1
00:00:11,060 --> 00:00:14,480
So in this lecture, we'll be summarizing what we learned in this section.

2
00:00:15,170 --> 00:00:18,590
This section introduced feedforward artificial neural networks.

3
00:00:19,190 --> 00:00:24,080
Although there was quite a bit of theory involved, the code was quite simple compared to what you already

4
00:00:24,080 --> 00:00:24,680
learned.

5
00:00:25,400 --> 00:00:28,310
Essentially, all you had to do was add another dense layer.

6
00:00:29,030 --> 00:00:34,620
However, this was only for the easy case where inputs were still represented using TFI Taf.

7
00:00:35,420 --> 00:00:40,820
Shortly thereafter, we learned how to prepare this text with Keros, which automatically does a lot

8
00:00:40,820 --> 00:00:43,100
of the same work we previously discussed.

9
00:00:43,850 --> 00:00:49,430
Importantly, we also learned about embeddings, which allow us to create a more granular representation

10
00:00:49,430 --> 00:00:54,410
of text in particular, instead of a whole document becoming a single vector.

11
00:00:54,740 --> 00:00:57,170
We map each token to a vector individually.

12
00:00:58,070 --> 00:01:02,780
Now for this section, we don't yet know what to do with all of these vectors, but we'll see how they

13
00:01:02,780 --> 00:01:05,060
can be handled with CNN's internet.

14
00:01:05,660 --> 00:01:11,570
What we did learn about was how embeddings could be used in an ends by taking the average of each individual

15
00:01:11,570 --> 00:01:12,320
word vector.

16
00:01:12,890 --> 00:01:16,460
In fact, this is what we use in the C bao variant of words of EC.

17
00:01:17,120 --> 00:01:22,190
In this section, we got to implement this version of words of EC, and we saw that it produces results

18
00:01:22,580 --> 00:01:26,930
similar to what we saw in our neural word embeddings demo earlier in the course.