Is there any other materials that derives the LSTM back propagation and carousel of error except the original paper? (I could not understand it, sorry).

I tried deriving and got stuck, and asked the following question: LSTMs - Data Science Stack Exchange question, however, it doesn't seems that there are much people interested in hand derivation of LSTMs.



One tool that may be helpful is Aiden Gomez's blog post. Its strength lies primarily in the fact that he runs through a toy/numerical example, which when paired with the original paper/thesis , serves as a great foundational tool.

I did take a look at the site you mentioned in your other question, it's actually an excellent resource. I'll hop on over and try and clarify what I can for you when I get a chance. It looks like you've misunderstood/overlooked notation which can happen since there are so many components involved.

It may also be worth taking a look at some code. Siraj Raval has a great video on LSTMs and includes the code in the link I've included. No libraries. I wouldn't dive too deep, but it's a great way to see the inner workings of the network.

As far as the CEC goes, there is a reddit post. If you're looking for a more rigorous handling of this topic you can either reference the original paper or often cited paper: On the Difficulty of Training Recurrent Neural Networks.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.