WEBVTT

00:00.330 --> 00:01.830
-: Hello and welcome to the section

00:01.830 --> 00:04.140
on Convolution Neural Networks.

00:04.140 --> 00:06.120
Super excited that you are joining us

00:06.120 --> 00:07.710
for this section as well.

00:07.710 --> 00:11.070
And today we're going to cover off the plan of attack,

00:11.070 --> 00:13.650
how we're going to learn everything in this section,

00:13.650 --> 00:15.343
there's so much to learn.

00:15.343 --> 00:17.460
Let's see how we're going to approach this.

00:17.460 --> 00:19.590
All right, what we'll learn in this section.

00:19.590 --> 00:20.423
First of all,

00:20.423 --> 00:23.460
we'll talk about what convolution networks actually are.

00:23.460 --> 00:25.563
Very important to understand the end goal

00:25.563 --> 00:27.150
that you're working towards

00:27.150 --> 00:28.800
before you actually start working towards it.

00:28.800 --> 00:30.360
So we'll talk about features,

00:30.360 --> 00:32.100
we'll have a look at a few little examples,

00:32.100 --> 00:33.300
we'll compare the human brain

00:33.300 --> 00:36.810
to artificial neural networks in terms of image recognition.

00:36.810 --> 00:40.680
So it'll be a fun, a light tutorial to get us started

00:40.680 --> 00:42.600
for this whole section.

00:42.600 --> 00:44.850
Then we'll talk about step one,

00:44.850 --> 00:47.400
diving straight into it, convolution operation.

00:47.400 --> 00:51.390
So this part of the course contains several steps

00:51.390 --> 00:52.410
that we need to go through

00:52.410 --> 00:57.030
in order to build a convolution neural network

00:57.030 --> 00:59.070
and that's how these tutorials are going to be broken up.

00:59.070 --> 01:00.540
So this one's gonna be step one,

01:00.540 --> 01:01.920
the convolution operation,

01:01.920 --> 01:06.480
we'll learn everything about feature detectors,

01:06.480 --> 01:08.820
we'll talk about, which are also filters,

01:08.820 --> 01:11.040
we'll talk about feature maps

01:11.040 --> 01:14.610
and what are the different parameters there, what they mean,

01:14.610 --> 01:17.490
and we'll have a look at some visual examples as well.

01:17.490 --> 01:21.540
Then we'll talk about step one part B, their RELU layer,

01:21.540 --> 01:26.540
or RELU layer, which is the rectified linear unit.

01:26.670 --> 01:31.050
And we'll talk about why linearity is not good

01:31.050 --> 01:33.210
and how we want more non-linearity

01:33.210 --> 01:36.840
in our network for image recognition.

01:36.840 --> 01:38.790
Then we'll talk about step two, pooling,

01:38.790 --> 01:41.370
and we'll understand how pooling works.

01:41.370 --> 01:43.200
We'll talk specifically about max pooling

01:43.200 --> 01:46.410
and we'll also mention a couple of things about mean pooling

01:46.410 --> 01:49.680
or some pooling and other approaches that you can take

01:49.680 --> 01:51.960
to the process of pooling.

01:51.960 --> 01:55.530
Also in this lecture, we'll have a really cool example.

01:55.530 --> 01:58.650
So there'll be a very visual, interactive tool

01:58.650 --> 01:59.730
that we're going to look at.

01:59.730 --> 02:02.340
So make sure to stick around to the end of that lecture

02:02.340 --> 02:05.010
because that's going to add a lot of value

02:05.010 --> 02:06.780
to your learning process.

02:06.780 --> 02:09.150
What we're going to discuss at the end there.

02:09.150 --> 02:10.680
Step three, flattening.

02:10.680 --> 02:11.880
So here we will,

02:11.880 --> 02:13.080
It's gonna be a quick tutorial

02:13.080 --> 02:15.630
on how to proceed from your pooled layers

02:15.630 --> 02:17.250
to your flattened layer,

02:17.250 --> 02:19.500
and then we're going to talk about full connection.

02:19.500 --> 02:23.430
So this is the very meaty tutorial

02:23.430 --> 02:24.570
that puts everything together

02:24.570 --> 02:26.040
and puts everything into perspective

02:26.040 --> 02:29.670
and actually shows you how everything works

02:29.670 --> 02:30.503
at the end of the day

02:30.503 --> 02:33.360
and how those final neurons understand

02:33.360 --> 02:34.290
how to classify image.

02:34.290 --> 02:35.826
Very, very important tutorial

02:35.826 --> 02:39.600
and hopefully that will summarize

02:39.600 --> 02:42.480
or kind of put everything together for you.

02:42.480 --> 02:43.920
And finally, we'll have a summary,

02:43.920 --> 02:46.530
which will summarize everything we've talked about.

02:46.530 --> 02:48.480
And as an extra little feature,

02:48.480 --> 02:52.230
I've included in a tutorial on softmax and cross entropy.

02:52.230 --> 02:54.090
So you don't have to take this tutorial,

02:54.090 --> 02:57.390
but I thought it'd be a great addition of knowledge

02:57.390 --> 02:59.370
because these are terms that you will come across

02:59.370 --> 03:02.040
when dealing with convolution neural networks.

03:02.040 --> 03:04.560
So maybe take it right away.

03:04.560 --> 03:07.200
Maybe when you come across these terms,

03:07.200 --> 03:09.900
you will always know, you can come back to this course

03:09.900 --> 03:12.690
and take this tutorial to understand better

03:12.690 --> 03:14.730
what softmax and cross entropy are.

03:14.730 --> 03:18.570
And also as always, throughout these tutorials,

03:18.570 --> 03:20.010
there will be lots of recommended reading

03:20.010 --> 03:23.340
for you to further upscale and get more knowledge.

03:23.340 --> 03:24.210
And on that note,

03:24.210 --> 03:26.040
I can't wait to see you on the first tutorial.

03:26.040 --> 03:29.040
This is gonna be a very fun and exciting section.

03:29.040 --> 03:31.443
And until next time, enjoy deep learning.