Photonic Stochastic Emergent Storage for deep classification by scattering-intrinsic patterns

Area of Science:

• Photonics
• Information Storage
• Computational Physics

Background:

• Natural systems exhibit pervasive disorder, offering complex, non-repeating patterns.
• Existing memory devices often rely on ordered structures, limiting storage capacity and speed.
• Emergent archetypes offer a paradigm for pattern instantiation from imperfect examples.

Purpose of the Study:

• To present a novel storage method, Stochastic Emergent Storage (SES), utilizing natural disorder.
• To demonstrate SES in the photonic domain using random transmission matrices.
• To achieve high-capacity, high-speed arbitrary pattern storage and classification.

Main Methods:

• Employed a Hopfield-like network and emergent processes to instantiate archetypes.
• Utilized random transmission matrices from light scattering in a turbid medium as prototypes.
• Implemented programmable hardware for experimental validation of the SES method.

Main Results:

• Successfully stored arbitrary archetypes and performed classification at the speed of light.
• Achieved simultaneous storage of thousands of memories without additional fabrication.
• Demonstrated collective assessment of stored memories against a given pattern (content addressable memory).

Conclusions:

• Stochastic Emergent Storage (SES) offers a powerful new approach for optical information processing.
• The method leverages mesoscopic diffusion for massive memory capacity.
• Spatial organization of memories enables hierarchical optical classification layers.