UCLA researchers have performed an in-depth analysis of nonlinear information encoding strategies for diffractive optical processors, providing new insights into their performance and utility. Their study, published in Light: Science & Applications, compared simpler-to-implement nonlinear coding strategies involving, e.g., phase coding, with the performance of repetition-based nonlinear information coding methods of data, shedding light on their advantages and limitations in the optical processing of visual information.
Diffractive optical processors, built using linear materials, perform computational tasks by manipulating light using structured surfaces. Nonlinear coding of optical information can improve the performance of these processors, enabling them to better handle complex tasks such as image classification, quantitative phase imaging, and encryption.
The UCLA research team, led by Professor Aydogan Ozcan, evaluated different nonlinear coding strategies using different datasets to evaluate their statistical inference performance. Their findings revealed that replicating data within a diffractive volume, while increasing inference accuracy, compromises the universal linear transformation capability of diffractive optical processors. As a result, diffractive blocks based on data iteration cannot serve as optical analogues for the fully connected or convolutional layers commonly used in digital neural networks. In general, diffractive processors based on data replication can be perceived as a simplified optical analog of the dynamic convolution kernel concept used in some neural network architectures. Despite the different features, the data replication architecture within a diffractive optical processor is still effective for inference tasks and offers advantages in terms of noise robustness.
Alternatively, phase coding of the input information, without data repetition, provides a simpler nonlinear coding strategy to implement with statistically comparable inference accuracy. Implemented through spatial light modulators or phase-only objects, directly, phase coding is a practical alternative due to its simplicity and effectiveness. Moreover, diffractive processors without data repetition do not need pre-processing of input information through a digital system, which is required for visual data repetition. Therefore, data replication can take time, especially for phase-only input objects, due to the need for digital phase recovery and preprocessing before visual data replication can occur.
The research team’s findings provide valuable insights into the push-pull relationship between material-based linear diffractive optical systems and nonlinear information encoding strategies. These results have potential for a wide range of applications, including optical communications, surveillance, and computational imaging. The ability to increase inference accuracy through nonlinear coding strategies can improve the performance of optical processors in various domains, leading to more advanced and efficient visual information processing systems.
Authors of this article include Yuhang Li, Jingxi Li, and Aydogan Ozcan, all affiliated with UCLA’s Department of Electrical and Computer Engineering. Professor Ozcan also serves as associate director of the California NanoSystems Institute (CNSI).
This research was supported by DOE (USA).
/Public Notice. This material from the original organization/author(s) may be current in nature and edited for clarity, style and length. Mirage.News does not take institutional positions or sides and all views, opinions and conclusions expressed herein are solely those of the author(s). Watch it in full here.