1
00:00:00,000 --> 00:00:01,000
Hello guys!

2
00:00:01,000 --> 00:00:07,000
So in our previous video we have already discussed about the architecture of LSTM, RNN and along with

3
00:00:07,000 --> 00:00:10,000
this we saw three important gates.

4
00:00:10,000 --> 00:00:12,000
One is for gate gate.

5
00:00:13,000 --> 00:00:16,000
Then we saw about input gate.

6
00:00:16,000 --> 00:00:20,000
And this was with respect to with along with the candidate memory.

7
00:00:20,000 --> 00:00:27,000
And finally we saw about output gate and we understood what forget gate was doing, input gate was doing

8
00:00:27,000 --> 00:00:28,000
and output gate was doing.

9
00:00:28,000 --> 00:00:28,000
Right right.

10
00:00:29,000 --> 00:00:37,000
Now let's take one specific example and let's see that how this entire data is basically trained in

11
00:00:37,000 --> 00:00:39,000
an LSTM RNN.

12
00:00:39,000 --> 00:00:39,000
Okay.

13
00:00:39,000 --> 00:00:42,000
So here uh, obviously I have one example.

14
00:00:42,000 --> 00:00:45,000
So this is my text paragraph and this is my output okay.

15
00:00:45,000 --> 00:00:54,000
Let's say that based on this information I need to find out whether this particular text based on this

16
00:00:54,000 --> 00:00:58,000
particular text, the food is in the restaurant is good or bad.

17
00:00:58,000 --> 00:00:59,000
Okay.

18
00:00:59,000 --> 00:01:05,000
So I need to predict whether the food is good or bad in the restaurant.

19
00:01:05,000 --> 00:01:06,000
Okay.

20
00:01:06,000 --> 00:01:10,000
And these are some of the reviews that have come with for this particular restaurant.

21
00:01:10,000 --> 00:01:14,000
So here let's read this text paragraph.

22
00:01:14,000 --> 00:01:19,000
It says, hey, I went to the restaurant and ordered burger.

23
00:01:19,000 --> 00:01:21,000
Okay, so in this paragraph, this is my first sentence.

24
00:01:22,000 --> 00:01:27,000
Then here you can see the burger looked tasty and crispy.

25
00:01:28,000 --> 00:01:28,000
Okay.

26
00:01:29,000 --> 00:01:31,000
But burger is not good for health.

27
00:01:32,000 --> 00:01:34,000
It has a lot of fats cholesterol.

28
00:01:35,000 --> 00:01:41,000
But this burger was made with whey protein.

29
00:01:42,000 --> 00:01:46,000
and only vegetables were used, so it was good.

30
00:01:46,000 --> 00:01:49,000
Now see all this information here.

31
00:01:49,000 --> 00:01:52,000
It shows that it has lot of fats and cholesterol.

32
00:01:52,000 --> 00:01:57,000
And here it is saying hey, it is made up of whey protein and only vegetables.

33
00:01:57,000 --> 00:01:58,000
So it was good.

34
00:01:58,000 --> 00:02:00,000
And finally my output is basically one.

35
00:02:00,000 --> 00:02:04,000
So this is that basically means this burger is good right.

36
00:02:04,000 --> 00:02:06,000
So this is my training data.

37
00:02:06,000 --> 00:02:12,000
So you can probably consider this as my training data okay.

38
00:02:13,000 --> 00:02:19,000
Now with respect to training data, how do we take this particular sentence and train it in our memory.

39
00:02:19,000 --> 00:02:25,000
What information needs to probably go to my memory cell or the long term memory cell and the short term

40
00:02:25,000 --> 00:02:26,000
memory cell?

41
00:02:26,000 --> 00:02:26,000
Okay.

42
00:02:27,000 --> 00:02:32,000
Uh, see, already we have discussed about this entire representation, right?

43
00:02:32,000 --> 00:02:34,000
Uh, with respect to the LSTM.

44
00:02:35,000 --> 00:02:40,000
Now, I will just give you a simplistic, uh, a simplistic representation.

45
00:02:40,000 --> 00:02:45,000
Now, with respect to LSTM, you know that there are two important line.

46
00:02:46,000 --> 00:02:49,000
There are two important line, okay.

47
00:02:49,000 --> 00:02:53,000
One is for the LSTM right.

48
00:02:53,000 --> 00:02:56,000
So this is for the long term memory.

49
00:03:00,000 --> 00:03:06,000
And there is also one more which is specifically for the short term memory.

50
00:03:07,000 --> 00:03:07,000
Okay.

51
00:03:07,000 --> 00:03:10,000
Short term memory.

52
00:03:10,000 --> 00:03:19,000
You know this this short term memory how it is basically denoted by this is nothing but c t minus one.

53
00:03:19,000 --> 00:03:23,000
This is nothing but c t minus one.

54
00:03:23,000 --> 00:03:26,000
Let me just go ahead and write it down c t minus one.

55
00:03:26,000 --> 00:03:31,000
And this is the previous hidden state HT minus one okay.

56
00:03:31,000 --> 00:03:36,000
Here we give the input okay x of t.

57
00:03:37,000 --> 00:03:44,000
As soon as we give the input the first important I'll, I'll just I'll put a box over here to make you

58
00:03:44,000 --> 00:03:45,000
understand.

59
00:03:45,000 --> 00:03:45,000
Okay.

60
00:03:45,000 --> 00:03:49,000
So the first important gate is nothing but for gate.

61
00:03:49,000 --> 00:03:50,000
Gate.

62
00:03:50,000 --> 00:03:51,000
For gate gate.

63
00:03:53,000 --> 00:03:56,000
Then the next important gate is my.

64
00:03:59,000 --> 00:04:02,000
Input and or candidate gate.

65
00:04:02,000 --> 00:04:04,000
Input and candidate gate.

66
00:04:04,000 --> 00:04:05,000
Uh, can.

67
00:04:05,000 --> 00:04:05,000
Sorry.

68
00:04:05,000 --> 00:04:09,000
Input gate and candidate memory okay.

69
00:04:09,000 --> 00:04:16,000
And finally third that you have is something called as output gate right.

70
00:04:17,000 --> 00:04:20,000
So I'm not writing what operation is basically happening over here.

71
00:04:21,000 --> 00:04:21,000
Right.

72
00:04:21,000 --> 00:04:23,000
But I'll add this one.

73
00:04:23,000 --> 00:04:27,000
So here you have this pointwise multiplication.

74
00:04:27,000 --> 00:04:30,000
Here you have this pointwise addition okay.

75
00:04:30,000 --> 00:04:30,000
Okay.

76
00:04:31,000 --> 00:04:35,000
And considering this I will just connect this I will just connect this.

77
00:04:35,000 --> 00:04:41,000
And finally you'll be able to see that I will be getting back to my output gate and coming back over

78
00:04:41,000 --> 00:04:41,000
here.

79
00:04:41,000 --> 00:04:43,000
So this is my CT.

80
00:04:43,000 --> 00:04:43,000
Okay.

81
00:04:44,000 --> 00:04:47,000
Now don't worry that I have missed some of the parts.

82
00:04:47,000 --> 00:04:50,000
If you really want to see the complete diagram, it looks something like this.

83
00:04:51,000 --> 00:04:55,000
But I've just combined this into some blocks and here I've actually shown it Okay.

84
00:04:56,000 --> 00:04:59,000
And this is nothing, but this is my next state of HD.

85
00:05:00,000 --> 00:05:03,000
Now, the reason why I'm showing you this, right?

86
00:05:03,000 --> 00:05:08,000
Because I need to pass all these words with respect to time to this particular gate.

87
00:05:08,000 --> 00:05:08,000
Okay.

88
00:05:09,000 --> 00:05:12,000
So how does the training specifically happen?

89
00:05:12,000 --> 00:05:19,000
See, when I have this text paragraph, you know that every word over here, every word, the first

90
00:05:19,000 --> 00:05:22,000
step is that it needs to be converted into vectors.

91
00:05:23,000 --> 00:05:25,000
It needs to be converted into vectors.

92
00:05:26,000 --> 00:05:27,000
Okay.

93
00:05:27,000 --> 00:05:30,000
First step we basically talk about it.

94
00:05:30,000 --> 00:05:36,000
And this layer also we say it as embedding layer.

95
00:05:37,000 --> 00:05:42,000
That basically means whenever I take a word first of all I need to convert this into vectors.

96
00:05:42,000 --> 00:05:47,000
Some vectors Okay, now already we have discussed about word two vec okay.

97
00:05:47,000 --> 00:05:48,000
So in word two vec what do we do?

98
00:05:48,000 --> 00:05:50,000
See word two vec.

99
00:05:50,000 --> 00:05:54,000
If I probably consider an example of word two vec as a vector okay.

100
00:05:55,000 --> 00:05:56,000
In word two vec.

101
00:05:56,000 --> 00:06:05,000
Let's consider that I am converting this word into three dimensions three dimension vector three dimension

102
00:06:05,000 --> 00:06:06,000
vector.

103
00:06:06,000 --> 00:06:11,000
Now, since I have these words, let's say see word two vec.

104
00:06:11,000 --> 00:06:12,000
How does it work?

105
00:06:12,000 --> 00:06:17,000
I hope you may be seeing you have already seen the previous videos of word two vec.

106
00:06:17,000 --> 00:06:22,000
So here what we do is that with respect to word two vec we we take one input.

107
00:06:22,000 --> 00:06:24,000
Let's say we have one word over here.

108
00:06:26,000 --> 00:06:32,000
And with respect to this particular word, if I want to probably convert this into a vector okay.

109
00:06:32,000 --> 00:06:40,000
If I really want to convert this into a vector, then, um, let's say the one of the word is, uh,

110
00:06:41,000 --> 00:06:41,000
tasty.

111
00:06:43,000 --> 00:06:43,000
Okay.

112
00:06:43,000 --> 00:06:46,000
Now in order to convert this word into a vector.

113
00:06:46,000 --> 00:06:53,000
So with the help of word two vec, what it does is that it takes some features one, two, three, like

114
00:06:53,000 --> 00:06:54,000
we are just using three dimension.

115
00:06:54,000 --> 00:06:59,000
So it will try to find out the relationship of this word with this particular feature.

116
00:06:59,000 --> 00:07:00,000
Okay.

117
00:07:00,000 --> 00:07:02,000
Now what feature it is, how it basically does it.

118
00:07:02,000 --> 00:07:06,000
I have already shown you how the entire word two vec basically training happens.

119
00:07:06,000 --> 00:07:06,000
Okay.

120
00:07:06,000 --> 00:07:10,000
But you can just consider see it's mostly like a black box.

121
00:07:10,000 --> 00:07:12,000
This is completely a black box.

122
00:07:12,000 --> 00:07:17,000
You will not be able to understand or get to know what is basically happening internally, but it will

123
00:07:17,000 --> 00:07:19,000
just try to select three words.

124
00:07:19,000 --> 00:07:22,000
And based on this relation it will assign some vectors.

125
00:07:23,000 --> 00:07:29,000
Right now most of my text is related to food.

126
00:07:29,000 --> 00:07:31,000
It is related to good bad.

127
00:07:31,000 --> 00:07:31,000
Something like this.

128
00:07:31,000 --> 00:07:32,000
Right.

129
00:07:32,000 --> 00:07:35,000
So I will just create three vectors okay.

130
00:07:35,000 --> 00:07:40,000
Over here one will be finding the relationship with respect to good.

131
00:07:40,000 --> 00:07:46,000
Then one will be finding the relationship with respect to bad and one will be finding the relationship

132
00:07:46,000 --> 00:07:48,000
with respect to a healthy.

133
00:07:48,000 --> 00:07:50,000
Okay, so let's say I have selected this three.

134
00:07:51,000 --> 00:07:58,000
And based on this three, we will try to take each and every word and we'll try to find out one three

135
00:07:58,000 --> 00:07:58,000
dimension.

136
00:07:58,000 --> 00:08:02,000
Let's say as soon as I read the burger looks tasty.

137
00:08:02,000 --> 00:08:10,000
So here when we say tasty I will be getting a vector for good somewhere as 0.9.

138
00:08:11,000 --> 00:08:16,000
Um, for bad since it is not bad right now, so it will be 0.0 for healthy.

139
00:08:16,000 --> 00:08:18,000
Still no information is given.

140
00:08:18,000 --> 00:08:20,000
So let's say that this is just point one.

141
00:08:20,000 --> 00:08:23,000
When we are saying tasty, it may be somewhat healthy.

142
00:08:23,000 --> 00:08:23,000
So.

143
00:08:23,000 --> 00:08:31,000
So as you all know in LSTM RNN what we do we keep on giving every input over here, right?

144
00:08:31,000 --> 00:08:32,000
One input at a time.

145
00:08:33,000 --> 00:08:37,000
Um, and let's say for tasty, this is my vectors that I'm specifically giving.

146
00:08:37,000 --> 00:08:38,000
Right.

147
00:08:38,000 --> 00:08:40,000
This is the vector that I'm giving and I'm considering.

148
00:08:40,000 --> 00:08:41,000
Hey it is three dimension.

149
00:08:42,000 --> 00:08:47,000
So similarly for every word what we do, we give this kind of dimension from here.

150
00:08:47,000 --> 00:08:47,000
Right.

151
00:08:48,000 --> 00:08:53,000
But at with respect to timestamp, first of all we will go ahead and let's say, uh, I go ahead and

152
00:08:53,000 --> 00:08:55,000
say is that I went to Restaurant and order burger.

153
00:08:55,000 --> 00:08:59,000
So every three, three, three dimensions will probably be going on over here.

154
00:08:59,000 --> 00:09:02,000
Now with the help of forget get what will get, what will basically happen.

155
00:09:03,000 --> 00:09:09,000
We need to find out like what information needs to be passed through this memory cell.

156
00:09:09,000 --> 00:09:10,000
Right.

157
00:09:10,000 --> 00:09:13,000
So here we have a memory cell.

158
00:09:13,000 --> 00:09:16,000
So here you can see I went to the restaurant and ordered burger okay.

159
00:09:16,000 --> 00:09:17,000
And ordered burger.

160
00:09:18,000 --> 00:09:24,000
So once I give this entire information right now I know I don't want to skip anything from the previous

161
00:09:24,000 --> 00:09:27,000
memory because nothing new has come.

162
00:09:27,000 --> 00:09:32,000
But when I say the burger looked tasty and crispy, I know because this tasty and crispy is going to

163
00:09:32,000 --> 00:09:33,000
be used somewhere.

164
00:09:33,000 --> 00:09:40,000
So what I will do when these words come from this input candidate kit, right input gate and candidate

165
00:09:40,000 --> 00:09:41,000
memory.

166
00:09:41,000 --> 00:09:44,000
We will try to add this information like tasty and crispy.

167
00:09:45,000 --> 00:09:51,000
We will add in the previous memory cell over here, because this tasty and crispy will be getting used.

168
00:09:51,000 --> 00:09:54,000
Now when this tasty and crispy will be coming, right.

169
00:09:54,000 --> 00:09:58,000
You'll be seeing that once I add in this input a candidate.

170
00:09:58,000 --> 00:09:59,000
Get this.

171
00:09:59,000 --> 00:10:03,000
All vectors will have more better values.

172
00:10:03,000 --> 00:10:06,000
Let's say over here it will be 0.9 here.

173
00:10:06,000 --> 00:10:08,000
Uh bad about bad.

174
00:10:08,000 --> 00:10:10,000
It has not been told anything but about healthy.

175
00:10:10,000 --> 00:10:12,000
Uh, is not good for health.

176
00:10:12,000 --> 00:10:14,000
It is basically, uh, the next sentence, it is actually coming up.

177
00:10:14,000 --> 00:10:15,000
Right.

178
00:10:15,000 --> 00:10:19,000
So right now, if I say the burger look tasty and crispy, so it will be just point one with respect

179
00:10:19,000 --> 00:10:20,000
to health.

180
00:10:21,000 --> 00:10:25,000
So this at this point of time, this information is going over here.

181
00:10:25,000 --> 00:10:29,000
And this will further get added to the memory because it is talking about tasty and crispy.

182
00:10:30,000 --> 00:10:33,000
Now when the next line goes that is.

183
00:10:33,000 --> 00:10:35,000
But the burger is not good for health.

184
00:10:35,000 --> 00:10:39,000
So when this comes, since it is saying not good for health, right?

185
00:10:39,000 --> 00:10:40,000
Not good for health.

186
00:10:40,000 --> 00:10:41,000
Okay, fine.

187
00:10:41,000 --> 00:10:43,000
This information will probably go now.

188
00:10:43,000 --> 00:10:49,000
The vector is going okay, but here you will be able to see that after this particular operation.

189
00:10:49,000 --> 00:10:56,000
Since we are talking about not good for health right now, my vectors will get changed to something

190
00:10:56,000 --> 00:10:57,000
like this.

191
00:10:57,000 --> 00:10:59,000
Now we are talking.

192
00:10:59,000 --> 00:11:00,000
See?

193
00:11:00,000 --> 00:11:01,000
Good is already there.

194
00:11:01,000 --> 00:11:02,000
Point eight.

195
00:11:02,000 --> 00:11:07,000
Okay, let's say the good will reduce a little bit, but bad if I talk about it will become point four

196
00:11:07,000 --> 00:11:09,000
because bad is increasing.

197
00:11:09,000 --> 00:11:09,000
Okay.

198
00:11:09,000 --> 00:11:11,000
It is saying not good for health right.

199
00:11:11,000 --> 00:11:12,000
Not good for health.

200
00:11:12,000 --> 00:11:17,000
So if I say not good for health again this will become point zero right.

201
00:11:17,000 --> 00:11:19,000
So here bad value is also increasing.

202
00:11:19,000 --> 00:11:21,000
Good value has decreased right.

203
00:11:21,000 --> 00:11:25,000
If I say good value may have decreased to point seven.

204
00:11:27,000 --> 00:11:28,000
Point seven.

205
00:11:29,000 --> 00:11:29,000
Okay.

206
00:11:30,000 --> 00:11:33,000
Because here we are trying to find out the relationship of every word.

207
00:11:33,000 --> 00:11:36,000
So if I say health with good bad and healthy right.

208
00:11:36,000 --> 00:11:38,000
If I try to find out this relation.

209
00:11:38,000 --> 00:11:40,000
So this value will keep on decreasing.

210
00:11:40,000 --> 00:11:41,000
Right.

211
00:11:41,000 --> 00:11:44,000
Then it says it has a lot of fats and cholesterol.

212
00:11:44,000 --> 00:11:49,000
So when we go to this sentence and then we pass the word now what will happen?

213
00:11:49,000 --> 00:11:53,000
This good value will still keep on decreasing because we are saying, hey, it has a lot of fats and

214
00:11:53,000 --> 00:11:53,000
cholesterol.

215
00:11:53,000 --> 00:11:55,000
So this becomes bad.

216
00:11:55,000 --> 00:11:56,000
Bad value will increase.

217
00:11:56,000 --> 00:12:03,000
Now the this healthy value, healthy value will also like increase or decrease, right.

218
00:12:03,000 --> 00:12:09,000
Initially let's say uh over here when we say it is not good for health, it is completely zero.

219
00:12:09,000 --> 00:12:10,000
Right.

220
00:12:10,000 --> 00:12:13,000
And here we have said, okay, it has a lot of fats and cholesterol.

221
00:12:13,000 --> 00:12:14,000
So it will be zero right?

222
00:12:14,000 --> 00:12:16,000
Initially it can be one.

223
00:12:16,000 --> 00:12:20,000
But this burger was made with whey protein and only vegetables.

224
00:12:20,000 --> 00:12:22,000
When this sentence comes what will happen.

225
00:12:22,000 --> 00:12:27,000
This vector will further change into something like let's say uh now we are talking about more good

226
00:12:27,000 --> 00:12:28,000
things, right?

227
00:12:28,000 --> 00:12:35,000
So this will increase, the bad part will decrease, and the good and the healthy part may become 0.9

228
00:12:35,000 --> 00:12:39,000
because we are seeing that, hey, it is made up of whey protein and only vegetables, right?

229
00:12:39,000 --> 00:12:43,000
So this vector will keep on changing and that is how word two vec work.

230
00:12:43,000 --> 00:12:50,000
Because we will try to find out the relationship between this word to this particular features given

231
00:12:50,000 --> 00:12:51,000
by word two vec okay.

232
00:12:52,000 --> 00:12:55,000
And then we will keep on giving that particular vectors.

233
00:12:55,000 --> 00:13:01,000
Now when this value changes we now know what information we need to remove from the previous state and

234
00:13:01,000 --> 00:13:03,000
what new information needs to be added.

235
00:13:03,000 --> 00:13:06,000
Because see, this tasty and crispy is again going to come somewhere.

236
00:13:06,000 --> 00:13:11,000
So we are just going to add some information about tasty and crispy over here okay.

237
00:13:11,000 --> 00:13:15,000
Let's say burger is not good for health.

238
00:13:15,000 --> 00:13:19,000
And later on we don't get anything about good for health or bad for health.

239
00:13:19,000 --> 00:13:19,000
Something like that.

240
00:13:19,000 --> 00:13:20,000
Right?

241
00:13:20,000 --> 00:13:25,000
Anything I will not be having a reference of not good or not bad over here.

242
00:13:25,000 --> 00:13:25,000
Right?

243
00:13:25,000 --> 00:13:27,000
So I may remove some of the information over there.

244
00:13:27,000 --> 00:13:32,000
So when, when see over here you will be able to see that I am reducing some value.

245
00:13:32,000 --> 00:13:33,000
I'm increasing some value.

246
00:13:33,000 --> 00:13:33,000
Right.

247
00:13:33,000 --> 00:13:35,000
How this is actually happening.

248
00:13:35,000 --> 00:13:37,000
This is actually happening based on the sentence.

249
00:13:37,000 --> 00:13:42,000
And this is only why when we reduce something that basically means we are removing something from the

250
00:13:42,000 --> 00:13:46,000
memory, maybe, or we are increasing something in the memory itself.

251
00:13:46,000 --> 00:13:46,000
Right.

252
00:13:46,000 --> 00:13:48,000
So this is for your short term.

253
00:13:48,000 --> 00:13:50,000
This is for your long term.

254
00:13:50,000 --> 00:13:50,000
Right.

255
00:13:50,000 --> 00:13:57,000
And this is how your entire forward propagation and the backward propagation will keep on happening

256
00:13:57,000 --> 00:13:58,000
with updation of the weights.

257
00:13:58,000 --> 00:13:59,000
Okay.

258
00:14:00,000 --> 00:14:03,000
So I hope you are able to make sense.

259
00:14:03,000 --> 00:14:09,000
Like how each and every vector changes and how your forget gate, input gate and output gate will keep

260
00:14:09,000 --> 00:14:14,000
on working with the help of forget gate we will just be doing one work.

261
00:14:14,000 --> 00:14:20,000
Is that what information I need to let through this particular memory and what information I need to

262
00:14:20,000 --> 00:14:21,000
forget?

263
00:14:21,000 --> 00:14:22,000
Make it forget.

264
00:14:22,000 --> 00:14:30,000
Okay, now as a training will happen initially when I say, uh, the burger look tasty and crispy,

265
00:14:30,000 --> 00:14:32,000
this entire information will go over here.

266
00:14:32,000 --> 00:14:38,000
Since this is a new information right from the previous information, I may remove something.

267
00:14:38,000 --> 00:14:44,000
Let's say over here also, we have a three dimensional vector, let's say about good bad or healthy.

268
00:14:44,000 --> 00:14:48,000
So good was 0.0, bad was 0.2.

269
00:14:48,000 --> 00:14:50,000
Healthy was let's say 0.1 previous cell state.

270
00:14:51,000 --> 00:14:54,000
Now the burger looked tasty and crispy.

271
00:14:54,000 --> 00:14:57,000
When this information is going now input gate what it is going to do.

272
00:14:57,000 --> 00:15:03,000
It is going to convert this entire vector where the good value will increase, okay.

273
00:15:04,000 --> 00:15:05,000
It will increase to 0.4.

274
00:15:05,000 --> 00:15:07,000
Let's say I'm just taking it as an example.

275
00:15:07,000 --> 00:15:11,000
When I talk about bad this will decrease okay.

276
00:15:11,000 --> 00:15:13,000
And when I'm talking about healthy right.

277
00:15:13,000 --> 00:15:15,000
This may have some value like 0.4.

278
00:15:16,000 --> 00:15:16,000
Right.

279
00:15:16,000 --> 00:15:21,000
Then when I pass this three this third sentence I need to add some more information.

280
00:15:21,000 --> 00:15:22,000
I need to skip some more information.

281
00:15:22,000 --> 00:15:23,000
Right.

282
00:15:23,000 --> 00:15:31,000
So from this entire cell, what I will do when we go to the next timestamp, I may I may decrease this

283
00:15:31,000 --> 00:15:31,000
value now.

284
00:15:31,000 --> 00:15:31,000
Right.

285
00:15:31,000 --> 00:15:35,000
Because now it is saying not good and not uh, not good for health.

286
00:15:35,000 --> 00:15:36,000
So what will happen?

287
00:15:36,000 --> 00:15:38,000
We will add some information.

288
00:15:38,000 --> 00:15:39,000
We'll forget some information.

289
00:15:39,000 --> 00:15:44,000
So if you're forgetting some information now we are saying that hey, this is not good.

290
00:15:44,000 --> 00:15:46,000
The burger is not good for health.

291
00:15:46,000 --> 00:15:48,000
So this value will become 0.2.

292
00:15:48,000 --> 00:15:51,000
Similarly, this second one is specifically for bad.

293
00:15:51,000 --> 00:15:53,000
This will increase to 0.4.

294
00:15:53,000 --> 00:15:59,000
And if I say healthy since I have written previous point for but it says it is not good for health,

295
00:15:59,000 --> 00:16:00,000
this may become 0.1.

296
00:16:00,000 --> 00:16:08,000
Okay, so this how the entire, uh, the forget gate, the input gate, candidate gate, and the output

297
00:16:08,000 --> 00:16:10,000
gate actually works.

298
00:16:10,000 --> 00:16:15,000
But you really need to remember forget gate will help you forget some of the information based on the

299
00:16:15,000 --> 00:16:20,000
context, switching only the important information that will be relevant for the long term context that

300
00:16:20,000 --> 00:16:22,000
will be used and it will keep on going out.

301
00:16:22,000 --> 00:16:22,000
Right.

302
00:16:22,000 --> 00:16:25,000
So I hope you got this complete idea.

303
00:16:25,000 --> 00:16:30,000
This was a quick, uh, way of making your understanding training data with LSTM, RNN.

304
00:16:31,000 --> 00:16:35,000
Um, I tried my level best, uh, to explain you all this things.

305
00:16:35,000 --> 00:16:36,000
Okay, so.

306
00:16:36,000 --> 00:16:36,000
Yes, uh, this was it.

307
00:16:36,000 --> 00:16:38,000
I will see you all in the next video.

308
00:16:38,000 --> 00:16:38,000
Thank.