Machine Learning Confronts the Elephant in the Room
quantamagazine.org
Machine Learning Confronts the Elephant in the Room
1–10 of 116 posts
Re: Machine Learning Confronts the Elephant in the Room
#2Somehow the right image where according to the description an elephant was introduced is now identical to the left photo in the article.
Re: Machine Learning Confronts the Elephant in the Room
#3Is that diagram right? The neural net does the segmentation? There's no pre-process, it literally feeds pixels in?
Re: Machine Learning Confronts the Elephant in the Room
#4Can't they implement a "virtual representation" or sort of ideal or archetypical scene in an AIs memory that's quickly accessible, and then use that to diff "new unexpected stuff" and refocus on that difference somehow?
Re: Machine Learning Confronts the Elephant in the Room
#5Wrong. They can do a second take. You should just code it in.
Re: Machine Learning Confronts the Elephant in the Room
#6Is this not a whole new attack surface?
Re: Machine Learning Confronts the Elephant in the Room
#7Somehow the right image where according to the description an elephant was introduced is now identical to the left photo in the article.
The elephant is on top of the guy's head, it's hard to see with all the boxes laid over the image.
Re: Machine Learning Confronts the Elephant in the Room
#8Somehow the right image where according to the description an elephant was introduced is now identical to the left photo in the article.
Look over the man's head ;)
Re: Machine Learning Confronts the Elephant in the Room
#9Wrong. They can do a second take. You should just code it in.
Is this sarcasm? The point of the second take is that humans often know they're confused and will go back to think about the image and remedy the confusion, whereas having a neural network just look at the image again isn't going to do anything (plus, existing architectures don't seem to have any capacity to say "I'm confused" anyway).
Re: Machine Learning Confronts the Elephant in the Room
#10Is this not a whole new attack surface?
Not really. In addition to object classes, object detection networks learn to ID teh context of an object's presence. In almost every scenario, this context is a desirable learning trait.