Data Compression with SVD — Topic 36 of Machine Learning Foundations

In this video, we take advantage of the singular value decomposition theory that we covered in the preceding video to dramatically compress data within a hands-on Python demo.

[pratyushharsh]

This is the best resource available on internet for brushing up your foundation and cover those topics that were not covered at university

[touheedkhalid]

@10:36 Show Stopper for this video. Jon smiling and saying "I bet you never thought linear algebra could be so cool" 😄
can't stop smiling. Love you Jon

[saadarman]

These are some amazing videos. Never had so enjoyable and full of knowledge course before this.

[aashishrana]

Pretty cool, you made me fall in love with linear algebra and machine learning a lot way more than earlier.
This is first time I am exploring these concepts and you have a given such a beautiful explanation in such a simplistic way of such a complicated concept that even a 5 year old guy can understand.
Thanks Man!

[justsimple]

I don't think it will be easy to learn the Linear Algebra but thanks to Jon Krohn for his amazing work.

:)

[westobaba]

Hey jon!
Can’t express how much of a blessing you’ve been..thank you so much!

[OscarBedford]

I feel like I've struck gold with your channel, man.

[gabrielkelvin]

Finally all that I've learned, makes sense

[shaiknayeem]

easy understandable videos .
Thank you sir

[rezabagherian]

i think i've found a machine learning gem here

[naeimwtg]

I cannot understand this point. How did you calculate this svd_rep64?The height and width of the reconstructed image are the same as the original image.
We use what data(U S V) for the input layer of a neural network to achieve dimensionality reduction ?

[subhanbasha]

Hey Jon,
I got a question.
Dimensionality reduction is the transformation of high dimensional data to lower dimensions without losing the information. Right?

So my question is how the transformed dataset will have same set of properties of original dataset?
Considering same mean and variance?

Thanks.

[toysandgames]

please ask why to switch to grayscale mode to compress the image, but not keep the original color to compress the image

[kuladipbhowmik]

Some of our Govt websites doesn't allow image size more than 5-10 KBs....I always have hard time compressing the images using online websites...Now I know how to do it...ha ha

[AyazHuseynov]

I think each pixel can be represented with 8 bit integers, but matrix/vector elements of SVD are not 8 bit integers right? You need at least 32 bit floating point numbers in order to store these elements. Therefore 3.7% compression ratio is not correct, we can say it is about 14-15% compression ratio from bitmap to "svd". Furthermore if we compare sizes png or jpeg representation of that bitmap vs "svd", we may get even worse results(svd will be bigger than png or jpeg).

[alexzambrano]

Impossible not to see the usefullness of SVD with that example

[gaurew]

Didn't thought matrices can do this much