I would like to know if it is possible to train a CNN to classify images (like nsfw detection) without receiving any intelligible information about it ?
By applying some kind of one-way mask before upload for example.
Edit: I found that it is possible to apply filters on encrypted images with homomorphic encryption, but I don't know if classification is possible.
The short answer is No.
The long answer is maybe, if you use Homomorphic encryption.
Homomorphic encryption is the conversion of data into ciphertext that can be analyzed and worked with as if it were still in its original form. Homomorphic encryption enables complex mathematical operations to be performed on encrypted data without compromising the encryption.
However, this is a very mathematically complex task, and the encryption is weak at best.
Homomorphic encryption schemes are inherently malleable. In terms of malleability, homomorphic encryption schemes have weaker security properties than non-homomorphic schemes.
Here's an image courtesy of Nvidia:
