Automating image filtering using neural networks
-
Has NetFree looked into using neural networks to approve pictures?
It sounds impossibly sci-fi that an algorithm can automatically and accurately do the kind of image filtering that NetFree requires, but it seems to me that it is actually possible with today's technology and relatively straightforward to do.
Feeding the algorithm with a set of input pictures and simple output values, such as NetFree's current image database, and later using the algorithm to guess the correct value for an unseen picture, sounds like quite a classic use case for neural networks.
Since neural networks are probability based, they could be used for example, over 60% suspicion: auto block. under 10%: auto allow. everything else filtered by humans and used to continue training of the network. This will dramatically decrease the load on NetFree's staff, and the percentages can be tweaked when and if confidence in the algorithm increases.
-
@SilverRampart said in Automating image filtering using neural networks:
Has NetFree looked into using neural networks to approve pictures?
I am under the impression that yes, they have looked into it and deemed it insufficiently accurate. (They aren't living in a cave and they are aware of recent advances in AI etc.)
If you are confident that your approach can rival the current approach and you are capable of implementing something with those success rates, I would suggest you contact the administration directly. Perhaps the matter would be reconsidered.(I don't speak on behalf of NetFree, this is my own conjecture)
-
@SilverRampart I would find it hard to believe that it is possible at this time. keep in mind that the image reviewers aren't just identifying items in the image, there can be images with thousands of items (such as crowds) and only a human can take it all in and identify a problem immediately.
(This is besides the fact that Google, Microsoft and others do have image identifying software and they are not nearly as accurate as is NetFree)
