direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments


Probabilistic Attack on Neural Cryptography
Zitatschlüssel Seoane:2012:PAN
Autor Luís Francisco Seoane Iglesias
Jahr 2012
Schule TU Berlin
Zusammenfassung The present Master Thesis deals with the implementation of a successful probabilistic attack to crack a Neural Cryptographic system. Neural Cryptography has been developed during the last decade with great success. Two systems implemented on artificial neural network have shown to be robust enough as key-exchange protocols when tested against different attacks. Existing attacks usually attempt to break the cryptographic protocols by mimicking it and show low performance. A probabilistic attack attempts to track the probability that the key-exchange protocol is converging towards each possible secret key and is radically different from those attacks tried before on Neural Cryptography. This attack was suggested before, as acknowledged in the body of the present work, but here it is presented the first implementation – to the best of our knowledge, including exhaustive literature – of a probabilistic attack on the two most researched Neural Cryptographic systems. In one of these systems – namely, the so-called Permutation Parity Machine (PPM) – the described algorithm shows an outstanding performance and allows us to conclude that the system is not safe enough for any cryptographic means. This result was published as a paper in the scientific journal Physical Review E and constitutes the core of the present Master Thesis. The results obtained so far for the probabilistic attack on the other relevant Neural Cryptographic system – the Tree Parity Machine (TPM) – are only of a speculative nature and do not allow to asses the security of this system under such an attack. These scarce results are included as a chapter of the present work with the hope that they can help to further improve any candidate probabilistic attack on TPM-based cryptography in future implementations.
Typ der Publikation Master Thesis
Link zur Publikation Download Bibtex Eintrag

Zusatzinformationen / Extras


Schnellnavigation zur Seite über Nummerneingabe