YouTube Transcript:
AWS Certified AI Practitioner (AIF-C01) – Full Course to PASS the Certification Exam

Skip watching entire videos - get the full transcript, search for keywords, and copy with one click.

AutoDub

Understand YouTube Foreign Videos

Immersive YouTube Voice Translation

Break language barriers, embrace global quality content

Solve Foreign Video Barriers Instantly

Video Transcript

Video Summary

Summary Requirements

Core Theme: The content is a comprehensive overview of the AWS Certified AI Practitioner (AIFC01) certification, detailing its curriculum, exam format, and the foundational AI/ML concepts covered, with a strong emphasis on AWS services like Amazon Bedrock and SageMaker.
Key Points:
- The course prepares individuals for the AWS Certified AI Practitioner (AIFC01) certification, covering traditional ML, managed AI services, and Generative AI/LLMs.

Mind Map

Click to expand

Click to explore the full interactive mind map • Zoom, pan, and navigate

hey this is Andrew Brown your favorite

Cloud instructor bringing you another

free Cloud certification course and this

one is the aabus certified AI

practitioner also known as the aif c01

and the way we're going to get certified

is we're going to be doing lecture

content hands on labs and as always I

provide you a free practice exam so you

can as that exam uh put it on your

resume or LinkedIn to go try to get that

job you've been looking to get uh if you

like courses like this one the best way

to support it is by purchasing the uh

optional paid materials on exampro doco

for this course it's at aif c01 this is

where you're going to get uh additional

practice exams uh cheat sheets um

downloadable lecture slides and more um

and you know if you do not know me I've

taught a lot of courses um so you know

Microsoft AWS gcp terraform kubernetes

you name it I've taught it and so

looking forward to jumping into the AI

okay hey this is angre brown and we're

at the start of our journey asking the

most important question first which is

what is the AI practitioner so this is

an AI certification teaching you the

foundational knowledge of AI Cloud

workloads adus offerings around

traditional ml pipelines adabs offerings

around managed AI Services offerings

around gen and large language models the

course code here is the AI aifc one so

make sure you check the course code so

that you know that you're using the

latest course uh consider the

certification if you want to become an

AI engineer or a data scientist or you

have to work with um AI stuff in your

developer job if you don't know what an

AI engineer is it's someone that builds

AI Solutions using manage AI Services it

could also be building uh ml pipelines

or working with data scientists um to

some degree you will want this uh

certification if you're looking to

architect business use cases for ml a

geni this certification is more focused

on the SE suite and decision makers to

help them buy into the ad ecosystem for

AIML but I'm going to cram in a bunch of

developer stuff because I know that

people want to do this for real and not

just talk about it um if you enjoy the

following tasks like stats and matths

working with data working with python

then this is a career path for you if

you don't you better watch out here

because this stuff creeps up on you uh

unexpectedly but for generative AI it's

not so much an issue here's our Ana

certification road map and you know

again this is just a suggestion you can

do these in any order that you want but

I strongly suggest that before you do AI

practitioner do the cloud practitioner

because a lot of those skills are

expected uh for this okay um and I just

need to remind you that ad

certifications do not validate

programming technical diagramming code

management and many other technical

skills that are required for obtaining

technical roles so do not assume that

when you get aert you can do the job

it's part of your Learning Journey yes

of course but uh you need to really make

sure that you can do the skills um how

long will it take to pass well if you're

beginner 20 hours if you're experienced

five hours this is not a hard

certification I probably made the

content harder than it had to be but I

want to prep you you know for your rules

in actually being able to do this stuff

you're looking about 10 hour average

study time spend half your time with

lecture in Labs other half with practice

exams recommending you study one or two

hours for possibly 14 days days again it

won't take long to get through this

course watch the lecture content do the

Hands-On Labs now this certification

doesn't require any hands-on experience

but I really think that you should do it

because uh in practice versus on paper

are to completely do different things

the labs are not hard here and it will

really help cement your knowledge and in

some cases I'm keeping the lecture

slides light because we're going to be

doing the lab so even if you don't do

them watch what I do so you get at least

that experience there get paid practice

exams because this one has new exam

question types uh and people said that

uh it was it threw them off right so um

you know you can go over to the exam Pro

platform get your free practice exam we

also have paid ones um and you can find

that aif c01 buy those paid ones support

more of this free content we really

appreciate your support and this stuff

is hard to make so um let's talk about

the domains for the exam there are five

domains each domain has its own

weighting this is deter how many

questions that will show up in domain uh

so for domain one we have fundamentals

of AI and ml domain two we have

fundamentals of gen domain three we have

apps applications of foundation

Foundation models I love just saying

applications is apps and by the way this

is not a spelling mistake I copy and

paste it it's Foundation models but

foundational models is also correct um

domain four is guidelines of responsible

Ai and domain five is security

compliance governance AI Solutions which

there's not a lot to talk about so they

really over emphasize it when there's

not much to say but these two categories

is all gen so I put a lot of gen in this

course Amazon Bedrock is done end to end

for this um so you're you're in really

good shape I probably have the best

course for um for the AI practitioner

for the Bedrock stuff okay sag maker uh

I do an okay job of it sagemaker um used

to be sagemaker Studio Classic and

they've migrated over to this new

experience which is not very good and so

you know I'm I'm kind of grumpy when

making the content for sage maker

because I miss the old experience and I

I I think ad has kind of U not done a

good job re reimagining that solution

but anyway where do you take this exam

in person or online uh adabs uses

Pearson view uh for their proctored

online exam system and also for their uh

their test Network PSI is gone if you

remember PSI from a long time ago ads is

not using them anymore the experience

with PSI hasn't been great but I also

think the reason why us is going with a

single provider now is just that they

can leverage that platform to its

maximum and add new new features like

exam question types which we'll talk

about in a moment a

proctor uh is someone that watches the

exam so the idea is they're there to

make sure you do not cheat so understand

that is a component in the test

experience the grading here is 700 out

of a th000 for a passing score I put an

aster there because it's around 70%

because it must uses scaled scoring you

could technically fail at 70% right so

always aim for higher um there are 65

questions on this exam 50 scored 15

unscored you can get uh 15 scored

questions wrong there's no penalty for

wrong questions format of questions

multiple choice multiple answer but also

uh ordering ordering matching and case

studies for this exam for sure so um

right now the exam is in beta at the

time to make this video so they might

change and get rid of those questions

because people don't like them but

understand that a US is trying new exam

type questions you'll experience this

our platform simulates them so you'll be

in good shape if you use our practice

exams not all providers can even similar

things like case studies we absolutely

have that in Spades and we've been doing

that well before this so it was just

coincidental that Aus decided to do that

uh 50 questions are on the exam or

unscored they will not count towards

your final score why are they unscored

unscored questions are used to evaluate

the introduction of new questions to

determine if the exam is too easy in the

passing score or questioned uh

difficulty needs to be increased

discover users who are attempting to

cheat Okay so there's lots of reasons

why they do this if you encounter

questions you've never studied for that

seem really hard keep your cool remember

they may be unscored questions the

duration of the exam is 2.5 hours you

get about 1.5 minutes per question um

there is 120 Minutes with 150 minutes

seat time seat time refers to the time

you should allocate for the exam this

includes reviewing the instructions

showing up for online uh Proctor uh on

showing up for the Proctor to look at

your workspace reading accepting the NDA

complete the exam provide feedback at

the end if it seems like I'm tired it's

cuz I shot this three times my

microphone wasn't on and so my voice is

kind of wearing out but we'll get

through this okay the uh the exam is

valid for 36 months and three years

before recertification I don't really

know that for certain because at the

time of this exam they didn't say that

but the general rule is that search for

8 us is always three years if you're

going to get recertified it's going to

be um You probably get it for free

through a of a skill Builder they're

always trying to do that let's have some

real talk talk about certifications I

have to remind you that cloud

certifications expect you to have

foundational technical skills like

programming scripting SQL it networking

Linux Windows servers project management

developer tools app development skills

compi algorithm skills and more if you

do not have these skills and you get

these searchs you cannot do the job

right this only teaches you how to do ml

AI on the adus platform but it's missing

a lot of stuff uh and adus likes to

position this certific a as a fund fun

fundamental exam but I find there's tons

of gaps with this one I'm producing my

own uh uh foundational generic uh a gen

certification to really fill the gaps

here but you know to fill the gaps

leverage fore Cod Camp their large

catalog of General technical content and

we at exam Pro also make additional uh

materials uh beside the certification to

really help you there that's only

available in the

subscription okay it itself does not

care about ad certifications for hiring

for their own technical roles

certifications serve as a structured way

of learning with a goalpost now

originally certification actually

mattered back in 2016 17 if you had Ana

certification we're talking about it

companies took notice but now it's more

of a learning path thing nurse

certifications can be more valuable so

the reception of the partitioner might

be more valuable uh but I don't know at

this point so I don't want to give you

false hope and considerations but you

still it's good to learn this

and and stuff like that understand that

you might need to add 250 to 500 hours

beside the certification to have the

developer knowledge to perform the stuff

here or AI knowledge if you will so you

know again just consider there's

additional work to be done if you want

to work as an AI engineer or data

scientist um we are going to add

Hands-On labs to help you fill the gaps

here so if you see me taking due TS and

it seems like we're doing long labs I'm

trying to help you out here you can

watch them and not do them if you want

but really you're really should do them

because I'm giving you Real World skills

here here and you folks keep saying that

you want it so I'm giving it to you so

some of the labs might even uh end up in

failed implementations uh not for this

certification I think there was only

like one that I that that was bust and

it wasn't my fault it was just yeah I

think we're trying to do fine t on

Amazon bedrock and so it just wasn't

clear the spend and I just did not want

to end up with a $5,000 bill or

something crazy so I did show the

process and I I did tell you that but uh

this one has next to no failures it's

just the one there but understand that

it's it's about seeing the problems

seeing what's worth using seeing what's

not worth using because these

certifications are marketing tools to

convince you to utilize them but I'm

here as your uh Community hero and I

actually am an aist Community hero to

tell you the real truth about these

services and which ones you should use

and and maybe avoid and I want to be

really clear with that uh we do try our

best to clean up infrastructure but you

should always be proac and check if the

resources are running you're responsible

for the cost and spending your ad

account in the

adus um practitioner course I show you

budgeting and stuff I'm not showing you

in this course but I do in this one and

by the way in this course I actually had

unexpected spend I usually don't have it

but I had it with stagemaker canvas it

was like almost $400 $500 Canadian

because it's US dollars converted

afterwards and uh it's just one of those

Services where they really really misled

you uh not intentionally but like

because the UI was so bad and I I really

pointed out and I even tell you don't

use stagemaker canvas and just watch me

do it so be very careful with Spen but I

do my best but you are responsible just

remember that [Music]

[Music]

okay hey this is Andrew Brown I'm on the

adabs training certifications pages and

we're looking at the adabs certified a

practitioner I do want to point out that

right now the exam is in beta um so

generally I would recommend for you to

wait for to go out of beta because beta

means that the exam questions are going

to change and often beta is for testing

to see whether the exam is good or not

not really forgetting that validation

but anyway if you want to go sit it

early you can but again my

recommendation is to wait the um exam

guide is very unlikely to change so uh

itus doesn't usually change too much

from the the beta experience it's more

about the exams so this is going to be

fine but let's scroll on down and take a

look here and see what they're

recommending so familiar with adus core

Services um share responsibility model I

am Global infrastructure this is all the

stuff that gets covered in the um adabs

Cloud petitioner so you should have your

Cloud petitioner before proceeding for

this certification things you don't have

to do develop coding ml algorithms

implementing data engineering feature

techniques hyper parameters building

deploying AI pipelines conducting math

or statistics basically nothing you

don't have to do any Hands-On but I'm

going to tell you I I pack back in

Hands-On stuff because I think that if

you do some Hands-On that it's going to

really help cement that information your

head there's no reason not to do it um

because you know we can read something

on a paper but it has nothing to do with

what's actually happening so you should

do Hands-On labs and I have hands on

labs for you I have a lot around Amazon

Bedrock just because I feel that that

was um should have been more

strengthened in this certification and

so or just knowledge in general because

it's such a large product but I spend a

lot of time in Bedrock let's scroll down

here and take look so we have multiple

choice multiple response ordering

matching case studies these three are

new not new if you're from Azure because

that seems like a similar thing from

over there um but yeah these are new

question types we have uh 15 unscored

questions we uh so we'll continue on to

here the results is between 100 and

1,000 with a minimum passing score of

700 okay your you score report can

contain tables of classifications of

performance which I'm not really

interested in we'll scroll on down let's

take a look at the domains we have

fundamentals of AIML fundamentals of gen

applications of foundational models

guiding of responsible Ai and then

security so we'll take a look here and

they uh rattle off a bunch of different

terms and so I do my best to cover as

much as I can here the problem is is

that it's not very succinct or exactly

what it is that they want you to know

and because we're right now in betas I

don't know exactly what's going to show

up on there but I did a lot of coverage

here have as much as I that

can over here we have recognize app AI

workload so they're just talking about

like when you should use them when you

should not use them do you know all the

manage services and we cover all those

manag Services here then we're talking

about Sage maker and the ml pipeline all

the steps and and all the core sagemaker

features and services you should

know um then there's about model

performances so I give this a bit of

extra time in the course just because

you know they become valuable later down

the road but they're not super technical

um so you're not going to have hard time

with that for Gen we have a lot of stuff

and I I really really dot the eyes on

gen not because well it's it's not like

I'm huge about gen but just I just

happen to be building a lot of J

projects so I just was able to pack in a

lot of good stuff here and I think a lot

of companies again this is where their

focus is going to be when they're taking

the AI practitioner um so a lot of

information on that then you know

they're talking about more of the other

services like Party Rock Bedrock

playground Amazon Q by the way Amazon Q

is a terrible terrible product um ad us

keeps telling me like oh it's it's new

and improved every like two weeks and I

I come back and doing it's just garbage

I don't know why they keep promoting it

but I guess they've invested a lot of

energy into that product and

unfortunately it's just not very good um

so sorry I don't have anything nice to

say of it maybe in the future it's

better but every time I look at it it's

bad applications of foundation model so

yeah we're talking about not just

Foundation models but just types of what

do they call this application Foundation

model but yeah just general generative

AI knowledge it's weird that they just

have this here because it's basically

that section as well uh and then

responsible AI you know there's isn't

there isn't a whole lot to say about

responsible AI it's kind of weird that

it has uh so much attention here but we

literally just spend one video on it and

there's three other videos of services

to look look at but um you can pretty

much guess like what is responsible and

what's not so it's not like that hard

we'll go down below here um you know not

a whole lot to talk about security I

mean they listed a bunch of stuff in

here but some a lot of the things that

they were listing don't even exist yet

so um and not I don't mean because it's

a beta I just mean like they're talking

about things that just don't exist like

or they haven't been implemented so you

know again I think that this is just

adabs is not not doing a very good job

putting together these exam guides as

they used to they're really throwing a

lot of stuff at the wall here but that's

okay I'm going to make sure you come

through this uh uh pretty well here with

no problems there's the appendix of a

lot of services and what is in scope and

what's out of scope so here you can see

a bunch of services here um you know and

not all of them are in the course but I

I listed the ones that I thought were

most relevant and what my experience was

and what logically made sense and uh

yeah so there you [Music]

[Music]

go hey this is Andrew Brown and we are

taking a look at the definition of what

is artificial intelligence and we really

want to put this against uh the terms of

machine learning deep learning and

generative AI so that it's very clear

what the differences are often people

just say AI when they mean ml or deep

learning so understand that um these

terms are uh not used correctly often

but people generally will understand

what you're trying to say so it's not a

big deal if you use them out of turn but

let's make sure that we know what they

are let's so let's first take a look

here at artificial intelligence also

known as AI these are maches that

perform jobs that mimic human behavior

okay that's the key thing here is that

they are humanik or doing tasks that

you'd expect a human to do um and that

is clearly a very broad term of what is

AI and so you can see why a lot of

things are attributed to being AI then

you have machine learning and machine

learning initialized as ml is machines

that get better at a task without EXP it

programming now of course we have to

code a machine learning model but once

we have that model and we pass things

into it it's able to complete its task

with its very complex algorithms um so

you could also just think of it as an uh

it's a a a special algorithm to perform

a task that would negate you the negate

you having to do calculations or

programming or things like that then we

have what is deep learning and when we

think of a lot of the AI stuff we're

usually thinking of deep learning

because it's these machines that have an

artificial neural network inspired by

the human brain to solve complex

problems so you probably have this uh

you probably seen a graphic of it of

like these nodes and they're

interconnected and they go through

layers that's deep learning a lot of

people call that machine learning or AI

but no that's that's the L then we have

gen so gen which is more of a u

marketing term but generative AI is a

specialized subset of AI that generates

out uh content such as images video text

and audio now I don't have it in the

graphic on the left because it's hard to

say where it is a go does it go here

right because it is a subset of AI but

technically um gen often utilizes deep

learning because when we think of it and

my Line's not dry here today but um when

we think of it there we go there's the

line is that a lot of gen techniques

like large language models or um or um

Vision models things like that are

utilizing neural networks so it is deep learning

learning [Music]

[Music]

okay all right so I know we keep talking

about what is AI what is Gen AI but

we're going to cover it again just so

that it becomes more clear from

different perspectives um so let's talk

about what is artificial intelligence so

AI is computer systems that perform

tasks typically requiring human

intelligence um so these include things

like problem solving decision making

understanding natural language

recognizing speech and images and an

ai's goal is to interpret analyze and

respond to human actions it's there to

simulate human intelligence in machines

when we use the word simulate we're

talking about mimics aspects resembles

behaviors but what we're not talking

about is emulation which is replicating

exact processes and mechanisms that's if

you created literally a ual human brain

that's what emulation would be um so AI

applications are vast and include areas

such as expert systems natural language

processing also known as NLP speech

recognition robotics uh and more AI is

using various Industries for tasks such

as uh we're talking about business to

Consumer so think of a customer service

chatbot if we're looking at e-commerce

think of a recommendation system if

we're talking about the Auto industry uh

maybe we're we're looking at automous

Vehicles if it's medical then medical

diagnosis there's a lot of applications

for AI but it's a broad application for

all sorts of things now let's take a

look at generative AI so generative AI

uh often initialized as geni or or said

as geni is a subset of AI that focuses

on creating new content or data that is

novel and realistic it can interpret or

analyze data but also Al generate new

data itself it often uh yeah so like

types of content produces would be text

images music speech and other forms of

media it often involves Advanced machine

learning techniques uh so it could be

using things like Gans it could be using

vae so variational Auto encoders um a

lot of current llms use the Transformer

architecture so if you're using um chat

GPT or Claud Sonet or any of the popular

ones they're basically all Transformer

architectures gener I has multiple

modalities and when we say modalities

it's like think about your your senses

you have touch taste hearing smell so

modalities are the kinds of content or

or um senses that a model has so we have

Vision so realistic images and videos

text generating humanlike text audio

composing music molecular which is more

of an interesting one so drug Discovery

via geomic data and uh I want to make it

clear again we're talking about large

language models but llms large language

models will generate out humanlike text

and is a subset of gen it's just one

modality of the many modalities um but

it's often conflated as being AI or gen

AI just because it's the most popular

and In Demand right now and the most

developed so just make sure that you

understand that geni and AI is not all

about large language models it's just

one modality one application of of the

broad sense of AI and gen AI now let's

just make sure we have a side by-side

comparison uh and then I'm sure after

this you'll definitely know uh

definitively the difference between Ai

and gen so in terms of functionality AI

focuses on understanding and decision

making whereas gen is about creating new

and original outputs for data handling

AI analyzes and makes decisions based on

existing data gen uses existing data to

generate new and unseen outputs in terms

of applications AI spans across various

sectors including data analysis

automation NLP and Healthcare where gen

and yes I see the spelling mistake uh

it's creative and Innovative focusing on

content creation synthetic data

generation defix and design so there you go

go [Music]

[Music]

let's talk about Jupiter so Jupiter

notebook is a web-based application for

authoring documents that combine Live

code narrative text equations and

visualizations and before it was called

jupyter notebook it was known as I

IPython notebook and jupyter notebooks

were overhauled and then turned into an

ID called Jupiter lab which we'll talk

about here in a moment but you generally

want to open notebooks in Labs um and

the leg the Legacy web-based interfaces

known as Jupiter classic notebook and to

be honest I get confused between juper

lab and classic I think most things that

you use these days are Jupiter lab um

but the confusion is because we just

call them notebooks even though Jupiter

classic notebook is the not nope the uh

the older one and the newer one is

Jupiter Labs let's go take a look at

jupyter Labs so jupyter lab is the next

Generation webbased user interface it

has all the similar features as the

classic juper notebook in a flexible and

more powerful user interface so it has

notebooks terminals text editor file

browser uh Rich outputs and the way you

I think that you know that you're using

jupyter lab is that it will have this uh

these tabs here on the side and a bunch

of functionality so Jupiter lab will

eventually replace the classic jupyter

notebook and that's kind of true because

um but not fully because in some places

I do come across classic notebooks

launching them up um but for the most

part functionally it has been

replaced then we have jupyter Hub so

jupyter Hub is a server to run jupyter

labs for multiple users it's intended

for a class of students a corporate data

science group scientific research groups

and so it has some components uh

underneath you will come across notebook

like experiences that are like Jupiter

Labs so some companies will um extend

the functionality of it one example is

Sage maker uh Studio Classic for

whatever reason ad us um spent all this

time creating extensions and extending

Jupiter lab and then they decided uh no

we're not going to have extensions

anymore and we're just going to use the

vanilla version um but uh there's also

things like vs code that has notebooks

or code lab that have notebooks and vs

code is like its own kind of notebook

thing it's not juper Labs but it's juper

lab compatible so just understand that

you'll come across things that are

notebooks that look like Jupiter lab but

they're not necessarily Jupiter lab okay [Music]

[Music]

let's take a look at natural language

processing also known as NLP and in

machine learning it's a technique that

can understand the context of a corpus a

corpus is a body of related text the

text that you are working with and NLP

intersects with computer science and

Linguistics so if you know a lot about

the the nature of uh spoken and written

language then uh computer science here

is going to meet in the middle here so

that we can um make sense of it using

algorithms so NLP enables us to do

things like analyze and interpret text

within documents emails and messages

interpret or contextualize spoken texts

like sentiment analysis synthesiz speech

uh such as using a voice assistant

talking to you automatically translate

spoken or written phrases and sentences

between languages in uh interpret spoken

or written commands and determine

appropriate actions another thing you'll

hear a lot is language understanding

which is supposed to be it's a it's more

like a specialized subset of NLP um uh

that just goes farther to understand uh

more traditional older ways of doing NLP

but uh anyway what I'll do is we'll just

take a look at this um very simple

flowchart to give you some idea of

things that are related with an NLP this

is mostly just get you exposed to some

terms it's not important to remember

what these are and I can't even describe

them off the top of my head um but again

just get you exposure to NLP terms so

that when you see them later you'll go

look up and be like oh I remember seeing

that term here so here we have like text

wrangling pre-processing language

understanding so structure and syntax

processing functionality which is what

the NLP uh does for you in the end but

text text Rand pre-processing is where

you are preparing uh text to be uh put

into possibly um a machine learning

model or maybe you're using it for um

some kind of analysis or something like

that and so this is basically taking

text and um formatting it changing it

and so what could we be doing here well

we could be doing conversions maybe

we're lower casing things maybe we're

upper casing things um maybe we're

turning contractions into their full

forms or vice versa sanitation this is

where you are maybe stripping out HTML

or special characters or you are

removing stop wordss when uh you have

stop wordss later on in your ml models

tokenization which is conver converting

um the text into uh Vector embeddings we

have stemming okay we have uh lonization

so there's a lot of things here but you

can see it's mostly just like formatting

the text to be utilized for something

else we have language understanding so

these are processes to make sense of the

text so part of speech tagging so is

this an adjective is this a noun things

like that chunking how can we uh break

up the text and then work with those

chunks later on down the road so that

still makes sense dependency parsing so

you know which word relies on other

words and what relationships do they

have to other ones uh

consti constitu parsing very hard for

word for me to say but like imagine a um

a a tra GRE tra green and so like you

know a noun has an adjective under it

which has another thing under it you

look up if you look it up and go to

Google Images you'll you'll know what

I'm talking about then we have

processing functionality what are we

using NLP 4 so we have name and

recognition this is where you have a

body of text and it's highlighting uh

important words like maybe important

nouns that it thinks you you care about

or things like that or personally

identifiable information we got engrams

sentiment analysis is this text positive

negative happy sad information

extraction what are we trying to get out

of a large body of text yeah um same

thing with information retrieval

questioning and answering topic modeling

so you know again not super important to

know these in depth right now but the

things that are important we will see

these terms again um and you'll know

what they are then so don't worry about

trying to memorize this now but just get

that exposure to NLP terms [Music]

[Music]

okay hey this is Andrew Brown and we're

looking at the concept of a regression

and this is a process of finding a

function to correlate a label data set

into a continuous variable or number so

imagine we need to predict a variable in

the future such as the weather what is

it going to be next week and so the idea

is that you're going to plot your data

onto a graph or vector space our dots

are represented as vectors um and we're

going to draw a line through it which we

call a regression line and the point of

the regression line is that is our

prediction so if this is going over time

based on the temperature um you know uh

that is how we are figuring out in the

future what things are going to be so

the distance of a vector from the

regression line going to just get out a

different colored pen tool other than

than red so maybe cyan so imagine this

dot here to the line that's what we're

going to call an error because the idea

is that um things that are closer to the

line is the prediction and things that

are farther away from the line are an

error from the line so hopefully that

makes sense there are different

regression algorithms used uh uh that

can uh that we use to predict future

variables so we have mean squared error

uh root mean squar error mean absolute

error and So based on the algorithm that

you use to draw your line that's going

to change um the prediction [Music]

[Music]

okay let's take a look at classification

this is the process of finding a

function to divide a label data set into

classes or categories so the idea here

is we're going to predict a category to

apply to the input of data so will it

rain next Saturday is it going to be

sunny or is it going to be raining so

the idea is uh we have our data we're

plotting it on a graph but we're drawing

a classification line that divides the

data set okay and the idea is that if it

falls on one side then it's sunny it

falls on the other side then it's rainy

and so again if you have a different

type of algorithm that's the thing

that's doing the division um it's going

to have different results you have a a

logistic regression a decision tree

random Forest you can use a neural

network you can use a uh a

Navy Bay I always say that wrong so I do

apologize or you can use KNN or you can

use a support Vector machine at or svm

so just understand that there could be

more algorithms of this but these are

the common ones and you know if you want

to learn more about how these different

algorithms will change just look up on

the Internet uh what that would look

like and there's definitely

[Music]

let's talk about clustering this is the

process of grouping unlabeled data based

on similarities and differences the key

word here is unlabeled when we looked at

uh um classification that was labeled

data so the idea here is that we're

grouping based on similar user

differences so imagine that this

grouping of dots that are close together

we determined that that is Windows and

this uh group of dots are Mac computers

and just like classification ression you

have different algorithms they're going

to give you different results and the

reason why I show you these algorithm

names is because when you have to do

classification regression or uh

clustering uh you're going to see these

names because you're going have to

choose what algorithm you want to

utilize right now it's not so important

to uh know them but when they are

important we will look at them uh in

more detail [Music]

[Music]

okay so we are going to dive into the

types of machine learning in other

slides in more detail but this is just

kind of an overview so that you can kind

of see these terms up front um so we'll

just quickly go through this here and

we're going to group them um based on

what they're trying to do so the first

is learning problems we have supervised

unsupervised reinforcement these are

three terms you're going to hear quite a

bit with machine learning uh the key

thing here is that supervised is where

you have labeled data and unsupervised

is where you're working with unlabeled

data for

reinforcement this is an agent an agent

that operates in an En environment and

must learn to operate using feedback and

this kind of sounds like agentic

workflows or agentic coding we're

talking about gen which we'll learn

about later but the idea is like imagine

you wanted to make a uh a machine

learning model that played the the Mario

or or the Sonic video game that'd be

using reinforcement learning okay then

we have hybrid learning problems so we

have semisupervised self- supervised

multi- instance so semisupervised is

where you have a mix of labeled and

unlabeled data you have a lot of

unlabeled data and a little bit of

labeled data and so that's kind of a a

mix between supervised and unsupervised

you have

self-supervised um and I believe that

this is where um the idea is that it can

label its own data I think but we'll

find out later on in future slides we

have multi- instance where we have um

examples of unlabeled data and so then

we just kind of bag them together um

again we'll cover that later on we have

statistical inference so here we have

inductive deductive and and transductive

so using evidence to determine the

outcome or then we have deductive using

general rules to determine the specific

outcomes and then we have transductive

used uh to predict specific examples

given specific uh specific things from a

specific domain okay then for learning

techniques we have multitask active

online transfer and Ensemble so

multitask is fitting a model on one data

set that addresses multiple related

problems active is the model is able to

query a human operator during the

learning process um online is using

available data and updating them mod

before prediction is made kind of sounds

like rag when we're talking about gen um

but again this is just general machine

learning right so we have transfer and

model is first trained on one task and

then sum are all the models used as a

starting point for uh for related task

and then we have uh Ensemble where uh

two or more models are fit on the same

data and the predictions from each model

are combined so yeah we're going to see

these terms again but just trying to get

it uh up front here for you [Music]

[Music]

okay let's take a look at the divisions

of machine learning this is just another

way to breakup machine learning um and

these terms you're going to see uh more

in how we're going to structure our

upcoming slides here so I just want to

give you a quick overview here so we

have classical machine learning and the

advantage of classical machine learning

is the data is simple you have clear

features um and generally classic

machine learning is extremely uh cost

efficient compared to other types of

machine learning but this is where you

have supervised unsupervised uh kind of

uh stuff so you know when you think of

classical machine learning think of

those two things supervised and

unsupervised um uh learning then you

have reinforcement learning this is uh

when there is no data and the idea is

that the model is going to through trial

and error figure out what is the right

thing to do this is where we have

real-time decision- making game AI so we

talked about Mario or sonic uh uh like

the ml model playing those games and

failing again and again and again until

it can pass the game a learning task or

robot navigation so think of automous uh

driving vehicles that would be a good

case for reinforcement learning we have

Ensemble methods when uh quality of data

is a problem so then you are going to

have different strategies to work with

multiple models or algorithms to have a

better outcome and here we have things

like bagging boosting stacking okay and

so you know you'll see those terms like

boosting you'll definitely see the word

boost more uh when we get to that then

we have neural networks and deep

learning you should just really think of

deep learning as neural networks this is

when the data is complicated and or the

features are unclear this is where you'd

use uh neural networks like a

convolutional neural network a

reoccurring neural network uh a gan so

generative adversar [Music]

[Music]

adversarial Network sorry multi-layer

percepton uh or perceptrons sorry MLP

Auto encoders and I just have a really

hard time pronouncing these things but

yeah you're going to see these terms

again so again don't worry about it right

right [Music]

[Music]

now let's take a look here at classical

machine learning and so when we say

classical we're talking about algorithms

that have existed for quite a while may

maybe as early as the 19 50s because we

had these mathematicians and they

figured these out and a lot of these

things actually relate to um statistics

right so we're taking statistics um and

utilizing them uh in these algorithms in

our Computing spaces so hopefully that

makes sense but yeah it's they're called

classical ml because we are dealing with

algorithms and one example would be

nearest neighbor algorithm which was

invented in

1967 and lots of companies today

definitely could utilize classical

machine learning uh to solve business

problems just because they're old does

not mean that they're not good it's just

a matter of organizations knowing how to

adopt uh classical machine learning so

let's talk about first supervised

learning so this is where we have data

that has been labeled into categories

and this is great when we are doing

something that is Task driven we're

trying to make a prediction because the

idea is we have this labeled data and so

then we can bring unlabeled data and

tell the machine to label it right so

here we have classification so we want

an outcome this would be to predict the

C what category something belongs to a

use case here would be identity fraud

detection we have regression this is

where maybe we want to predict a

variable in the future so we're we're

trying to figure out a market forecast

um and we cover you know classical

regression so you should know what these

are um if not you will know about what

they are soon enough because we'll cover

them more than once um then for

unsupervised learning we have data that

has been not been lab laed okay this is

where things are datadriven so we

recognize a structure or a pattern we're

not making a very specific prediction um

here we have clustering so the outcome

of something so you group data based on

similarities or differences example here

would be targeted marketing Association

so find a relationship between variables

through Association the use case here

would be a custo a customer

recommendation we have dimens

dimensionality reduction so here help

reduce the amount of data pre-processing

this is a problem you have a lot of data

um and this a use case here would be big

data visualization so um yeah there you [Music]

[Music]

go all right let's compare supervised

versus unsupervised learning and I know

we've already talked about it like twice

before but we're going to talk about it

again and then again because I'm just

trying to give it to you in different

perspectives so that you really know the

difference between these so let's talk

about what is supervisor learning so

this is a machine learning task or

function that needs to be provided

training data and the training data is

when you provide labeled data the

correct answers and the Machine can

learn from those results so show me how

to do it and then I can do it on my own

that's what's happening here and so for

supervised learning models we have classification

classification

regression what about unsupervised

learning this is a machine learning

Tasker function that needs no existing

training data uh for this it will take

the unlabeled data into discover its

patterns applying its own labels so I am

an independent worker I can figure this

out on my own right uh and for this

these unsupervised learning models we

really should have put the unon that

there let me just fix that there

unsupervised we have clustering

Association Dimension dimensionality

reduction and so supervised learning

tends to be more accurate than

unsupervised learning but requires more

upfront work whereas unsupervised

learning still requires human

intervention to validate the results so

hopefully that is clear [Music]

[Music]

okay okay let's review it one more time

I know it's getting tiresome but it's

very important that you remember the

difference between supervis unsupervised

and reinforcement so supervised learning

is where the data has been labeled for

training it's task driven and you're

making prediction this is when the

labels are known and you want a precise

outcome when you need a specific value

return and so here we use classification

ression as examples of supervised

learning there's more than just those

two but that's what I want you to know

for now we have unsupervised learning

data has not been labeled the ml model

needs to do its own labeling it is Data

driven you're recognizing a structure or

a pattern when the labels are not known

the outcome does not need to be precise

when you're trying to make sense of data

here we have clustering dimensionality

reduction Association then you have

reinforcement learning so there's no

data and there's an environment and an

ml model generates data and many

attempts to reach the goal this is

decision driven you have game AI

learning task robot navigation so

hopefully that is clear and it's in your

head um we are going to repeat these

again but it's going to be less of this

um and more detail [Music]

[Music]

okay let's talk about supervised

learning models and we're going to cover

classification and regression

again um just so that we really know

that we know what these things are so

classification is a process of finding a

function to divide a data set into

classes or categories so imagine will it

be cold or will it be hot tomorrow right

so very clear it's either one or the

other it's going to fall on one side of

the line or the other one we have

different algorithms we can use like log

Logistics regression K nearest neighbor

support Vector machines colel SP spms uh

Navy's uh bay decision stre

classification random Force

classification so we're listing a lot

more here we have what is regression

regression is a process of finding a

function to correlate a data set into a

continuous variable number so what is

the temperature going to be tomorrow and

here we have uh things like s uh simple

linear regression multiple linear

regression polom regression support

Vector regression decision tree

regression random Force regression just

again want to continuously repeat that

so you know what these things are [Music]

[Music]

okay let's take a look at unsupervised

learning uh so what can we do here we

have clustering and again we've covered

these prior but I just really want to

make sure that you know what they are so

clustering is a process of grouping

unlabeled data based on S similarities and

and

differences right so we used an example

previously um you know is this a Mac or

is it a Windows here it's about age and

something else and so it's saying you

know is are these people do these people

have cholesterol are they highrisk or

low risk um for chering algorithms we

have K means uh DB scan K modes then we

have Association so Association is the

process of finding relationship between

variables through Association um so the

idea is that if somebody buys breads

then suggest butter because based on

previous cont combinations we know what

people want um so there are different

algorithms for that I cannot say those

words so I'm not going to attempt it you

can see them here on the right hand side

we have dimensionality reduction this is

where we're reducing the amount of data

we retaining the data Integrity often

used as a pre-processing stage and we

have lots of algorithms for this

principal component analysis linear

discriminant analysis generalized

discrimin analysis singular value

decomposition uh Laden uh direct I can't

say that word there's just too many

words that are too hard to say but

there's a lot there's a lot for

Dimension dimensionality reduction um

yeah and so hopefully you can remember

those things classification regression

clustering Association Dimension

dimensionality reduction [Music]

[Music]

okay let's take a look here at neural

networks and deep learning first

defining what are neural networks so

these are often described as mimicking

the brain you have a neur neuron or node

that represents an algorithm the data is

inputed into the neuron and based on the

output the data will be passed to one of

the many connected neurals the

connections between neurons is weighted

the network is organized into layers

there will be an input layer multiple

hidden layers and an output layer you

could technically have one hidden layer

but often you have multiple layers if

you have three or more now we're talking

about deep learning if you have less

than three then it's just a neural

network um and just look at the visual

for here for a moment because each node

or uh neural remember that it has its

own um its own algorithm like how it's

going to process that data and I'm

pretty certain that most neural networks

the the algorithm is going to be same

for all the nodes but we'll talk about

that as we dig deeper into the neurons

themselves um but then there's the

concept of a feed forward neural network

which is initialized as fnn I don't know

why it's not ffnn but whatever so these

are neural networks where connections

between between nodes do not form a

cycle that means that they always move

forward so data moves forward okay we

don't have neural networks going back

and this way and that way they're just

going One Direction which is forward

then you have back propagation this is

where after um things ran into like

everything's ran through it's going to

move backwards through the neural

network and adjust the

weights okay to improve the outcome on

the next iteration so after it's ran it

actually has to update all the weights

and that is back propagation this is how

a neural network learns it has to do

back propagation okay then we have a

loss function so it's a function that

compares the ground truth to the

prediction to determine the error rate

so how bad the network performed ground

truth right is data that is labeled that

you know to be correct okay now we're

talking about how these neurons are

going to have their own algorithm right

because up here we say that uh it

represents an algorithm so this is where

we have these um algorithms which we

call activation functions so an

activation function is an algorithm

applied to a hidden layer node it's one

of these things right here let me just

get my pen out again one of these that

affects the connected output and so an

example of that would be R L U or reu I

don't know how to pronounce it properly

but I recognize it uh but we will be

looking at activation functions when we

look at Neons a bit uh a bit soon here

um there's the concept of D so when the

network layer increases the amount of

nodes we call it more dense uh and when

the layers decrease the the amount of

nodes we call it sparse okay so when we

see increase it's dense if it's

decreasing it's sparse um

and for deep learning algorithms we have

supervised and unsupervised just like

with classical machine learning um and

so on the supervised side we're going to

see things like uh fnn RNN CNN so you

are passing in labeled data for this to

work for unsupervised learning we we

have uh dbn's SES rbms and not important

to really remember this but I just

wanted you to know that they have

supervised and unsupervised learning um

uh for for deep learning [Music]

[Music]

okay let's take a look at what a

perceptron is so a perceptron is an

algorithm for supervised learning of

binary classifiers invented in

1943 and then the machine was built in

1957 so the mark1 perceptron which is

the name of the machine um it was able

to do some form of image recognition

uh what that would be I don't know I I

wasn't able to extrapolate that but you

can see all of the interconnected uh

work just kind of like the human brain

would have where you have these uh

connections and layers and so this is

kind of where the idea of a um of a

neural network you know came from and

the fact that it's so old just shows you

that we've been doing ml longer than you

think but yeah hopefully that lays the

ground of of the word perceptron but

we'll take a look now at a perceptron [Music]

[Music]

network all right so let's take a look

at a basic perceptron Network and you

might be saying why are we so interested

in this very

old um type of network it's not old this

is neural networks it they are

perceptron networks um so you know just

as goes to show you that the concept is

not new it's just that we have now

scaled it and we have a lot more compute

and we're not connecting everything by

hand right so a basic perceptron has an

input and output layer each layer

contains a number of nodes nodes between

layers have established connections that

are weighted so here is that example the

amount of nodes in the input layer the

input layer right I'm going get my pen

out here over here is determined by the

number of dimensions of the inputed

vector what does that mean the number of

dimensions of an inputed Vector so a

vector remember our our graph we're

taking a DOT and putting it somewhere so

if you had a graph um or a vector space

that had an X and A Y then you have two

inputs for the node right you'd have X

and Y and it doesn't have to be X and Y

it could be different kinds of values

but that's the point there okay so the

input layer is just connection points

okay this input layer nothing that this

layer does will modify the data okay

just the starting point for it so the

amount of nodes in the output layer is

determined by the application of the

neural network so if you have a yes and no

classification uh then you would only

have one output node because you just

want to know is it yes or is it no is it

zero or is it one so it would not matter

if there was a thousand input nodes but

if your classification is yes or no you

only need a single node for that right

the output nodes and other layers can

modif Y and compute new values based on

the inputed

data okay and so data moving between

nodes are uh are multiplied by the

weights right so that is what a weight

does it it affects uh the the strength

or the weakness of the number of what

you want to adjust it for the weights

will be modified during the training

process to produce a better outcome so

hopefully that is clear but the only

thing that's you don't see here is those

hidden layers those additional layers

but anyway we'll move on to now talking

about how the algorithm of the actual uh

neural or the neuron works [Music]

[Music]

okay let's take a look at activation

function so when data arrives to a node

that that can perform a computation all

arriving inputed data is summed and then

an activation function is triggered so

the idea here is you have uh let's say

you have two um uh nodes and you have

connections to the the out the output

node notice that it's summing that is

the mathematical symbol for a sum and

then we have a

mathematical um uh symbol for a function

right so it's going to sum it and then

trigger the activation function so the

activation function acts as a gate

between nodes and determines whether

output will proceed to the next layer

the activ activation function will

determine if a node is active or in

active based on its own output which

could be a range between 0 to 1 to1 to

zero and there's all sorts of activation

functions you can put in here um and

this is not the full list and depending

on if you're watching a a beginner like

because I'm going to have this video in

more than one course so if you're in a

beginner course we will not show you uh

the types of activation functions like

literally how they work but in a more

advanced ml1 we will because you will

want to know them there so just

understand that um you know if you don't

see exactly what these look like it

doesn't matter right now okay so we have

linear activation functions so it can't

do back

propagation um that's what linear

activation functions can't do so here it

just passes along the data then we have

nonl linear activation functions so can

do back propagation can stack and have

many layers here we have binary steps so

if greater than threshold then activate

we have sigmoid used in binary classif

ation susceptible to to the vanishing

gradient problem these are things again

if you are doing real ml with me here

then we will talk about them if you

don't see it in the course it's because

I'm trying to make things easy on you

okay we have tan or ton H I'm not sure

how to pronounce it this is a modified

scill version of sigmoid still

susceptible to the vanishing gr uh

gradient problem which is something we

really want to avoid uh reu again I

don't know how to say it properly uh uh

mostly and we're missing an L there

nobody tell me that okay mostly commonly

used activation function will treat any

negative value as a zero we have leaky

relo this counters the dying REO problem

with a small slope of negative values

parameterz relo so type of leaky relo

where the negative slope is fixed at

0.01x exponential linear unit similar to

reu no dying Ru problem saturates

negative large numbers we have switch

this is is an alternative to uh to the

REO by the Google brain team max out use

it in a max out layer choose the output

uh to be the max of inputs inputs soft

Max this is something you'll see a lot

if you're looking at architectural

diagrams like if you look at the

Transformer architecture look for the

word softmax you'll always see these

near the outputs converts the outputs of

probabilities for the multiple

classifications so yeah you know I might

cover these or we might not uh based on

that course but anyway uh that that is

the activation functions [Music]

[Music]

okay all right so we're taking a look at

activation functions the first being the

linear activation function it is also

known as the identity function it's a

straight line as you can tell here the

model is not really learning it does not

improve upon uh the error term it cannot

perform back propagation it cannot stack

layers only ever has one layer this

means your model will behave if it's

linear so no longer handle complex

nonlinear data uh the range is that it's

Unbound so it's infinite it's derivative

one what you put in is what you get out

um so you know why would you want to use

this I think that it's used for inputs

um because you know if you're just

passing something along then that's

totally fine there but if you had

multiple hidden layers with this it's

not going to be very useful but there you

you [Music]

[Music]

go let's take a look at binary step

activation function so this function

will either either return Z one if the

value is zero or less it will return zero the value is greater than zero

zero the value is greater than zero it'll be uh it'll be one and that's why

it'll be uh it'll be one and that's why it's called a binary step function

it's called a binary step function because it's clearly in one place or or

because it's clearly in one place or or the other it can only handle binary

the other it can only handle binary classification so on or off or true or

classification so on or off or true or false it has a range of zero or one it

false it has a range of zero or one it is bound so it's not infinite it's one

is bound so it's not infinite it's one of the earliest used activation

of the earliest used activation functions not used much today but you

functions not used much today but you know when we were looking at that

know when we were looking at that example of like uh producing a yes or no

example of like uh producing a yes or no you could see that this would be the

you could see that this would be the activation function on the output

activation function on the output function right because that'd be very

function right because that'd be very clear but you can see this is very very

clear but you can see this is very very simplistic

simplistic [Music]

[Music] okay let's take a look at the sigmoid

okay let's take a look at the sigmoid activation function which is a logistic

activation function which is a logistic curve that resembles an S shape so there

curve that resembles an S shape so there it is it can handle binary multic

it is it can handle binary multic classifications so think Cow Horse pig

classifications so think Cow Horse pig as we are looking at multiple types of

as we are looking at multiple types of classific classification we can now

classific classification we can now stack layers

stack layers uh we have ranges between Z and one it

uh we have ranges between Z and one it tends to bring the activations to either

tends to bring the activations to either side of the curve with clear

side of the curve with clear distinctions on prediction one of the

distinctions on prediction one of the most widely used functions near the end

most widely used functions near the end of the function y responds less to X so

of the function y responds less to X so this causes the vanishing gradient what

this causes the vanishing gradient what we're talking about we say Vanishing

we're talking about we say Vanishing gradient like look at this it just goes

gradient like look at this it just goes and it vanishes into the gradient that's

and it vanishes into the gradient that's what it's talking about the network

what it's talking about the network refuses to learn further or is

refuses to learn further or is distractedly slow so if values are over

distractedly slow so if values are over here then you're going to run into

here then you're going to run into some trouble so sigmoid is analog

some trouble so sigmoid is analog meaning almost all neurons will fire be

meaning almost all neurons will fire be active activation will be both dense and

active activation will be both dense and slow slowly and costly so think about

slow slowly and costly so think about that um binary step because if it's

that um binary step because if it's binary step it's either on or off um

binary step it's either on or off um because remember that the the purpose of

because remember that the the purpose of it is that if it's zero it's not going

it is that if it's zero it's not going to pass data along if it's one it is so

to pass data along if it's one it is so because this it I mean it it could

because this it I mean it it could technically be zero but like even if

technically be zero but like even if it's here it's a little bit off on right

it's here it's a little bit off on right it's always on or it's it's like really

it's always on or it's it's like really on or it's teeny tiny on right so um

on or it's teeny tiny on right so um there you

there you [Music]

[Music] go all right I want to admit something

go all right I want to admit something that's really embarrassing but when we

that's really embarrassing but when we initially listed out those activation

initially listed out those activation functions I think I swapped the h&n so I

functions I think I swapped the h&n so I called it tan H when it's just ton and

called it tan H when it's just ton and that's why I was saying ton before

that's why I was saying ton before because I'm like in my mind I knew it

because I'm like in my mind I knew it was Tom but like the H was off so I said

was Tom but like the H was off so I said tan H so I do apologize for that but it

tan H so I do apologize for that but it is ton it is the same as a sigmoid

is ton it is the same as a sigmoid function but it's scaled and it's made

function but it's scaled and it's made larger so it looks really really similar

larger so it looks really really similar so it can handle binary multi

so it can handle binary multi classification because it's analog just

classification because it's analog just like the other one we can stack layers

like the other one we can stack layers we have ranges between1 and one the

we have ranges between1 and one the gradient is stronger so it has a a

gradient is stronger so it has a a steeper curve it still has a vanished

steeper curve it still has a vanished and gradient problem like the sigid um

and gradient problem like the sigid um but versus taon and sigmoid is based on

but versus taon and sigmoid is based on your use case so ton can assist in to

your use case so ton can assist in to avoid bias in gradients ton can

avoid bias in gradients ton can outperform sigmoid so you know it's

outperform sigmoid so you know it's depends if you need to do it or not

depends if you need to do it or not [Music]

[Music] right let's take a look here at relo so

right let's take a look here at relo so relo stands for rectified linear unit

relo stands for rectified linear unit activation function where the positive

activation function where the positive axis is linear and the negative axis is

axis is linear and the negative axis is always zero so it looks like that and

always zero so it looks like that and again just remember the point of

again just remember the point of activation functions is that it's either

activation functions is that it's either on or off or always on to to some degree

on or off or always on to to some degree or not um so here the range is zero to

or not um so here the range is zero to infinite so we have a positive axis that

infinite so we have a positive axis that is

is Unbound um so with sigmoid andon it

Unbound um so with sigmoid andon it fires almost all the neurons and this

fires almost all the neurons and this leads to things being dense remember we

leads to things being dense remember we said dense as in um there's it's adding

said dense as in um there's it's adding more information as it goes as opposed

more information as it goes as opposed to being the same or less it's slow it's

to being the same or less it's slow it's costly um so the uh reu is Will Will

costly um so the uh reu is Will Will sparsely trigger activation functions

sparsely trigger activation functions because of its negative AIS gradient

because of its negative AIS gradient being zero so you have um you know if

being zero so you have um you know if something is really low it's going to be

something is really low it's going to be zero it's not going to um be a teeny

zero it's not going to um be a teeny tiny bit on it's less costly but it's

tiny bit on it's less costly but it's more uh efficient so it's a lot faster

more uh efficient so it's a lot faster the negative axis uh with a zero grading

the negative axis uh with a zero grading has a side effect called the REO dying

has a side effect called the REO dying gradient so the gradient will go towards

gradient so the gradient will go towards zero and will be stuck in zero because

zero and will be stuck in zero because variations adjusting due to input or

variations adjusting due to input or error will have nothing to uh nothing to

error will have nothing to uh nothing to adjust to so the nodes essentially die

adjust to so the nodes essentially die okay

okay [Music]

[Music] let's take a look at leaky REO

let's take a look at leaky REO activation function so leaky rectified

activation function so leaky rectified linear unit activation function is where

linear unit activation function is where the positive axis is linear and the

the positive axis is linear and the negative axis has a gentle gradient

negative axis has a gentle gradient closer to zero do you notice that every

closer to zero do you notice that every time we look at one of these it's trying

time we look at one of these it's trying to solve a problem and and try to be

to solve a problem and and try to be better so hopefully you're seeing that

better so hopefully you're seeing that as we go through these activation

as we go through these activation functions so is similar to the REO but

functions so is similar to the REO but it reduces the effects of the REO d

it reduces the effects of the REO d gradient it's leaky because the negative

gradient it's leaky because the negative axis leaks which causes some nodes not

axis leaks which causes some nodes not to die uh we have also paramed relo

to die uh we have also paramed relo which is leaky uh REO where the negative

which is leaky uh REO where the negative slope is

slope is 0 uh or

0 uh or z01 we have reu 6 uh where we have relu

z01 we have reu 6 uh where we have relu where the positive axis has an upper

where the positive axis has an upper limit so it's not infinite uh so the

limit so it's not infinite uh so the idea here it's bound to a max value okay

idea here it's bound to a max value okay [Music]

[Music] let's take a look here at exponential

let's take a look here at exponential linear unit also known as elu it has a

linear unit also known as elu it has a slope towards a negative one axis it has

slope towards a negative one axis it has a linear gradient in the positive axis

a linear gradient in the positive axis so that's what it looks like kind of

so that's what it looks like kind of like um uh what was the last one I

like um uh what was the last one I pretty forgot it was called but uh you

pretty forgot it was called but uh you know the one where it was uh zero in in

know the one where it was uh zero in in the uh One Direction there but anyway so

the uh One Direction there but anyway so something between yeah reu and and leaky

something between yeah reu and and leaky reu um so elu slope slopes towards the

reu um so elu slope slopes towards the Nega one negative value it pushes the

Nega one negative value it pushes the mean of the activation closer to zero

mean of the activation closer to zero meaning activation closer to zero causes

meaning activation closer to zero causes faster learning and convergence uh elu

faster learning and convergence uh elu avoids the dying uh elu problem it

avoids the dying uh elu problem it saturates for larger negative numbers so

saturates for larger negative numbers so everything is a trade-off with these

everything is a trade-off with these things

things [Music]

[Music] okay let's take a look at the swish

okay let's take a look at the swish activation function so it has a slope

activation function so it has a slope that dips and eases out to zero in the

that dips and eases out to zero in the negative axis it has a linear gradient

negative axis it has a linear gradient in the positive axis so kind of looks

in the positive axis so kind of looks similar but like a little bit different

similar but like a little bit different swish was proposed by the Google brain

swish was proposed by the Google brain team as a replacement for REO it's

team as a replacement for REO it's called swish because of its switching

called swish because of its switching dip it looks similar to relu but it's a

dip it looks similar to relu but it's a smooth function it never abruptly

smooth function it never abruptly changes Direction it it is non monotonic

changes Direction it it is non monotonic so it does not remain stable similar to

so it does not remain stable similar to Ru will have sparity very negative uh

Ru will have sparity very negative uh very negative values will Zero out there

very negative values will Zero out there are other variants in the swish family

are other variants in the swish family so we have Mish hard Swish and hard

so we have Mish hard Swish and hard [Music]

[Music] let's take a look at max out so this is

let's take a look at max out so this is a function that uh that will take

a function that uh that will take multiple inputs and it will select the

multiple inputs and it will select the maximum value and return the value so um

maximum value and return the value so um the max out is a generalization of relu

the max out is a generalization of relu and the Leaky reu functions max out

and the Leaky reu functions max out neuron would have all the benefits of

neuron would have all the benefits of relu neurons without having the dying

relu neurons without having the dying reu max out is uh is that it's expensive

reu max out is uh is that it's expensive as it doubles the number of parameters

as it doubles the number of parameters for each

for each [Music]

[Music] neuron all here's our last one the soft

neuron all here's our last one the soft Max activation function this is uh it

Max activation function this is uh it will calculate the probabilities of each

will calculate the probabilities of each class over all possible classes when

class over all possible classes when used for multi classification models it

used for multi classification models it Returns the probabilities of each class

Returns the probabilities of each class and the target class will have the high

and the target class will have the high probability uh the calculated properties

probability uh the calculated properties Pro probabilities will be in the range

Pro probabilities will be in the range of zero and one the sum of all

of zero and one the sum of all probabilities is equal to one softmac

probabilities is equal to one softmac functions is generally used in multiple

functions is generally used in multiple classifications on the output layer so

classifications on the output layer so again I said if you look at the

again I said if you look at the Transformer architecture which probably

Transformer architecture which probably is in this course you will see it there

is in this course you will see it there and you'll see it in other ml models uh

and you'll see it in other ml models uh diagrams for sure you can only assign a

diagrams for sure you can only assign a single label to a probability for this

single label to a probability for this [Music]

[Music] okay let's define a algorithm and a

okay let's define a algorithm and a function so an algorithm is a set of

function so an algorithm is a set of mathematical or computer instructions to

mathematical or computer instructions to perform a specific task and an algorithm

perform a specific task and an algorithm can be composed of several smaller

can be composed of several smaller algorithms you're basically saying how

algorithms you're basically saying how do you do something that's what an

do you do something that's what an algorithm is right how are we going to

algorithm is right how are we going to do something um so I want to take a look

do something um so I want to take a look here at the K nearest neighbor knnn

here at the K nearest neighbor knnn algorithm which can be used to create a

algorithm which can be used to create a supervised classification machine uh

supervised classification machine uh learning algorithm so tell me who are

learning algorithm so tell me who are your closest neighbors and we will infer

your closest neighbors and we will infer that that I can be considered of the

that that I can be considered of the same class so within KNN you can use

same class so within KNN you can use different distance metrics uh such as uh

different distance metrics uh such as uh idian Hamming uh manowski Manhattan so

idian Hamming uh manowski Manhattan so there's all different ones that you can

there's all different ones that you can utilize a function is a way of grouping

utilize a function is a way of grouping algorithms together uh so you can call

algorithms together uh so you can call them to compute a result so sounds like

them to compute a result so sounds like a machine learning model model right

a machine learning model model right where you have a grouping of algorithms

where you have a grouping of algorithms so you know look at this K and N just

so you know look at this K and N just here for a moment because we do uh see

here for a moment because we do uh see this happen a lot but K nearest neighbor

this happen a lot but K nearest neighbor is just like how close am I from here to

is just like how close am I from here to here to here to here it's literally in

here to here to here it's literally in the name how who are my nearest

the name how who are my nearest neighbors okay so KNN itself is not

neighbors okay so KNN itself is not machine learning but when applied to

machine learning but when applied to solve machine learning problem it makes

solve machine learning problem it makes it a machine learning algorithm okay

it a machine learning algorithm okay [Music]

[Music] let's take a look at what a machine

let's take a look at what a machine learning model is but before we do that

learning model is but before we do that let's define what a model is in general

let's define what a model is in general terms so in general terms a model is

terms so in general terms a model is information representation of an object

information representation of an object person or system models can be concrete

person or system models can be concrete so they have a physical form think a

so they have a physical form think a design of a vehicle a person posing for

design of a vehicle a person posing for a picture then you have abstract so

a picture then you have abstract so Express as behavioral patterns think

Express as behavioral patterns think mathematical computer code written words

mathematical computer code written words so what is a machine learning model then

so what is a machine learning model then an ml model is a function that takes uh

an ml model is a function that takes uh in data performs a machine learning

in data performs a machine learning algorithm to produce a prediction the

algorithm to produce a prediction the machine learning model is trained not to

machine learning model is trained not to be confused with the training model

be confused with the training model which is learning to make correct

which is learning to make correct predictions uh an ml model can be the

predictions uh an ml model can be the training model that is just deployed

training model that is just deployed once it has been tuned to make good

once it has been tuned to make good predictions so normally you'd have

predictions so normally you'd have training data let's say labeled data and

training data let's say labeled data and here you are going to have your learning

here you are going to have your learning algorithm and you're going to put it

algorithm and you're going to put it through training so that's your training

through training so that's your training model and then you have hyper tuning

model and then you have hyper tuning where you are continuously tweaking the

where you are continuously tweaking the model to get it to where you want it to

model to get it to where you want it to be okay then once you deploy the model

be okay then once you deploy the model that is your trained model your machine

that is your trained model your machine learning model which can go and produce

learning model which can go and produce predictions um and from here you could

predictions um and from here you could then provide it unlabeled data because

then provide it unlabeled data because you know its goal is to make predictions

you know its goal is to make predictions and that could be labeling data or doing

and that could be labeling data or doing other things okay and we call uh uh uh

other things okay and we call uh uh uh the interaction with the deployed

the interaction with the deployed machine learning model inference right

machine learning model inference right so when you are inferring something you

so when you are inferring something you are providing you're providing data and

are providing you're providing data and saying hey can you uh make a prediction

saying hey can you uh make a prediction for me and that's what inference is

for me and that's what inference is [Music]

[Music] okay so let's take a look at what a

okay so let's take a look at what a feature is so a feature is a

feature is so a feature is a characteristic extracted from our

characteristic extracted from our unstructured data set that has been

unstructured data set that has been prepared to be ingested by our machine

prepared to be ingested by our machine learning model to infer a prediction so

learning model to infer a prediction so ml models generally only accept

ml models generally only accept numerical data and so we prepare our

numerical data and so we prepare our data into machine readable format by

data into machine readable format by encoding which we'll revisit later in

encoding which we'll revisit later in more detail um so let's talk about what

more detail um so let's talk about what is feature engineering so feature

is feature engineering so feature engineering is the process of extracting

engineering is the process of extracting features from our provided data sources

features from our provided data sources so imagine you have your data sources

so imagine you have your data sources which you have then your raw data you're

which you have then your raw data you're going to clean and transform them into

going to clean and transform them into features turning them into machine

features turning them into machine readable format information for your

readable format information for your machine learning models and then you

machine learning models and then you know you go from there

know you go from there [Music]

[Music] okay so what is inference inference is

okay so what is inference inference is the act of requesting and getting a

the act of requesting and getting a prediction and when we're talking about

prediction and when we're talking about in the context of machine learning we're

in the context of machine learning we're inputting data into a machine learning

inputting data into a machine learning model that has been deployed for

model that has been deployed for production use to then output a

production use to then output a prediction so imagine our raw data is a

prediction so imagine our raw data is a banana and we tell we say tell me what

banana and we tell we say tell me what this is to the machine learning model

this is to the machine learning model it's going to bring back information so

it's going to bring back information so saying it's a yellow banana and it has a

saying it's a yellow banana and it has a confidence score of 0.9 so if we talk

confidence score of 0.9 so if we talk about the inference textbook definition

about the inference textbook definition it's steps in reasoning or moving from

it's steps in reasoning or moving from premise to logical consequence but I I

premise to logical consequence but I I think that it's easy to remember as the

think that it's easy to remember as the act of requesting and getting a

act of requesting and getting a prediction

prediction [Music]

[Music] okay let's talk about parameters and

okay let's talk about parameters and hyperparameters so a model parameter is

hyperparameters so a model parameter is a variable that configures the internal

a variable that configures the internal state of a model and whose value can be

state of a model and whose value can be estimated the value of parameter is not

estimated the value of parameter is not manually set and will be learned

manually set and will be learned outputed after training parameters are

outputed after training parameters are used to make predictions then we have

used to make predictions then we have model hyp or model hyperparameter this

model hyp or model hyperparameter this is a variable that is external to the

is a variable that is external to the model and whose value cannot be

model and whose value cannot be estimated the value of the

estimated the value of the hyperparameter is manually set before

hyperparameter is manually set before the training of the model hyper

the training of the model hyper parameters are used to estimate model

parameters are used to estimate model parameters and so we have things like

parameters and so we have things like learning rate Epoch and batch size and

learning rate Epoch and batch size and here's kind of a diagram hopefully it

here's kind of a diagram hopefully it helps make sense but imagine you have a

helps make sense but imagine you have a variable and you want to input it into

variable and you want to input it into your model right and we'll just make a

your model right and we'll just make a box here to indicate that this is the

box here to indicate that this is the model it's going to go into layers right

model it's going to go into layers right and we'll talk about this again later on

and we'll talk about this again later on but uh par

but uh par are the connections between uh nodes

are the connections between uh nodes okay so the idea is that this will have

okay so the idea is that this will have a variable or a value and it'll have a

a variable or a value and it'll have a weight and those are those internal

weight and those are those internal State those parameters okay so hopefully

State those parameters okay so hopefully uh that is very clear there because the

uh that is very clear there because the idea is that when you want to uh utilize

idea is that when you want to uh utilize something for training right you're

something for training right you're going to pass um very like a a Content

going to pass um very like a a Content or variables it's going to go through

or variables it's going to go through all those layers and then all these

all those layers and then all these connections have to be set these

connections have to be set these parameters

parameters of these connections have to be set so

of these connections have to be set so you get the result that you want to get

you get the result that you want to get so hopefully that is clear but we will

so hopefully that is clear but we will cover it again um if it's not clear

cover it again um if it's not clear later on

later on [Music]

[Music] okay hey this is Andrew Brown let's take

okay hey this is Andrew Brown let's take a look at responsible AI specifically

a look at responsible AI specifically for ad of us and often you'll see like a

for ad of us and often you'll see like a list of things like fairness

list of things like fairness explainability privacy and security

explainability privacy and security safety

safety controllability veracity robustness

controllability veracity robustness governance and transparent so this is

governance and transparent so this is the one that adab us defines other ones

the one that adab us defines other ones like Microsoft and other people have

like Microsoft and other people have similar lists so they're more or less

similar lists so they're more or less the same but for the exams for the AI

the same but for the exams for the AI practitioner they might give you a list

practitioner they might give you a list of these so you might want to remember

of these so you might want to remember those key terms let's go ahead and see

those key terms let's go ahead and see what we have in terms of resources for

what we have in terms of resources for responsible AI here so we have model

responsible AI here so we have model evaluation on Amazon Bedrock we have

evaluation on Amazon Bedrock we have Amazon sagemaker clarify we do look at

Amazon sagemaker clarify we do look at that later that's for explainable AI to

that later that's for explainable AI to determine what's going on there and

determine what's going on there and again we have guard rails we have a on

again we have guard rails we have a on that so we look at that we have clarify

that so we look at that we have clarify again clarify again model monitor which

again clarify again model monitor which is more about monitoring the degration

is more about monitoring the degration of a model we do talk about that Amazon

of a model we do talk about that Amazon augmented AI that is a human reviewing

augmented AI that is a human reviewing the end points so all these things are

the end points so all these things are covered um yeah it doesn't look like

covered um yeah it doesn't look like they have a whole lot here let's see ACI

they have a whole lot here let's see ACI service cards provides transfering

service cards provides transfering document intended use cases for fairness

document intended use cases for fairness so I know Microsoft has something very

so I know Microsoft has something very similar um but uh yeah I guess they're

similar um but uh yeah I guess they're just down below

just down below here not super exciting to be

here not super exciting to be honest yeah you got a bunch of stuff you

honest yeah you got a bunch of stuff you can read through so you can see how

can read through so you can see how they're being responsible with it I

guess and yeah so nothing super super exciting here but um yeah I guess

exciting here but um yeah I guess clarify is their big thing here

clarify is their big thing here remembering this list

remembering this list [Music]

[Music] okay let's take a look at labeling so

okay let's take a look at labeling so data label is the process of identifying

data label is the process of identifying raw data images text files videos and

raw data images text files videos and adding one or me more meaningful and

adding one or me more meaningful and informative labels to provide context so

informative labels to provide context so machine learning model can learn from

machine learning model can learn from with supervised Lear uh machine learning

with supervised Lear uh machine learning labeling is a prerequisite to produce

labeling is a prerequisite to produce training data and each piece of data

training data and each piece of data will generally be labeled by human on

will generally be labeled by human on left- hand side that's an example of um

left- hand side that's an example of um Amazon recognition where it's trying to

Amazon recognition where it's trying to identify bounding boxes or classifying

identify bounding boxes or classifying image under particular categories that's

image under particular categories that's an example of supervised machine

an example of supervised machine learning that requires labeled data with

learning that requires labeled data with unsupervised machine learning labels

unsupervised machine learning labels will be uh produced by the machine and

will be uh produced by the machine and may not be human readable then there's

may not be human readable then there's this concept of ground truth this is a

this concept of ground truth this is a uh a properly labeled data set that you

uh a properly labeled data set that you use as an objective standard to train

use as an objective standard to train and assess a given model and is often

and assess a given model and is often called Ground truth the accuracy of

called Ground truth the accuracy of train models will depend on the accuracy

train models will depend on the accuracy of your ground truth and so ground truth

of your ground truth and so ground truth data is very important uh for uh you

data is very important uh for uh you know successful models okay

[Music] let's take a look here at data mining

let's take a look here at data mining this is the extraction of patterns and

this is the extraction of patterns and knowledge from large amounts of data not

knowledge from large amounts of data not the extraction of data itself and so the

the extraction of data itself and so the industry has this thing called Chris DM

industry has this thing called Chris DM which defines it in six phases first is

which defines it in six phases first is business understanding so what does the

business understanding so what does the business need data understanding what do

business need data understanding what do we have and what data do we have we have

we have and what data do we have we have data preparation so how do we organize

data preparation so how do we organize the data for modeling the modeling which

the data for modeling the modeling which is what modeling Tech techniques should

is what modeling Tech techniques should we apply

we apply evaluation what data model best meets

evaluation what data model best meets the business objectives deployment how

the business objectives deployment how do people access the data so that gives

do people access the data so that gives you an idea about working with data

you an idea about working with data mining

mining [Music]

[Music] okay let's take a look here at data

okay let's take a look here at data mining methods um these are ways that we

mining methods um these are ways that we find valid patterns in relationships in

find valid patterns in relationships in huge data sets and they're important

huge data sets and they're important when we're talking about machine

when we're talking about machine learning because sometimes that is what

learning because sometimes that is what the model is trying to do it's trying to

the model is trying to do it's trying to find a pattern of relationship it's

find a pattern of relationship it's trying to predict ICT that so I'm not

trying to predict ICT that so I'm not going to read through all of this

going to read through all of this because you can read through it if you

because you can read through it if you want but these are terms that we've seen

want but these are terms that we've seen already like classification clustering

already like classification clustering regression sequential Association rules

regression sequential Association rules outer detection and prediction uh and

outer detection and prediction uh and notice down here when we have prediction

notice down here when we have prediction it says uh use a combination of other

it says uh use a combination of other data mining techniques such as transends

data mining techniques such as transends clustering classification to predict

clustering classification to predict future data which is fine but we have

future data which is fine but we have classification clustering regression and

classification clustering regression and Association these four are going to show

Association these four are going to show up again and again when we're looking at

up again and again when we're looking at um classical models okay so machine

um classical models okay so machine learning models but anyway I just wanted

learning models but anyway I just wanted to include that even though this is more

to include that even though this is more of a data a data slide

of a data a data slide [Music]

[Music] okay let's take a look here at knowledge

okay let's take a look here at knowledge mining this is a discipline in AI that

mining this is a discipline in AI that uses combination of intelligent services

uses combination of intelligent services to quickly learn from vast amounts of

to quickly learn from vast amounts of information it allows organizations to

information it allows organizations to deeply understand and easily explore

deeply understand and easily explore information uncover hidden insights and

information uncover hidden insights and find relationships and patterns at scale

find relationships and patterns at scale this is a term that was kind of coin

this is a term that was kind of coin over at Microsoft you don't hear about

over at Microsoft you don't hear about it over at Azure or gcp but it still is

it over at Azure or gcp but it still is a good concept to know the other thing

a good concept to know the other thing is that when we look at rag so that's

is that when we look at rag so that's retrieval augmented generation there is

retrieval augmented generation there is a lot of overlap with this or in many

a lot of overlap with this or in many cases you can look at rag being

cases you can look at rag being knowledge mining um but let's talk about

knowledge mining um but let's talk about what we have here so the first thing is

what we have here so the first thing is ingest then we have enrich and we have

ingest then we have enrich and we have explore so inest is ingest content from

explore so inest is ingest content from a range of sources using connectors to

a range of sources using connectors to fir uh uh to first and third party data

fir uh uh to first and third party data stores so we have structured data like

stores so we have structured data like databases csvs unstructured data like

databases csvs unstructured data like PDF video images and audio we have

PDF video images and audio we have enrich so enrich the content with AI

enrich so enrich the content with AI capabilities and let you extract

capabilities and let you extract information find patterns and deep

information find patterns and deep deepening understanding so for manage AI

deepening understanding so for manage AI Services we have Vision Services

Services we have Vision Services language Services speech services

language Services speech services decision services and search Services

decision services and search Services now those literally map to Azure uh AI

now those literally map to Azure uh AI managed services but we're talking about

managed services but we're talking about AWS uh when we're talking about Vision

AWS uh when we're talking about Vision we're talking about recognition we're

we're talking about recognition we're talking about language um I guess that

talking about language um I guess that could be something like um I'm trying to

could be something like um I'm trying to remember the service that does NLP here

remember the service that does NLP here uh okay remember off the top of my head

uh okay remember off the top of my head but for speech we have poly um for for

but for speech we have poly um for for search this could be um not necessarily

search this could be um not necessarily an AI well it could be Kendra right so

an AI well it could be Kendra right so there's a lot of manag AI services that

there's a lot of manag AI services that can be utilized at that level then we

can be utilized at that level then we have Explorer so the newly indexed data

have Explorer so the newly indexed data via search Bots or existing business

via search Bots or existing business applications and data visualizations so

applications and data visualizations so here it could be used in a CRM it could

here it could be used in a CRM it could be in a wrap system it could be powerbi

be in a wrap system it could be powerbi and I didn't list it here but it could

and I didn't list it here but it could also be used to return back to an llm to

also be used to return back to an llm to interpret and then complete rag so there

interpret and then complete rag so there you

you [Music]

[Music] go let's take a look here at data

go let's take a look here at data wrangling this is the process of

wrangling this is the process of transforming mapping data from one raw

transforming mapping data from one raw data form into another format with the

data form into another format with the intent of making it more appropriate and

intent of making it more appropriate and valuable uh for a variety of Downstream

valuable uh for a variety of Downstream purposes such as analytics also known as

YouTube TranscriptPreparing your results…

YouTube Transcript:AWS Certified AI Practitioner (AIF-C01) – Full Course to PASS the Certification Exam

AutoDub

Video Transcript

Summary Requirements

YouTube Transcript:
AWS Certified AI Practitioner (AIF-C01) – Full Course to PASS the Certification Exam