Article
Multidisciplinary Sciences
Tao Yan, Rui Yang, Ziyang Zheng, Xing Lin, Hongkai Xiong, Qionghai Dai
Summary: Photonic neural networks use photons instead of electrons to perform brain-like computations, leading to significantly improved computing performance. However, current architectures are limited to handling data with regular structures and cannot generalize to graph-structured data beyond Euclidean space. In this study, a diffractive graph neural network (DGNN) is proposed to address this limitation by utilizing diffractive photonic computing units (DPUs) and on-chip optical devices. DGNN achieves complex feature representation by capturing dependencies among node neighborhoods during light-speed optical message passing over graph structures. It demonstrates superior performance in node and graph-level classification tasks with benchmark databases, providing a new direction for high-efficiency processing of large-scale graph data structures using deep learning.
Article
Optics
Md Sadman Sakib Rahman, Jingxi Li, Deniz Mengu, Yair Rivenson, Aydogan Ozcan
Summary: Researchers in the USA have made significant improvements in the performance of diffractive optical networks, signaling a major advancement in their use for optics-based computation and machine learning. By utilizing feature engineering and ensemble learning, they were able to substantially enhance the statistical inference capabilities of these networks.
LIGHT-SCIENCE & APPLICATIONS
(2021)
Article
Engineering, Multidisciplinary
Hang Chen, Jianan Feng, Minwei Jiang, Yiqun Wang, Jie Lin, Jiubin Tan, Peng Jin
Summary: This study extends the diffractive deep neural network (DNN)-N-2 to visible wavelengths and proposes a general theory to solve contradictions between wavelength, neuron size, and fabrication limitations. The novel visible light (DNN)-N-2 classifier successfully recognizes handwritten digits and altered targets.
Article
Nanoscience & Nanotechnology
Deniz Mengu, Yifan Zhao, Anika Tabassum, Mona Jarrahi, Aydogan Ozcan
Summary: This study presents the use of diffractive optical networks to perform permutation operations with a large number of input-output connections. By using deep learning techniques, the capacity of the diffractive optical network to approximate permutation operations increases with the number of diffractive layers and trainable transmission elements in the system. The authors also addressed challenges related to physical alignment and output efficiency by designing misalignment tolerant diffractive designs. The proposed diffractive permutation networks have potential applications in security, image encryption, and data processing, and can serve as channel routing and interconnection panels in wireless networks.
Review
Optics
Yichen Sun, Mingli Dong, Mingxin Yu, Xiaolin Liu, Lianqing Zhu
Summary: The UCLA research group developed the world's first all-optical diffraction deep neural network system, which can perform classification tasks at near-light-speed. They used a terahertz light source as the input, optimized the model parameters, and utilized 3D printing technology to construct the system. This research opens up new possibilities for optical neural networks.
JOURNAL OF THE OPTICAL SOCIETY OF AMERICA B-OPTICAL PHYSICS
(2023)
Article
Nanoscience & Nanotechnology
Zhengyang Duan, Hang Chen, Xing Lin
Summary: Photonic neural networks utilize photons instead of electrons to perform AI tasks, but current architectures are not capable of multiplexing different tasks in parallel. This paper proposes an optical multitask learning system using multiwavelength diffractive deep neural networks (D(2)NNs) with joint optimization. By encoding multitask inputs into multiwavelength channels, the system achieves high accuracy in parallel multitask performance. Numerical evaluations show that multiwavelength D(2)NNs achieve significantly higher classification accuracies for multitask learning compared to single-wavelength D(2)NNs. This work opens new possibilities for high-throughput neuromorphic photonic computing and parallel execution of multiple tasks in AI systems.
Article
Automation & Control Systems
Md Sadman Sakib Rahman, Aydogan Ozcan
Summary: This study demonstrates for the first time a time-lapse image classification scheme using a diffractive network, improving classification accuracy and generalization performance by leveraging the lateral movements of the input objects and/or the diffractive network. Numerical exploration reveals a blind testing accuracy of 62.03% on the optical classification of objects from the CIFAR-10 dataset using time-lapse diffractive networks, achieving the highest inference accuracy so far. Time-lapse diffractive networks will be widely beneficial for spatiotemporal analysis of input signals using all-optical processors.
ADVANCED INTELLIGENT SYSTEMS
(2023)
Article
Optics
Dongyu Yang, Junhao Zhang, Ye Tao, Wenjin Lv, Shun Lu, Hao Chen, Wenhui Xu, Yishi Shi
Summary: Deep CDI is a physics-driven untrained learning method that can reconstruct complex scenes from a single diffraction pattern with high confidence and fast reconstruction of dynamic processes.
Article
Multidisciplinary Sciences
Tingzhao Fu, Yubin Zang, Yuyao Huang, Zhenmin Du, Honghao Huang, Chengyang Hu, Minghua Chen, Sigang Yang, Hongwei Chen
Summary: Researchers have developed an on-chip diffractive optical neural network (DONN) using 1D dielectric metasurfaces, achieving 90% classification accuracy, computing at 10^16 flops/mm^2, and consuming 10E-17 J/Flop. The proposed DONN is based on a silicon-on-insulator platform and offers high integration and low power consumption for machine learning tasks.
NATURE COMMUNICATIONS
(2023)
Article
Nanoscience & Nanotechnology
Deniz Mengu, Yair Rivenson, Aydogan Ozcan
Summary: The recent research efforts in optical computing have shifted towards developing optical neural networks that take advantage of the processing speed and parallelism of optics/photonics in machine learning applications. Diffractive Deep Neural Networks (D(2)NNs) utilize light-matter interaction over trainable surfaces, designed using deep learning, to perform statistical inference tasks as light waves propagate through them. The new training strategy introduces input object transformations as uniformly distributed random variables to improve the network's resilience against such transformations, leading to scale-, shift-, and rotation-invariant designs that are crucial for dynamic machine vision applications.
Article
Optics
Tiankuang Zhou, Xing Lin, Jiamin Wu, Yitong Chen, Hao Xie, Yipeng Li, Jintao Fan, Huaqiang Wu, Lu Fang, Qionghai Dai
Summary: Researchers propose a reconfigurable diffractive processing unit based on the diffraction of light, which supports different neural networks and achieves high model complexity with millions of neurons. By developing an adaptive training approach to overcome system errors, they achieve excellent experimental accuracies for high-speed image and video recognition.
Article
Computer Science, Artificial Intelligence
Lu Shi, ChangYuan Wang, Feng Tian, HongBo Jia
Summary: The paper introduces an integrated pupil tracking framework LVCF based on deep learning, consisting of VCF and LSTM, which was trained and evaluated on multiple datasets and outperformed the state of the art technologies.
Article
Optics
Tingzhao Fu, Yuyao Huang, Run Sun, Honghao Huang, Wencan Liu, Sigang Yang, Hongwei Chen
Summary: Integrated diffractive optical neural networks (DONNs) have the potential for high-speed and ultra-low energy consumption complex machine learning tasks. However, their on-chip implementation is limited by input dimensions. This article proposes a space-time interleaving technology based on arrayed waveguides to realize an on-chip DONN with high-speed, high-dimensional, and all-optical input signal modulation. Experimental results demonstrate the feasibility of this method.
CHINESE OPTICS LETTERS
(2023)
Article
Computer Science, Artificial Intelligence
Junfeng Zhou, Baigang Huang, Wenjiao Fan, Ziqian Cheng, Zhuoyi Zhao, Weifeng Zhang
Summary: This paper proposes a new text-based person search model called CM-LRGNet, which achieves fine-grained alignment of visual and linguistic modal data. By extracting Cross-Modal Local-Relational-Global features in an end-to-end manner and performing cross-modal alignment on different feature levels, it achieves state-of-the-art performance on the CUHK-PEDES benchmark dataset.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Shiv Ram Dubey, Satish Kumar Singh, Bidyut Baran Chaudhuri
Summary: Neural networks have experienced significant growth in recent years and activation functions play a crucial role. This paper provides a comprehensive overview and survey of activation functions in deep learning, covering different types, characteristics, and performance comparison of these functions. The insights gained from these studies benefit further research and practical decision-making.
Article
Engineering, Electrical & Electronic
Christoph Prokop, Steffen Schoenhardt, Bert Laegel, Sandra Wolff, Arnan Mitchell, Christian Karnutsch
JOURNAL OF LIGHTWAVE TECHNOLOGY
(2016)
Article
Engineering, Electrical & Electronic
Christoph Prokop, Steffen Schoenhardt, Tanveer Mahmud, Arnan Mitchell, Christian Karnutsch
JOURNAL OF MICROMECHANICS AND MICROENGINEERING
(2016)
Article
Optics
Thach Giang Nguyen, Guanghui Ren, Steffen Schoenhardt, Markus Knoerzer, Andreas Boes, Arnan Mitchell
LASER & PHOTONICS REVIEWS
(2019)
Article
Optics
Elena Goi, Xi Chen, Qiming Zhang, Benjamin P. Cumming, Steffen Schoenhardt, Haitao Luan, Min Gu
Summary: Optical machine learning utilizes the advantages of optical signals to process data at the speed of light. In this study, optical devices in the form of single-layer nanoscale holographic perceptrons were presented for optical inference tasks. The combination of machine learning and on-chip integration in these devices shows promise for transformative impacts in optical decryption, sensing, medical diagnostics, and computing.
LIGHT-SCIENCE & APPLICATIONS
(2021)
Article
Optics
Steffen Schoenhardt, Andreas Boes, Thach G. Nguyen, Arnan Mitchell
Summary: Investigated the influence of excitation beam width and optical losses on the spectral response of ridge resonators, finding that the space required for the excitation beam is the limiting factor on the highest achievable Q-factor.
Article
Optics
Steffen Schoenhardt, Andreas Boes, Thach G. Nguyen, Arnan Mitchell
Summary: Integrated photonic resonators based on bound states in the continuum (BICs) on the silicon-on-insulator (SOI) platform have the potential for novel, mass-manufacturable resonant devices. To overcome the size limitation of traditional resonator structures, this study investigates the translation of BIC-based ridge resonators into a guided mode system with finite lateral dimensions. Numerical simulations demonstrate that, similar to BIC-based resonators, such a waveguide system can exhibit spectrally narrow-band inversion of its transmissive behavior.
Proceedings Paper
Optics
Elena Goi, Mengxiang Chen, Steffen Schoenhardt, Min Gu
Summary: Optical machine learning is an important research field that utilizes the advantages of optical signals to enable information processing at the speed of light. Diffractive neural networks, fabricated using nanoprinting methods, have the potential to revolutionize adaptive optics and data processing. However, fabrication errors limit the performance of these networks. In this study, the performance of optical implementations of neural networks for complex data processing is investigated, and the impact of fabrication errors on the system performance is analyzed.
HOLOGRAPHY, DIFFRACTIVE OPTICS, AND APPLICATIONS XII
(2022)
Proceedings Paper
Engineering, Electrical & Electronic
Elena Goi, Steffen Schoenhardt, Min Gu
Summary: In this study, aberration detectors based on multi-layered perceptrons printed by two-photon nanolithography are presented. By utilizing all-optical inference, these perceptrons are able to collect phase information from a point spread function, enabling direct aberration detection in a single step.
2021 CONFERENCE ON LASERS AND ELECTRO-OPTICS (CLEO)
(2021)
Proceedings Paper
Engineering, Electrical & Electronic
Thach G. Nguyen, Guanghui Ren, Steffen Schoenhardt, Markus Knoerzer, Andreas Boes, Arnan Mitchell
2019 CONFERENCE ON LASERS AND ELECTRO-OPTICS (CLEO)
(2019)
Article
Optics
Guanghui Ren, Didit Yudistira, Thach G. Nguyen, Iryna Khodasevych, Steffen Schoenhardt, Kyle J. Berean, Joachim M. Hamm, Ortwin Hess, Arnan Mitchell
Proceedings Paper
Engineering, Electrical & Electronic
Guanghui Ren, Thach Nguyen, Iryna Khodasevych, Steffen Schoenhardt, Kyle J. Berean, Paul M. Z. Davies, Joachim M. Hamm, Ortwin Hess, Arnan Mitchell
2016 CONFERENCE ON LASERS AND ELECTRO-OPTICS (CLEO)
(2016)