connor @notwa@cybre.space

#introductions hi, i'm connor.

i'm interested in #programming, #machinelearning #ai, #hacking, #reversing, learning and playing with math. i enjoy writing #cpp, #python, and #lua.

sometimes i hack #zelda games for the #n64, but more generally i follow wherever my interests take me. i tend to avoid big projects because of that. i ramble a lot about whatever i'm working on or learning about.

i've helped my sister raise a dozen pet #rats! i also love #tea, #cats, #anime, and #music.

better but still not enough circuits :/

the singer is a crook, wo-ay-yo
the singer is a crook, wo-ay-yo
the kitchen is the cook, wo-ay-yo
the scissors are the barbers, wo-ay-yo
the singer is a crook, wo-ay-yo
the chorus, full of actors…

also the stuff on the right (red and green science) didn't really change at all because i had blueprinted it before my playthrough :P everything else (besides the plastic) was freehanded though

there needs to be way more iron pressure and green circuit assemblers, so the middle will probably change the most

i copied my factory into creative mode and cleared out the miners/smelters to fix it up. there's enough things i want to change that it's almost embarrassing to post, but ah well.

cool anxiety for no reason again. guess i'll try firing up factorio creative mode until my mind has settled down

it doesn't really make sense to crank up analog gain when some things could get away with a digital left shift :/

I really wish digital (dynamic range) compressors were commonplace because audio levels are all over the place, especially when watching videos on phones etc

my problems are usually so poorly conditioned than gaussian blur might be the best bet regardless

if you only have savestates every n>1 frames then you have a fraction of starting places to base your inputs off of

i think i sorta prefer the jankiness of non-finely-grained reversal in emulators

i think i'm way more comfortable with these concepts now that i have something simple to bring it back to

it's funny because the wikipedia page on QR makes it look complicated at a glance, but the actual algorithm on the gram-schmidt page is dead simple

i think tef was saying the "not QR" thing i was doing was gram-schmidt so i checked that out today and yep that's exactly what it is

well conjecture is the wrong word since they have proof under assumptions but y'know what i mean

at a glance it looks like this relies on n+1 samples i.e. one for each input axis plus an unperturbed sample, so it might not be useful for some of my problems. anyway i guess their conjecture is that it's best to sample perturbations from a random orthonormal matrix.

"Linear interpolation gives better gradients than Gaussian smoothing in derivative-free optimization"
you have my attention

i think google scholar's citations for this paper are broken because it uses an old title :/
or there might just not be any. but the title is definitely wrong

plus more abstraction = more pushing errors up the stack, which C is pretty happy to let you mess up

i'm looking at the data structure code i wrote in ciel and i want to refactor it but i don't know if anything i do will be as simple/minimal. i kinda have to handle all the ugly edge cases *somewhere*