b3ta.com links
You are not logged in. Login or Signup
Home » links » Link 1318017 | Random (Thread)

This is a normal post It's an interesting question
As I understand it, the technology works by recognition. So presumably if it says you look like a gorilla, you do look a bit like a gorilla on some objective level.
 (SnowyTheWereRabbit the Leporid from Hell, Wed 1 Jul 2015, 9:03, , Reply)
This is a normal post "Objective" in the sense of "according to the algorithms that someone wrote".
Not in any god's-eye-perspective sense.
 (I helped save b3ta! Enzyme is powered by sunlight, Wed 1 Jul 2015, 9:06, , Reply)
This is a normal post They most likely use machine learning
not that it's an excuse for not testing it thoroughly

It is possible the training data set didn't include enough photos of black people.
 (I helped save b3ta! Mudface The web isn't callipygian, Wed 1 Jul 2015, 9:12, , Reply)
This is a normal post Machine learning isn't de novo, though.
You can't just pop a Raspberry Pi next to your window and expect it to know more about the world by the end of the day.
 (I helped save b3ta! Enzyme is powered by sunlight, Wed 1 Jul 2015, 9:22, , Reply)
This is a normal post I think with the new deep learning methods they may not know excatly which features
the system has learnt to recognise objects.

( www.b3ta.com/links/Computer_neural_networks_on_acid comes to mind )

But again, they had to train it on a comprehensive set of photos + accurate captions, and test it properly.
 (I helped save b3ta! Mudface The web isn't callipygian, Wed 1 Jul 2015, 9:42, , Reply)
This is a normal post As I understand it,
The algorithm will look at the photograph, and use imaging techniques to recognise certain attributes, e.g. eyes, mouth, nose etc.

They then record the sizes, colours, distances, positions etc. of all of these data points, and then run statistical analysis of all of that data, building multidimensional arrays out of it. Given enough data, you can chuck in a new set of points, and it will give you a percentage value for which pre-defined categories it's likely to fit into.
 (I helped save b3ta! HoratioFellatio Taking the the out of psychotherapist, Wed 1 Jul 2015, 14:37, , Reply)
This is a normal post More about the black box issue:
stats.stackexchange.com/questions/93705/meaning-of-a-neural-network-as-a-black-box
 (I helped save b3ta! Mudface The web isn't callipygian, Wed 1 Jul 2015, 15:25, , Reply)