Long Short-Term Memory (LSTM), Clearly Explained

why is tanh(new long term memory), considered potential short term memory?

Kaboom!! and Poof!

I never regret watchin your videos for my Phd thesis relating to LSTM,CNN and LSTM. QUADRPLE BAM!!!

beautifully explained

i love you

Thank You! it was really helpful..

maaaaan you just saved my Introduction to Deep Learning exam... thnk uuuuuu!

🥲 this is amazing

The best explanation of LSTM

Quadro bam!!!

Lovely

so does it mean we dont need to do any back-propagation?

You are a unique, super talented artist Professor.
Can't explain how finally I understand Gradients, RNN, & LSTM. To the Transformers QUADRUPLE BAM!! <3

Literalmente la cabra

great video, i was able to easily understand LSTM.

How you are explaining the complicated concepts in much easier way to understand by viewers you have the extraordinary teaching skill thanks man

Tripple BAM 💥💥💥

loved it

Thanks so much! really helped me to understand concept for LSTM!! definitely gonna recommend this channel to my friends ...

It is a very nice video with intuitive explanation. However concepts like constant error carusal are not described.

Sorry, but I'm confused. What's the difference between "% long term memory to remember" vs "% of potential long term memory to remember" ?

love josh, bro makes me ai engineer😂

A good amount of resources out there that document how LSTMs work, don't even go into how the numbers change throughout the network, leading to even more confusion around the topic. This video takes all that information, jargon, and confusion, and shapes it up into an extremely intuitive and concise manner. I thought that I wouldn't be able to understand this new technological breakthrough, but now I see that most literature out there fails to present the information in ways that everyone can understand. I don't like how there's an implicit gatekeeping in the world of AI, but you really broke through that wall with this video.

Thank you so much for saving many days of confusion and frustration!

yknow many people should watch and learn a thing or two with this channel, even if they got no use for it, just to realize how much of learning is accepting you don't know something, then leaving your pride behind and letting someone be very patient and understanding with you. this seems like a childs show but it has worked and works for many people. maybe we don't need to take ourselves that seriously, specially when learning it seems

why, conceptually, would you say that in stage three the short term memory absorbs information from the long term memory? whereas for the rest of the stages it makes sense how long term memory is (not)taken into account and updated, I can't intuitively understand how the short term memory has to do with the long term memory and not with only just the previous short-term memory and input. Hope the question makes sense, if not I can clarify, thanks in advance!

i have a question, is initializing the long term and short-term memories of the first cell to 0 the standard and what works? or is it just an example and there's a criteria? thanks!

Thank You for beautifully explanation.

exceptional animation. Thanks

i'm so grateful, thank u

So LSTM can predict day 5 in the example you provided. But why couldnt it predict day 4?

Josh Starmer is a gift to humanity

Extremely helpful video and astonishing ability to express difficult concept in clear terms. I have one question concerning the sigmoid formula you used. I saw the sigmoid function formulated as 1 / (1 + e^(-x)) in many text books and I was wondering why, although inputs are still mapped in the same codomain, you used a different formulation and which are the main differences in terms of function behaviour. Thanks so much anyway!

this is the best explanation video I have ever seen thank you.

thanks alot

Its a masterpiece to understand LSTM and its gates

hi, I've heard about the hidden size term in LSTM which is the number of LSTM units, can you tell me what it is?

will there be a learning algorithm for this LSTM?

Thank you josh, the best teacher i've ever had!

broo, you just saved ma dayy!!!

KABOOM!!!

Triple bam Thank you

fight night....Tomorrow is my exam

hi josh, u are a great teacher. but could you explain why the last sigmoid is necessary to the overall architecture? it seems to me that the forget gates and input gates already gives fine tuned control over how much long term information to retain, if it were v relevant, isnt it okay to just pass that through the tanh activation as the new short term memory?

Best explainer video on LSTM, I saw so many videos but only this one could clear my concepts!!
BAMMMM!!