Nonlinear optical encoding enabled by recurrent linear scattering

Area of Science:

• Optics
• Information Processing
• Computational Science

Background:

• Nonlinearity is crucial for optical computation but remains a significant challenge.
• Existing methods often require high power or complex setups.

Purpose of the Study:

• To introduce a novel design for passive optical nonlinearity using a multiple-scattering cavity.
• To demonstrate optical data compression and its application in information processing tasks.

Main Methods:

• Utilizing a multiple-scattering cavity with a continuous-wave laser at low power.
• Leveraging nonlinear random mapping induced by scattering events.
• Implementing optical data compression with a digital decoder.

Main Results:

• Successfully induced passive optical nonlinear random mapping.
• Demonstrated information retention during dimensionality reduction for data compression.
• Achieved high performance in classification, image reconstruction, keypoint detection, and object detection.
• Showcased real-time pedestrian detection at extreme compression ratios.

Conclusions:

• The multiple-scattering cavity design offers a low-power, efficient solution for optical nonlinearity.
• Optical data compression via this method enables powerful optical information processing.
• This work opens new avenues for optical computing algorithms and architectures.