madhadron

@grimalkina@mastodon.social It's almost certainly them just assuming that their good p-values and R^2's mean that their choice of factors is correct. My experience is that if the statistics in a paper don't seem to make sense somewhere, it's because they're wrong.

@mark_histed@fediscience.org Oh, I remember that. It started when I was still working at Facebook. It was clearly a hobby project and I couldn't take it seriously.

@tyrell_turing@fediscience.org @NicoleCRust@neuromatch.social (2/N) Counterexample: a cofunction that receives some other cofunction in its codata and replaces itself with that. There is no corresponding function. Given that brains learn behavior from the outside world, I am inclined to think that functions are insufficient. It's possible that cofunctions over codata are a useful way to model brain activity, but computable functions aren't the right tool.

@tyrell_turing@fediscience.org @NicoleCRust@neuromatch.social I'm afraid this will get long, so (1/N). This is something we dealt with a lot in the functional programming community. Functions act on data. Data is fixed and fully defined at the start. Online systems are described by cofunctions acting on codata. You're asserting that every cofunction has a corresponding function under some kind of mapping. That is not true.

@NicoleCRust@neuromatch.social @knutson_brain@sfba.social @neuralreckoning@neuromatch.social @tyrell_turing@fediscience.org @awaisaftab@mastodon.social @summerfieldlab@qoto.org @Neurograce@mastodon.online @eikofried@mastodon.online I don't think definition 1 is correct, though. Computable functions describe a process taking an input state into a desired output state. The theory doesn't describe online data processing which is what the brain does.