Types of Recurrent Neural Networks (RNNs)

what's going on everyone this is jay patel and in this video we will discuss the types of rnn in the previous video we discussed how the rnn looks like and we also saw what is the use of rnn and why can't we make a natural language processing application with the help of a simple artificial neural network so if you haven't watched my video on that you can find its link by clicking on the upper i button or you can also find the link to the entire playlist of recurrent neural network down in the description box

now in this video we will discuss the types of recurrent neural network and we'll see what kind of rnn is used for what kind of application now in the previous video we discussed these named entity recognition application whose job was to determine the entity occurring in a given sentence and we also saw how the recurrent neural network looks like now this is an example where the input length is exactly of the same length as the output length but most of the application might not be following this pattern for example we can create an application as

a movie rating system where the objective is to rate the movie based on a given comment so in this kind of applications the input can be of any size while the output is only a one single number between one to five so the recur neural network will be slightly different and it will look something like this now here we will only produce the output at the last word which means that we will pass the all the which means that we will pass all the words one by one and meanwhile we will also pass the activations

to the next time step and once we reach the end of the word at that time we will produce the output and this output will be any number from 0 to 5. now again you should consider that this output was given after considering all the words one by one in a proper sequence and these activations will change its value where at every different time stems now some applications can be exactly opposite of this for example an application such as maybe name generation can be made whose job is to generate a baby name by from based

on a given input for example if we pass mail as an input then it must generate a mail name or and or if we pass a female as an input then it must generate a female name and in this kind of application the input size will be only one and the output size can be anything and the model will look something like this as once we pass the input it will keep on generating characters until it has generated the end of the word character which means that here the model should stop now here the output

from each timestamp is also passed as an input to the next time stem and that is how the model knows which character has occurred first and which character to produce for example the vowels mostly come after the words like j or so on and if the first two characters are g and o then the third character is likely to be h or s so that the model can create john johnson or joseph or any word like these that is why the recurrent neural networks are very powerful in identifying the next words based on the words

that it has seen already now and another application is music generation now there can be an another application which might have variable input length and the variable output length and both these are not same for example in the english to spanish translation the input english sentence can be of any length and the output spanish sentence can also be of any length so at that time the rnn will look something like this where once we have passed all the input words only then it will start generating the output words and in this kind of model the

first part which only takes the input is called encoder and the second part is called as a decoder to summarize everything some applications might just have fixed output length or fixed output shape and those kind of model can be called as a many two one applications and some can be of exactly opposite type which can be of one too many and then some might be of type many too many where it can also be categorized into two categories here the input length t x is same as the output length t y and here t x

and t y are not equal so this was about the type of iron but i we haven't discussed the mathematical details behind rnn yet i have told you that this looks like a neural network but what does exactly it is actually it is called as a cell in rnn and in the next video that is what we are going to discuss like how does this cell looks like or what are the mathematical equations that goes behind this rnn cell so i hope to see in the next video and you can find its link by clicking

on the left side of this video or down in the description box and i will see you in the next one