4 ms·
What is noise anyway? I can program a generative model and inverse generative model couple and then you can transmit data as a image that doesn't look "noisy" b
by darkmighty 11y ago
What is noise anyway? I can program a generative model and inverse generative model couple and then you can transmit data as a image that doesn't look "noisy" but has incompressible information (what I referred to as noise), which could be a cryptogram. The model might generate anything, a innocuous conversation, a picture of a house, whatever. For good models the "inquisition" cannot distinguish you from your neighbor in terms of suspicious activity.
- DanBC 11y agoThis is just RubberHose, with the same flaws. (Increased bandwidth, and "they" don't care if it's noise or cipher text, they'll happily beat you until you give them what they want.) https://web.archive.org/web/20121029045140/http://marutukku.org/current/src/doc/maruguide/t1.html https://web.archive.org/web/20121029045140/http://marutukku.... https://en.wikipedia.org/wiki/Rubberhose_%28file_system%29 https://en.wikipedia.org/wiki/Rubberhose_%28file_system%29 https://en.wikipedia.org/wiki/Rubber-hose_cryptanalysis https://en.wikipedia.org/wiki/Rubber-hose_cryptanalysis
- darkmighty 11y agoNo, you're misunderstanding my proposal (w.r.t. Generative Models). This is different from deniable encryption (where an eavesdropper will detect you may be encrypting something), this is traffic literally indistinguishable from normal traffic. It only fails if the government knows a priori you are a criminal and inspects your hardware. I thought about making a few proof of concept models for this, but I don't think there's real demand for it (and I'm lazy) -- steganography is probably sufficient for what most people use.