Ponente
Descripción
In recent years, neural networks have been extensively employed within our group to address challenges in adaptive optics, leading to several innovative approaches, including sensorless wavefront correction and the prediction of atmospheric turbulence. Despite their potential, a major limitation of these methods lies in the substantial computational demands associated with real-time implementation in large-aperture telescopes—demands that remain challenging even for modern GPU-based systems.
To overcome this constraint, we are currently investigating the use of Optical Neural Networks (ONNs), which offer the promise of ultra-fast computation by leveraging the intrinsic parallelism of light propagation. Specifically, we aim to implement ONNs through Spatial Light Modulators (SLMs), enabling the execution of neural network operations at photonic speeds, thereby significantly reducing latency and enhancing the feasibility of real-time adaptive optics correction.
