Types of RNN | Recurrent Neural Network Types | Deep Learning Tutorial 34 (Tensorflow & Python)

Key Takeaways

in this short video we'll quickly cover different types of recurrent neural networks in our last video we looked at the example of named entity recognization where given a statement you can tag each of the word whether it's a person or not so in this case uh for example double baby and yoda these are the person so there is one as an output so this is the case of many too many rnn because you have many inputs or x and you have many outputs or y so generic way you can say something like this x1 x2 all the way till xt and y 1 y 2 all the way till y t so this is a many to many recurrent neural network the other use case is language translation here in this statement there are three input words and three output words but actually you know that the output words could be different okay and in order to translate this you need to pass the whole statement so the architecture will look something like this where in your rn and hidden layers you would input all your words initially after you are done with your last word then your rnn will start producing the output and we looked into this a little bit in the last video so if you have not seen my last video on what is rnn i highly recommend you watch it because this video is a continuation of that video this is also another case of many to many and generic way of representing it would be using this notation so here xt means the last word t is the last word and x1 x2 these are like different words in a statement sentiment analysis in here the input would be a paragraph and the output would be your review for example these are like product reviews okay so given a text of productivity you can save it it is one star or two star so the rnn would look something like this it will be many to one so you have many words as an input but the output is one which is your product review and generic way of representing this would be this x1 x2 all the way xt and then there is y hat which is a predicted output the other use case is music generation where you can pass a simple like a sample note or like a seed node and then you can ask rnn to produce a music melody you know rnn can sing a song or even poetry writing you can feed a single word or like a seed word and it can it can write a poem so in this case it this is the case of one too many where your input node is just one sometimes you it sometimes you know you don't have input at all and you can ask rnn to just produce some random music and it will do it for you and the output has you know y 1 y hat y 2 hat y 3 hat and so on that's why it's one too many and this is a generic way of representing this architecture so that's all i had i think for this video i hope that clarifies your understanding on different types of rnn we will be looking into lstm gru those kind of special cases uh in future videos but in this video we covered one too many many too many and many to one type of rnn i hope you like this if you do please give it a thumbs up and share it with your friends thank you