• Technical Conference: 

    15 – 20 May 2022

  • Exhibition: 

    17 – 19 May 2022

Home > Schedule > Post Session

Post Session


AW3E

A&TTR on Integrated Photonics in Neural Networks II

Presider: Yasha Yi, University of Michigan

Authentication Required

Access to this content is only available to registered attendees. Registered attendees, please login now to access the content.

Presentations

Optical Processing for Artificial Neural Vision (AW3E.1)
Presenter: David Brady, Duke University

Convolutional neural networks have become established as the primary mechanisms for image processing over the past decade. While general purpose optical neural networks remain a long term project, in the near term optical prefilters act as the first layers of electronic deep convolutional networks and enable 10-100x reduction in system power per reconstructed voxel.

Authors:David Brady, Duke University

  Paper

All-Optical Neural Network With Programmable Linear Transformation (AW3E.2)
Presenter: Xue Feng, Tsinghua University

In our previous works, programmable arbitrary linear optical operations have been demonstrated on discrete phase-coherent spatial modes. Thus, in this work, we proposed and demonstrated a programmable ONN scheme for various image identification tasks.

Authors:Yidong Huang, Tsinghua University / Xue Feng, Tsinghua University

Single-Pixel Image Classification via Nonlinear Optics and Deep Neural Network (AW3E.3)
Presenter: Santosh Kumar, Stevens Institute of Technology

We propose and experimentally demonstrate a hybrid system which utilizes a nonlinear mode-selective optical method to extract the features with single-pixel detection and subsequently recognize the high-resolution images from a deep neural network
<gdiv></gdiv>

Authors:Santosh Kumar, Stevens Institute of Technology / Ting Bu, Stevens Institute of Technology / He Zhang, Stevens Institute of Technology / Irwin Huang, Stevens Institute of Technology / Yuping Huang, Stevens Institute of Technology

  Paper

A Codesigned Photonic Electronic MAC Neuron With ADC-Embedded Nonlinearity (AW3E.4)
Presenter: Lorenzo De Marinis, Scuola Superiore Sant'Anna

We present a reduced-precision integrated photonic electronic multiply-accumulate (MAC) neuron with ADC-embedded nonlinearity. The proposed device trades off speed with resolution, outperforming both analog and digital electronic solutions in terms of speed and energy consumption

Authors:Lorenzo De Marinis, Scuola Superiore Sant'Anna / Alessandro Catania, University of Pisa / Piero Castoldi, Scuola Superiore Sant'Anna / Paolo Bruschi, University of Pisa / Massimo Piotto, University of Pisa / Nicola Andriolli, National Research Council of Italy

  Paper

Massively-Parallel Amplitude-Only Fourier Optical Convolutional Neural Network (AW3E.5)
Presenter: Volker Sorger, George Washington University

Here we introduce a novel amplitude-only Fourier-optical processor paradigm and demonstrate a prototype system capable of processing large-scale ~(2,000x1,000) matrices in a single time-step and 100 microsecond-short latency, for accelerating machine-learning applications.

Authors:Mario Miscuglio, George Washington University / Zibo Hu, George Washington University / Shurui Li, UCLA / Jonathan George, George Washington University / Roberto Capanna, George Washington University / Hamed Dalir, Optelligence LLC / Philippe Bardet, George Washington University / Puneet Gupta, UCLA / Volker Sorger, George Washington University

  Paper

Conditional Machine Learning-Based Inverse Design Across Multiple Classes of Photonic Metasurfaces (AW3E.6)
Presenter: Christopher Yeung, University of California, Los Angeles

We present a machine learning-based photonics design strategy centered on encoding image colors with material and structural data. Given input target spectra, our model can accurately determine the optimal metasurface class, materials, and structure.

Authors:Christopher Yeung, University of California, Los Angeles / Ryan Tsai, University of California, Los Angeles / Benjamin Pham, University of California, Los Angeles / Yusaku Kawagoe, University of California, Los Angeles / David Ho, University of California, Los Angeles / Julia Liang, University of California, Los Angeles / Mark Knight, Northrop Grumman Corporation / Aaswath Raman, University of California, Los Angeles

  Paper