Neural Cryptography

1
Neural Cryptography
Group A4 Chris Sun Bryan Maden Min Fang
2
Problem of Symmetric Key
Key distribution
3
Key and security
The more we use a same key
the more easily to beak the code
4
One-time pad
Unbreakable
5
Key 1
Plain 1
Plain X
Key X
Key 2
Plain 2
Cipher
Key 3
Plain 3
6
Problem of key distribution
No secure channel
Contact in person
Too expensive
7
NN to implement one-time pad
If this is true we have a perfect code
8
Basic NN
Input sample
Hidden layer
Output
9
Weight
Activation function
Propagation rule
10
Training
Input sample
Hidden layer
Output
Modify weights
compare
11
Public random bit stream
Output 1
NN 1
NN 2
Output 2
12
NN Cryptography

• All input and output are known to the public.
• All weights are unknown to the public, even the
  partner.

13
NN Cryptography -2

• Each NN trains by the output from his partner.
• After synchronization, the final weights is the
  key.

14
Example of NN used in cryptography
15
Question?
16
The End