Professor asks me to do research on deep complex neural networks, as in neural networks that perform on complex numbers.

Meanwhile me: "Google, what are complex numbers?"

They're what you multiply Santa by to make him real

Complex numbers are those numbers that identity themselves as non univocally quantifiable. It has something to do with gender math

@Root I promise, I've been a good researcher

@Root I think the number you are looking for is 42.

@Mosesrocks If you want complexity, take the square root of a salesman's IQ.

@willhertz nowadays math has gender too and you are not allowed to assume their gender

I think my early christmas gift this year is that @C0D4 , @Root and @devTea all thought that my rant was worthy of their recognition.

Feeling #*blessed*. XD

@Mosesrocks but it's a zero... 😢

@C0D4 ssshhh sub 5mins ninja edit, you saw nothing

@Root I ran out of decimal points for the sales team, some days I think π would be easier to solve.

Complex Numbers are a joke that got far too real.

@Root "If you want complexity, take the square root of a salesman's IQ."

I will definitely use this at some point. This is a good one. :)

@C0D4 badass insult. Love it!

@frankot trust me they are not always covered during precollege time.

@frankot I'm not saying that I am completely oblivious to what complex numbers are, and we did in fact study them in the last two years of school, but I didn't care that much at the time (If you would have told 17 year old me that I would need complex numbers in 7 years I might still have not paid attention).

I'm joking about it, not because I don't know what complex numbers are, but because I'm not sure I really understand what they signify in the real world. I understand the fourier transform for example and how powerful and useful a domain conversion can be.

About the research thing; I guess it pays off to be somewhat decent at programming stuff in keras and pytorch, as they make it easy enough to code up stuff and try it out.

@frankot I studied about fourier transforms last semester by myself, since I needed to visualize some audio signals as spectrograms. And I didn't actually know what they actually were.

I don't have a strong background in mathematics since I most of my Bachelor studies focused on Software development. But now I find ML much more creatively stimulating, so I'm trying to pry my way into the field XD

I know you weren't implying anything, so no offense taken. I like talking about this stuff, Cheers!

forget googling it.

watch this 4 minute video, and then the 24minute one right after it in the playlist, both by 3Blue1Brown, and get your mind blown at how awesome they are and how you will understand them almost intuitively afterwards.

(correction: watch the second one, the 24minute long one first, that's a very gradual and detailed explanation, THEN watch the 4minute one, which is condensation of the 24minute one)

https://youtube.com/watch/...

you're welcome. and subscribe to him, he's amazing.

Is it even possible to do a fourier transform without complex numbers? You are literally multiplying a rotating unit complex by f(x). Is there another way to do it that I missed out on?

@Lor-inc since the whole point of complex numbers is that they're two-dimensional, as in, non-vector-like form of expressing 2d coordinates*, i'd assume no?

but I don't really know what fourrier transform does, just assuming from your question that it needs (at least) 2dimensional space.

*...now you made me have a question - are there also 3d, 4d, n-d complex numbers? if yes, how are they notated? a+bi+cj+dk...? if no, why not?

@Midnigh-shcode Not for 3d, but 4d complexes are called quaternions

@Midnigh-shcode But quaternions are pretty fucked in many ways. For example, multiplication isn't commutative.

@Lor-inc Hello matrix multiplication my old friend....

@Lor-inc Also yes you can fourier transform without complex numbers, you just have to convert it to trig thingies. Not pretty but might work.
Complex numbers in general are just a abstraction (or just a different way) of joining normal and polar coordinates

@Gregozor2121 Sounds pretty unintuitive but I see how it could be done.