Adversarial input

Do any of you ever insert adversarial perturbations into your memes in an attempt to obfuscate them for the CNN employed by the censors and algorithms at the various major social media companies?

Attached: 11_carrara1.jpg (600x233, 49.15K)

elaborate further. I am slow.

Quick guide?
I want to lace stock images with merchants.

We need a tutorial OP

checked. you add noise to memes to confuse AI. in OP context, CNN is convolutional neural network

i only make memes for posting on Any Forums
i dunno wtf ur talking about

My apologies. I should mention that I am looking for methods or software to do this.

this, this is an anonymous imageboard. if you want credit for your memes try redditt.

Come on now, is it really that hard to add a new transparent layer, apply a noise mask to it with high turbulence, and then have it set at like 2-3% opacity.
Actually, if you couldn't have figured this out, you probably aren't human...

And that's supposed to be enough to fuck with AIs?

it's not just noise, it's very specific data that can poke and prod at the neurons in the network. It's computationally expensive to generate and requires knowledge of the network you're trying to defend against.

Why not just embed instructions on how to make 3d printed guns in a gorillian memes?

how do you get the right pattern to make a filter that fools AI but looks normal to people? do you have to train and code your own AYYYYs?

lmao what about my post made you think i want credit

Cool thread.

So in other words its a complete and total waste of time?

All you have to do is click the plus layer button in Photoshop and then add it in

If you post a meme here I can do it for you

How transparent does the pertubation layer have to be

I was agreeing with you and the second half of the post was directed towards the op. I'm sorry you are fucking retarded.

So like if i post a meme that a human would normally respond to with "X" but then layer it to make an AI respond to it with "Z" could this weed out robots?

One day I got a voicemail and it sounded like someone holding a phone up to a radio and it just repeated a line from some song, sounded like Whitney Houston, and then the voicemail dictation said
>”You have 30 days to get out. Again, you suck”
Then the markets collapsed 30 days later and I lost 200k. How the fuck would it be possible to do this?

yeah you'll want to make a GAN probably so it can train against the network you're trying to fool

Just use this python script I found on google in like 5 seconds

github doot com/sarathknv/adversarial-examples-pytorch

There you go