jm + cryptanalysis   2

Decoding the Enigma with Recurrent Neural Networks
I am blown away by this -- given that Recurrent Neural Networks are Turing-complete, they can actually automate cryptanalysis given sufficient resources, at least to the degree of simulating the internal workings of the Enigma algorithm given plaintext, ciphertext and key:
The model needed to be very large to capture all the Enigma’s transformations. I had success with a single-celled LSTM model with 3000 hidden units. Training involved about a million steps of batched gradient descent: after a few days on a k40 GPU, I was getting 96-97% accuracy!
machine-learning  deep-learning  rnns  enigma  crypto  cryptanalysis  turing  history  gpus  gradient-descent 
26 days ago by jm
How Advanced Is the NSA's Cryptanalysis — And Can We Resist It?
Bruce Schneier's suggestions:
Assuming the hypothetical NSA breakthroughs don’t totally break public-cryptography — and that’s a very reasonable assumption — it’s pretty easy to stay a few steps ahead of the NSA by using ever-longer keys. We’re already trying to phase out 1024-bit RSA keys in favor of 2048-bit keys. Perhaps we need to jump even further ahead and consider 3072-bit keys. And maybe we should be even more paranoid about elliptic curves and use key lengths above 500 bits.

One last blue-sky possibility: a quantum computer. Quantum computers are still toys in the academic world, but have the theoretical ability to quickly break common public-key algorithms — regardless of key length — and to effectively halve the key length of any symmetric algorithm. I think it extraordinarily unlikely that the NSA has built a quantum computer capable of performing the magnitude of calculation necessary to do this, but it’s possible. The defense is easy, if annoying: stick with symmetric cryptography based on shared secrets, and use 256-bit keys.
bruce-schneier  cryptography  wired  nsa  surveillance  snooping  gchq  cryptanalysis  crypto  future  key-lengths 
september 2013 by jm

Copy this bookmark: