Payaso
Cast
gollark: (Apparently you can maybe get somewhat better performance from image recognition neural networks by feeding them "DCT" things which also conveniently happen to be what JPEG images contain, but almost nobody does this?)
gollark: Also that.
gollark: The image is just 3 matrices of R/G/B values.
gollark: There are 129057189471894718247141491807401825701892912 random details and things but that's the gist of it.
gollark: Then, you just move it a little bit toward lower loss (gradient descent).
This article is issued from Wikipedia. The text is licensed under Creative Commons - Attribution - Sharealike. Additional terms may apply for the media files.