Neural cryptography with queries for co-operating attackers and effective number of keys

This work is a new proposal of neural synchronization, which is a communication of two Tree Parity Machines (TPMs) for agreement on a common secret key over a public channel. This can be achieved by two TPMs, which are trained on their mutual output, which can synchronize to a time dependent state of identical synaptic weight vectors. In the proposed TPMs random inputs are replaced with queries, which are considered. The queries depend on the current state of A and B TPMs. Then, TPM's hidden layers of each output vectors are compared. That is, the output vectors of hidden unit using Hebbian learning rule and dynamic unit using Random walk learning rule are compared. Among the compared values, the output layer receives one of the best values. In this paper, the increased synchronization time of the co-operating attacker against the flipping attack is also analyzed.