacker gets control?
  • 113. Benefits in using Machine Learning! •Take your phone as an example. •Your translating app uses ML, your keyboard uses ML, even your camera probably uses ML! •The advancement of ML and AI also brought new jobs on the scene. •Machine Learning consultant, researcher and expert are now well paid positions on the market. •And the fact that automation is increasing, the request for more qualified jobs increases as well.
  • 114. Benefits in using Machine Learning! •With ML learning impaired people can also try to improve their life quality, with speech and visual aids. •New entertainment ideas can be brought to life, in new and improved ways.
  • 115. Evil ML Think about an algorithm that allows a robot to learn how to make tea. You show the procedure (take the teabag, boil the water and put them together in a cup) and the robot learns how to make tea. But of course, you don’t have to tell him that he doesn’t have to hurt humans, it’s just a simple tea-making robot!
  • 116. Evil ML What if you are standing in front of the kitchen drawer that contains the cup, and you told the robot to make tea. How is it going to take the cup, it needs to serve you! Well, he could simply ask you, or if you didn’t tell it how, push you on the ground potentially hurting you! Of course, this is just a simple and funny example, but still, there are serious implications especially in robotics!
  • 117. Fact Check •True or False? Even a simple task could bring bad unexpected results if every aspect of it isn’t taken into consideration.
  • 118. Fact Check •Correct answer is: Even a simple task could bring bad unexpected results if every aspect of it isn’t taken into consideration. True - Especially when ML needs to interact with humans
  • 119. ML - nearby future •Researchers believe that at this rate, smartphones will have more and more importance on your lives (even more than now). •This translates in more problems to be solved, tasks to be automated and new fields of research. •The issue is, smartphones are not as powerful as laptops and computers, so researchers need to adapt their models.
  • 120. Lightweight ML •Take into account object recognition, something that ML models can achieve already pretty well. •It requires a lot of memory to load everything needed, and a big chunk of processing power is needed for predictions. •In order to bring even more ML technology into smart devices, current models need to be optimized.
  • 121. Lightweight ML •Reducing the complexity of a model reduces its performance, and of course they want the best results. •This results in compression, pruning and re-design research, that is already giving neat results. Neural network compression, for example, allows to reduce the memory footprint a model, still achieving the same results. Pruning removes unnecessary and redundant parameters, merging them with others. Re-design can be achieved by learning how models work, and improving them consequently.
  • 122. Results, right now! •For example, there are models that can generate a description of an input image, in natural language. •“There is a red car parked in front of some trees”, this is something that you may get as an output. •How amazing would be to have a computer that can tell you in real-time what happens? •Speeding up models started to be a more involved competition among researchers, and improvements are made every day! •Another field that would benefit from speed improvements are games, autopilots, art creation and more!
  • 123. Quantum computing •Quantum computing is basically a computer that, instead of using just 0s or 1s, can also use both at the same time for the same variable. •Quantum computers have really neat properties that help certain particular tasks that require randomness. •Future applications for ML will consist of what is called Quantum Machine Learning.
  • 124. Quantum Machine Learning (QML) •Why Quantum Machine Learning? Because quantum computers could give a solid base to develop new approaches. •And on the other hand, QML could be used to study how quantum systems work. •This is the newest and still emerging.
  • 125. ML – Advance concepts •There are much more amazing concepts to discover in ML listed below: Want to know how computers play video games? Reinforcement Learning! A totally different approach on optimization based on evolution? Genetic Algorithms! Ways to generate data without any knowledge on what it looks like? Generative Adversarial Networks!
  • 126. Thank You LETS DO PRACTICAL NOW !!!