효율적인 신경망 기반 암호키 교환 기술 = Practically Secure Key Exchange Scheme based on Neural Network

다국어 초록 (Multilingual Abstract)

Key exchange is one of the major aspects in cryptography. Recently, compared to the existing key exchange schemes, more efficient key exchange schemes have been proposed based on neural network learning. After the first key exchange scheme based on neural network was proposed, various attack models have been suggested in security analysis. Hebbian learning rule is vulnerable to majority attack which is the most powerful attack. Anti Hebbian learning rule is secure against majority attack has a limitation in efficiency, so we can only use key exchange scheme based on random walk learning rule which is more secure and efficient than the others. However, if we use random walk learning rule, the efficiency which is advantage about neural cryptography is reduced than the other learning rules. In this paper we analyze random walk and neural cryptography, and we propose new learning rule which is more efficient than existing random walk learning rule. Also, we theoretically analyze about key exchange scheme which is uses new learning rule and verify the efficiency and security by implementing majority attack model.

국문 초록 (Abstract)

키교환 문제는 암호학에서 중요한 관심사 중 하나이다. 최근 신경망 학습을 기반으로, 기존의 키 교환 방식보다 효율적인 키교환 기법들이 제안되었다. 최초의 신경망 기반 키교환 기법이 제안된 이후 많은 안전성 분석과 공격 기법들이 연구되었다. 공격들 중 가장 강력한 다수 공격(Majority attack)에 대해 기존에 제안된 헤비안 학습(Hebbian learning)은 취약점이 존재한다. 다수 공격에 안전한 안티 헤비안 학습(Anti Hebbian learning)은 효율성에 한계가 존재하며, 결론적으로 랜덤 워크 학습(Random walk learning)에 기반한 신경망 암호만이 안전하고 효율적인 방법으로 우리가 사용할 수 있음이 보여졌다. 하지만 랜덤 워크 학습을 사용하면 신경망 암호의 장점인 효율성이 다른 학습을 사용하는 것보다 감소한다.
이에 본 논문에서는 기존의 랜덤 워크 학습과 이를 사용한 신경망 암호에 대해 분석하고, 이것을 바탕으로 기존의 랜덤 워크 학습보다 효율적인 새로운 방식의 학습을 제안한다. 또한, 새로운 학습을 사용한 키교환 기술에 대한 이론적 분석과 더불어 다수 공격을 직접 구현하여 제안 방식의 효율성과 안전성을 검증한다.

참고문헌 (Reference)

1 X. Lei, "Twolayer tree-connected feed-forward neural network model for neural cryptography" 87 (87): 2013

2 A. Ruttor, "Synchronization of random walks with reflecting boundaries" 37 (37): 2004

3 L. F. Seoane, "Successful attack on permutation-parity-machine-based neural cryptography" 85 (85): 2012

4 A. Engel, "Statistical Mechanics of Learning" Cambridge Univ. press 2001

5 I. Kanter, "Secure exchange of information by synchronization of neural networks" 57 (57): 2002

6 O. M. Reyes, "Permutation parity machines for neural synchronization" 42 (42): 195002-, 2009

7 A. Allam, "On the improvement of the neural cryptography using erroneous transmitted information with error prediction" 21 : 1915-1924, 2010

8 A. Ruttor, "Neural synchronization and cryptography" 2006

9 A. Ruttor, "Neural cryptography with queries" 2005 : P01009-, 2005

10 A. Ruttor, "Neural cryptography with feedback" 69 (69): 2004

11 A. Jensen, "Lecture Notes on Difference Equations" Aalborg University 2011

12 M. Abadi, "Learning to protect communications with adversarial neural cryptography"

13 A. Ruttor, "Genetic attack on neural cryptography" 73 (73): 2006

14 S. Pattanayad, "Encryption based on neural cryptography" 321-330, 2017

15 P. Xie, "Crypto-nets: Neural networks over encrypted data"

16 L. N. Shacham, "Cooperating attackers in neural cryptography" 69 (69): 2004

17 A. Klimov, "Analysis of neural cryptography" Springer Verlag 288-298, 2002

18 W. Feller, "An Introduction to Probability Theory and Its Applications" Wiley 1968

19 R. Mislovaty, "A Secure key-exchange protocol with an absence of injective functions" 66 (66): 2002