1
00:00:00,030 --> 00:00:06,299
In the previous video, we learned about
convolutional neural networks, which are

2
00:00:06,299 --> 00:00:11,160
supervised deep learning models that
have revolutionized the field of

3
00:00:11,160 --> 00:00:16,680
computer vision, especially object
detection in images. In this video, we

4
00:00:16,680 --> 00:00:20,670
will learn about another supervised deep
learning model, which is the recurrent

5
00:00:20,670 --> 00:00:28,590
neural network. So far, we have seen
neural networks and deep learning models

6
00:00:28,590 --> 00:00:34,649
that see datapoints as independent
instances. However, let's say you want to

7
00:00:34,649 --> 00:00:38,070
build a model that can analyze scenes in
a movie.

8
00:00:38,070 --> 00:00:43,309
Well, you cannot assume that scenes in a
movie are independent, and therefore,

9
00:00:43,309 --> 00:00:49,140
traditional deep learning models are not
suitable for this application. Recurrent

10
00:00:49,140 --> 00:00:54,000
neural networks overcome this issue. Recurrent neural networks or (RNNs)

11
00:00:54,000 --> 00:00:59,340
for short, are networks with loops that
don't just take a new input at a time,

12
00:00:59,340 --> 00:01:04,619
but also take in as input the output from
the previous dat point that was fed

13
00:01:04,619 --> 00:01:09,659
into the network. Accordingly, this is how
the architecture of a recurrent neural

14
00:01:09,659 --> 00:01:14,280
network would look like. Essentially, we
can start with a normal neural network.

15
00:01:14,280 --> 00:01:24,520
At time t = 0, the network takes in
input x0 and outputs a0. Then, at time t = 1,

16
00:01:24,520 --> 00:01:30,509
in addition to the input  x1,
the network also takes a0 as input,

17
00:01:30,509 --> 00:01:38,250
weighted with weight w0,1, and so on and so forth. As a result, recurrent neural

18
00:01:38,250 --> 00:01:42,479
networks are very good at modelling
patterns and sequences of data, such as

19
00:01:42,479 --> 00:01:49,229
texts, genomes, handwriting, and stock
markets. These algorithms take time and

20
00:01:49,229 --> 00:01:54,630
sequence into account, which means that
they have a temporal dimension. A very

21
00:01:54,630 --> 00:01:59,969
popular type of recurrent neural network
is the long short-term memory model or

22
00:01:59,969 --> 00:02:06,259
the (LSTM) model for short. It has been
successfully used for many applications

23
00:02:06,259 --> 00:02:11,110
including image generation, where a model trained on many

24
00:02:11,110 --> 00:02:16,470
images is used to generate new novel
images. Another application is

25
00:02:16,470 --> 00:02:22,170
handwriting generation, which I described
in the welcome video of this course. Also

26
00:02:22,170 --> 00:02:27,670
LSTM models have been successfully
used to build algorithms that can

27
00:02:27,670 --> 00:02:35,290
automatically describe images as well as
streams of videos. I think this is a good

28
00:02:35,290 --> 00:02:39,760
overview of recurrent neural networks.
Given that this is just an introductory

29
00:02:39,760 --> 00:02:45,380
course I will leave it here. This
concludes this video

30
00:02:45,380 --> 00:02:47,260
on recurrent neural networks. I will see
you in the next video

31
00:02:47,260 --> 00:02:51,360
we will switch to unsupervised deep
learning models and talk about

32
00:02:51,360 --> 00:02:54,360
autoencoders.