email


rulu   ruru
 

Warning:
This site sometimes contains or links to sites that contain adult material.
It is not safe for work. If you are under your community's legal age (18 or 21),
please do not scroll down and instead leave this page immediately.

 
 
 

post Out And About

October 10th, 2017

Filed under: Uncategorized — admin @ 7:58 am

British Bum, Yo…

_______

In The Name Of Science

2 Comments »

  1. “Dr. Kosinski and Mr. Wang say that the algorithm is responding to fixed facial features, like nose shape, along with “grooming choices,” such as eye makeup.”

    I hate to break it to people, but humans can typically make this guess too. Shaped eyebrows on a man? I have at least a 50/50 shot saying they’re gay or in Hollywood (both?). So my success in certain circumstance will be high too.

    Beyond this, working with people using AI to render images, it’s really easy to lead the algorithm based on what you’re exposing it to. In fact your own data source can really undercut what you’re trying to do, this is because AI networks have to be trained. Chances are, like training your dog, it will pick up on things you didn’t expect and hardcode them unintentionally. So in training the algorithm if they used stereotypical “gay cues” like the “grooming choices” you consider gay then you’ve automatically biased the algorithm. (Bias is introduced through weights and if you’ve used them correctly.) So anyone gay that doesn’t shape their eyebrows will probably get flagged as heterosexual, for example. It’s weighted, but those weights can be fucked up.

    https://stackoverflow.com/questions/2480650/role-of-bias-in-neural-networks

    Comment by VFXer — October 11, 2017 @ 12:08 am

  2. If anyone is interested in reading a great book on algorithm’s pick up a copy of Weapons Of Math Destruction. Highly recommend!

    Comment by Ethan — October 11, 2017 @ 6:25 pm

RSS feed for comments on this post. TrackBack URI

Leave a comment

ruldrurd
©2014 Dragon Media
Entries (RSS) and Comments (RSS)