Transcripts
1. Introduction: Ai, what is it and
how does it work? Hello and welcome
to my A101 Course. My name is Robin, and this is a beginner's introduction to AI when no previous
knowledge is what. This course has been reviewed by British AI professionals to show that the inpatient taught
is accurate and up-to-date. In this course, we'll be
learning about types of AI. What an N, N is. Three progressively
more complex networks, training AI and lead networks. And at the end of the course, I've gotten many
project in mind if we take part to test
your understanding. To listen, concentrate, replay
the video if you have two, because this is a condensed and intense short-course for now.
2. #1 - Types Of AI: There are two types of AI, narrow AI and general AI, also known as AGI. Strong AI. Narrow AI is an AI
that is made for a specific task like predicting
stock price movements. General AI is an AI that can adapt to almost anything
just like a human. However, this technology
has not been developed yet. And there's a three
on the direction of where AI development
is likely to go. Therefore, we will only be
learning about narrow AI.
3. #2 - What is an ANN?: What is an ANN? The brain has 86
billion neurons, which exchange information
with one another at a very fast rate through
small electrical pulses. A given neuron doesn't exchange information with all
the other 86 billion. Rather, they are connected to the surrounding neurons to
perform specialized functions. These structures of neurons are called biological
neural networks. I'll cut the biology
lesson here, but all we need to know is that the brain learns from
it to experience. Each time you learn a new skill, like playing a new video game, the structure of
your neurons change, the way that they connect
to each other changes. This is a really
important part of AI. It's about emulating this
behavior on computers. This is called an
artificial neural network or ANN for short. So when you hear ANN, think for program
that mimics the way nerve cells work in
the human brain. But why do we want to emulate these biological
neural networks? Well, it's simple. They can be automated, easily, use advanced statistics, and
can make fewer mistakes. And humans. This an already trained
artificial neural network with lots of neurons inside. I will explain how ANN are
trained later on this course. But for now, let's figure out how to use this
already trained to n. Well, we give it an input
and then produce an output. So for example,
let's say this N, N is built to describe to blind people their surroundings. So we'll give it a 920 by
180 picture of a house, which means there are
2,072,600 pixels, or in other words,
2,072,600 inputs. And if you're ANN
works properly, it will tell you that
it's looking at house. This is amazing, but it's good to note that this
is a narrower AI. So it can only do
limited functions. It can't do everything. We can just give it anything unexpected to output
anything we want. Let's do another example. Now. Let's pretend we have a different ANN that contain
your voice into words. We say hello. It inputs our audio signal and it has output the word we said without us
needing to use the keyboard. Join me in the next
episode where we'll look inside an ANN and
see how it's built.
4. #3 - AI Scenario I: This is the node which
has a value called X. To build a neural network, remember, we have
to have an input. We'll use a node
to represent this. So the value x in this case is any number
we want it to be. We can connect this
node to another node using a connection
which has a weight. Let's call this weight w. But remember, we
also need an output. This weighted connection is
connected to an output node. This output here is determined upon the input times the weight. And for future reference, here, we're assuming that
the function is linear. If you don't understand
what this means, don't worry, just
keep on listening. This is probably the most simple neural network
you can make. Regardless. Let's test its limitations. Let's say we have a monochrome
picture and we want to invert the shades
so that black turns to white and white terms the black or darker shades tend to lighter shades and light shades tones, dark shades. The first question is, how can we pass the whole
picture into one node, despite the node only
having one value. The last day then we looked
at had over 2 million inputs. And here we only have one. What we do is pass the
pixels in one at a time. Now, moving on, just
using my own intuition, I'm going to assign
values to shade so that black is minus
one and white is one. This is a system
I've come up with on the spot and doesn't
need to be learned. Now, for the weight, we're going to decide that the weighted value is minus one. This, if you think about it, actually makes a lot of sense. As x is times by this weight. When x is times by this weight, the x inverts its value. So if it's negative, it will turn positive
and vice versa. Great. Now let's test it. Let's pretend we have a program that passes
in one pixel at a time. The first pixel is white and therefore
has a value of one. This means that one
times minus one is of course minus one. Great. Now the pixel is inverted. Now the next pixel is dark gray. It too gets inverted. I think you get the point. When you've done all
the pixels There, we have it, an inverted picture. But this simple AI
won't get us very far. It has no predictive
power and is essentially just a function and not really recognized
as the proper AI. Once you've understood
this lesson, join me next time. When we start introducing
multiple inputs.
5. #4 - AI Scenario II: This is an ANN with two
inputs and one output. Notice how I've distinguished
between the first, second x's and w's using
X1 and X2 and W1 and W2. You may be asking, how do we calculate the output,
which x is in W? We need to answer
is all of them. You do x times w
for both of them, then add them up. The equation is this, X1 times w1 plus x2 times w t. Note that if we
had a 100 inputs, we would expand the equation
with plus x three times W3 plus x4 times W4 plus X5 times W5 and so
on and so on and so on. Let's test this AI
within example. A manager of a factory has two employees who work
different number of hours. He wants to ask me how
many shoes will be made in a given working day,
IEEE total work. Jimmy makes one shoe and our will is faster it
makes to choose an hour. The x's will be the number
of hours they work that day. And the wait till
like efficiency or how a node affects the
nodes following it. Let's say Jimmy did
five hours of work. Well, the three hours of work, meaning using the equation, they create a total of 11
choose or 11 total work. Also notice how
even though we'll do two hours less than
Jimmy due to his weighting, is node has more of an
effect on the output. Then Jimmy's. Next lesson, we'll cover multiple inputs and
multiple outputs.
6. #5 - AI Scenario III: This is an ANN, two inputs and two
outputs. Here. I'm going to change the
notation for the last time. This is because we have
duplicate variable names. We have to W1 and W2 is. All I'm doing here is
adding an actual number to distinguish between
the reaction's going to the same node. We're going to use the
same formula as last time, despite it looking a little different due to
the new notation. I hope you can see now why
you've changed the notation. Otherwise, both formula
would be the same, which makes no sense. Now, Let's have the same
scenario as last time. But this time the
manager also wants to know the total wage
he has to pay them. Let's test that out. Let's add some appropriate
weights and inputs. The answer to total output is
the same as last time, 11. The total wage is five times
five plus three times eight, which is of course 49. This ai is starting
to become more useful as it allows
the manager to visually compare the total work to the wages by
changing weights. Here's a question for you. The input values are the number of hours
they individually do. The second output node represents
the total wages given. What did the weight to one
and wait two to represent. Have a think and find
our next lesson.
7. #6 - Training AI: In this lesson, we'll be
learning about training AI. But before we do so, the answer to the question in the last lesson
is that weight to one and weight to 2% Jimmy's and wells pay
per hour respectively. Now let's get back on
topic training AI. Last time I mentioned that
the manager can change the weights to see how the total work and
total wages change. However, there are some
problems with this. If there were 100 inputs
go into 100 outputs, It's practically impossible to go through and change
all the weights. In response to this problem, the manager might decide
to train this ANN, the weightings of salary
is already know for certain because it's just
how much he pays them. However, the weightings
w1 one and W12, which is their work power, are not known when
building this. And then what we
do to train an ANN is to give it lots of
examples, inputs and outputs. And it will automatically
figure up on its own weightings that give realistic and even
predictive outputs.
8. #7 - Layered Networks: This here is a neural network with four inputs
and four outputs. But instead of the
inputs connecting directly to the outputs, there's an intermediate layer. This is what we call
a hidden layer. And a lot of the time
when using an AI, you don't know how the weights the AI is
producing its outputs. Hidden layers, add complexity to the AI and allow it went and trained to produce
more desirable results and do more complex activities. For example, let's
set up a scenario. Let's say we have a car in the
center for maize trying to get away from the center of
the maze as far as possible. The car has three sentences which input three
numbers into our ANN, which tells the n, n, how far the sensors are. All the chi is on different
sides from the wolves. The car also has a receiver, so it knows how far away
it is from the center. We can let the air
I have control of the car by the accelerator, breaks, wheels, and reverse. Notice that each
input and output only takes or events
a single value. For example, the steering wheel could be a single
value by having 0 at resting position minus numbers to turn left and
positive numbers, right. The hope with this Ai
is that the car makes its way by itself
away from the center. Whether or not this
actually works depends upon how
you've built the AI, including how and what the training data
you've given it is. And that's it. See you for the final lesson where I talk
about the class project.
9. Class Project: Hello, I'm well
done on completing my AI one-on-one course. I hope you've learned
something and hopefully you understand the AI better and
want to pursue it further. I've got a class, a
challenge for you. I want you to draw
digitally or on paper your own AI
with its connections. It can have as many inputs
and outputs as you want. And the hidden
layer if you wish, then I want you to come
up with your own imagine a scenario that you
can use your AI for. Just let my car in the maze. Then I want you to label the
inputs and output nodes. According to your scenario. Feel free to add any extra
explanation and get creative. More details about the class
project in the description. After you've done upload
your class project, and I'll have a look
at it personally and give you some feedback. Thank you for
watching and goodbye.
Robin
