SICP for neural nets

[Anonymous No.12273078]

What is the SICP for neural networks? Does it even exist yet?

[Anonymous No.12273236]

>>12273078

[Anonymous No.12273251]

>>12273078
training a network is just minimization and the output of the trained network is just matrix arithmetic.

ML is just a very lazy way to do fitting. the hallmark of NNs is that you use a fit that has way too many parameters with highly nonlinear terms and rely on huge statistics in the training to get a fit that’s not shit. anyone who ever did human-tier science or statistics knows exactly why this is a very shitty way to do things and why even using the word “model” to describe what they are doing is severely deluded

[Anonymous No.12273273]

>>12273078
>SICP for neural nets
what do you mean by this? You mean an introductory text that's good, and in some instances great, but by far does not teach nor demonstrate the depth or breadth of its subject, despite codemonkeys holding it as the bible of all things related to computers?

[Anonymous No.12273530]

>>12273273
I learned a lot from SICP, like "oh that's what a computer is doing" in a more profound way. Was hoping for a similarly enlightening book about NNs. I see a few textbooks on Amazon but they're ~$hundos and I don't wanna buy garbage. Until then I'm just working on calc and vector algebra in spare time. I guess if one can progress to literacy in vector calculus that's enough to start doing things on your own in this space.

[Anonymous No.12273849]

>>12273251
>this is a very shitty way to do things
Oh Jeez, guess all the success of the last decade is delusional.

>>12273530
I was also looking for a comprehensive source, but found none. I read Bishop, Goodfellow, Deisenroth and some other one I can't remember because it sucked, several blog posts and papers. The mixture of all of these sources gave me good understanding, but I'm 99% that a comprehensive guide doesn't exist yet. The reason might be that the field is advancing so quickly that information today is already outdated next month.

[Anonymous No.12273877]

>>12273530
>buying books

[Anonymous No.12274656]

>>12273078
i am using as a starting point

https://cs.stanford.edu/people/karpathy/convnetjs/index.html
https://cs.stanford.edu/people/karpathy/convnetjs/demo/automatic.html

everything in browser and lots of ui to understand what's going on, i hope when i get my bearings i'll have an easier time moving forward to a "heavy" cuda enabled environment. with preference to some sort of array language.

[Anonymous No.12274686]

>>12273530
>I learned a lot from SICP, like "oh that's what a computer is doing" in a more profound way.
there is no such book, not just because it hasn't been written, but because we don't know what NNs do in a similar way to the SICP explanations.
How exactly NNs do and why they are effective is still an active and arduous area of research(frankly moving at snails pace, as it's legit difficult).

[Anonymous No.12275060]

Why has nobody asked what SICP is?
What the hell is SICP?

[Anonymous No.12275103]

>>12275060
A really fucking good book.

[Anonymous No.12275472]

>>12275103
Ah I see, thanks.

[Anonymous No.12275832]

>>12275060
if you cant google "SICP" you dont have enough intelligence to read the book so dont bother

[Anonymous No.12275925]

>>12275832
>hurrr durrr Google know eberthing
>just Google everything I'm too lazy to communicate normally
You're the LIP of the SC.

[Anonymous No.12275955]

>>12275103
so good that even its author said the book is irrelevant