3 ms·
> Apple genuinely believe this is a novel and unique method to catch CSAM without invading people's privacy What’s novel about this? The technique and issues w
by mirker 5y ago
> Apple genuinely believe this is a novel and unique method to catch CSAM without invading people's privacy
What’s novel about this? The technique and issues with it are fairly obvious to anyone with experience in computer vision. It seems not too different from research from 1993 https://proceedings.neurips.cc/paper/1993/file/288cc0ff022877bd3df94bc9360b9c5d-Paper.pdf https://proceedings.neurips.cc/paper/1993/file/288cc0ff02287...
The issues are also well known and encompass the whole subfield of adversarial examples.
- shuckles 5y agoPresumably the system is novel enough that a very similar proposal which lacked one important insight (using perceptual hashes to turn image similarity problems into set intersection problems) was accepted to USENIX this year: https://www.usenix.org/conference/usenixsecurity21/presentation/kulshrestha https://www.usenix.org/conference/usenixsecurity21/presentat...
- mirker 5y agoThe privacy aspect does seem novel, and that paper evaluates the various aspects of privacy and the quality of perceptual hashes. However, anything concerning privacy (of client or server data) is orthogonal to the vision methods underpinning the system, which are well understood and are the primary cause of the post. For that matter, I don’t see how the paper’s proofs of privacy and/or formalization of the problem help with hash collisions. And that particular paper doesn’t use deep learning methods which are probably easier to attack with off-the-shelf methods.