Neural synchronization and cryptography [Elektronische Ressource] / vorgelegt von Andreas Ruttor

Neural Synchronizationand CryptographyDissertation zur Erlangung desnaturwissenschaftlichen Doktorgradesder Bayerischen Julius-Maximilians-Universitat Wurzburgvorgelegt vonAndreas Ruttoraus WurzburgWurzburg 2006Eingereicht am 24.11.2006bei der Fakultat fur Physik und Astronomie1. Gutachter: Prof. Dr. W. Kinzel2.hter: Prof. Dr. F. Assaadder Dissertation.1. Prufer: Prof. Dr. W. Kinzel2. Prufer: Prof. Dr. F. Assaad3. Prufer: Prof. Dr. P. Jakobim PromotionskolloquiumTag des Promotionskolloquiums: 18.05.2007Doktorurkunde ausgehandigt am:AbstractNeural networks can synchronize by learning from each other. For that pur-pose they receive common inputs and exchange their outputs. Adjusting discreteweights according to a suitable learning rule then leads to full synchronization ina nite number of steps. It is also possible to train additional neural networks byusing the inputs and outputs generated during this process as examples. Severalalgorithms for both tasks are presented and analyzed.In the case of Tree Parity Machines the dynamics of both processes is drivenby attractive and repulsive stochastic forces. Thus it can be described well bymodels based on random walks, which represent either the weights themselves ororder parameters of their distribution. However, synchronization is much fasterthan learning.