Parts | Days | Selection | Search | Updates | Downloads | Help
DY: Fachverband Dynamik und Statistische Physik
DY 24: Networks - Statistics and Dynamics (joint session DY/ BP/ SOE)
DY 24.2: Talk
Wednesday, April 2, 2014, 15:15–15:30, ZEU 118
Pattern-matching via a network of phase oscillators of different frequency: A novel Architecture — •Daniel Heger and Katharina Krischer — Technische Universität München, Physikdepartment
Oscillatory networks can in principle be used for pattern recognition. Nevertheless, current architectures either lack scalability towards large numbers of oscillators or need the external input of complex time-dependent coupling functions. In our talk, we will present a novel architecture for pattern matching with oscillatory neural networks. A system of oscillators of different frequencies and coupling functions is used whose dynamics average to the dynamics of an all-to-all coupled oscillator network. In contrast to previous approaches, the necessary coupling functions are automatically generated inside the network and the output pattern can easily be read out binary. By additionally choosing a new type of coupling function, the matching mechanism is stable even for high coupling strengths and the degenerate attractive limit set containing the memorized patterns transforms to a system of seperated attractors for each memorized pattern. Although the system's dynamics do not average to the dynamics of simple coupled Kuramoto oscillators, the appealing mathematical structure permits determination of the stability of all fixed points using nonlinear stability analysis and a dynamic equation solely in pattern space can be derived.