i'm probably only going to actually read like 50 pages but still
-
-
Show this thread
-
okay so it looks the the thesis is that we have an algorithm for copying aesthetics and it's called autoencoding lossless inputs=canon, feature function=worldview, projection function=mimesis more on what this means later in the paper
Show this thread -
okay so some good definitions canon => set of privileged objects worldview => schema of interpretation mimesis => capacity for reproduction and representation they fuse together, making each other stronger
Show this thread -
-
autoencoding => "the process of determining a worldview and a canon in developing a method of mimesis" okay then there's a section on how autoencoders work but i think i'm better off reading the wikipedia article due to my cs background https://en.wikipedia.org/wiki/Autoencoder …
Show this thread -
so autoencoders mathematically compress inputs to their most efficient signal:noise ratio the paper translates this to mean that they create likenesses of their inputs, in order to get the gist across without spending too many resources this is about right
Show this thread -
there's a joke about the authors of the paper being good autoencoders here somewhere
Show this thread -
okay now they're explaining data science and machine learning 101 the explanation passes my sniff test, looks like they aren't talking out of their asses actually a pretty good explanation, i highly recommend reading it if you want to know the super basics of how ml algos work
Show this thread -
lots of "hey are you even allowed to make mathematical abstractions of real world things? turns out: yes" i repeat: this is a good intro to data science stuff for non-data science people
Show this thread -
so one big thing autoencoders do is dimensional reduction this means they draw out the most salient features of a type of thing, and what best differentiates examples from each other in order to understand an aesthetic, one must do a dim. red. on it
Show this thread -
-
an aesthetic is some scheme of understanding such that a certain type of thing becomes very easy to describe you can "train" an aesthetic on just the canon and it'll work just as well as training it on the whole thing
Show this thread -
okay there's a bunch of stuff on modernism I lack the context to understand just gonna skim past this
Show this thread -
"ambient mood" is the sort of vibe or aesthetic or worldview a work embodies they're fake understandings of reality (standard map/territory stuff here) they can't be talked about directly, they're illegible and exist only as abstract patterns of sense-data
Show this thread -
Aesthetics have vocabularies, and these are what the autoencoders find
Show this thread -
multiple items sharing an aesthetic makes it easy to compare them, because you can do it using the terms of the aesthetic i.e. you already ran the dimensional reduction, so you know the important variables immediately
Show this thread -
we can define an aesthetic as a small number of transformations, that allow you to turn one item into another, with the set of things that can be created with these transformations being in the aesthetic
Show this thread -
okay thats all i got from reading this going to add "can read arcane academic papers" to the list of things i brag abouthttps://twitter.com/imhinesmi/status/1225907259219337219 …
Show this thread
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.