Projektdaten

---

Hybride photonische Rechnerarchitekturen in (rück-) gekoppelten nicht-linearen Systemen mit Speicher

---

The project aims at realizing non-linear optical networks with reconfigurable topology, enabled by combining feedback-coupled optical amplifiers with coherent optical memories. The potential of these systems for neuro-inspired information processing in the reservoir computing approach is explored. The goal is to realize a novel neuromorphic computation scheme based on an extension of Reservoir Computing to solve problems that are hard on current digital von-Neumann computers. Examples of investigated problems are sequence prediction, pattern recognition and classification of irregular time series. The two main components of this new scheme are an optical non-linearity and coherent optical memories. The investigation will benefit greatly from close collaboration between theoretical simulations and experiment. The versatile coherent optical multi-cell memory with random access is implemented in warm Cesium vapor. Together with a pre-defined non-trivial read-out order of the memory cells, this allows for constructing a highly connected optical neural network, which greatly improves upon the state-of-the art in delay-based Reservoir Computing. This in turn also requires extensive numerical simulations to identify critical parameters and allow for efficient use of the experiment time. Our project introduces for the first time the concept of hybrid neural networks in optical hardware, which combines neurons for information processing with a random access memory to store the machines state. In particular, the adaptive nature of the memory allows for an on-the­ fly evolution of the network topology, which allows a greater refinement of the computational properties. Our project will benchmark and quantify the properties of this novel analog, brain-inspired optical computing mechanism. In our project two groups join their complementary expertise which is experimental quantum optics group (Wolters) and theory of optical reservoir computing (Lüdge).