Dataset Information

Meta-optic accelerators for object classifiers.

ABSTRACT: Rapid advances in deep learning have led to paradigm shifts in a number of fields, from medical image analysis to autonomous systems. These advances, however, have resulted in digital neural networks with large computational requirements, resulting in high energy consumption and limitations in real-time decision-making when computation resources are limited. Here, we demonstrate a meta-optic-based neural network accelerator that can off-load computationally expensive convolution operations into high-speed and low-power optics. In this architecture, metasurfaces enable both spatial multiplexing and additional information channels, such as polarization, in object classification. End-to-end design is used to co-optimize the optical and digital systems, resulting in a robust classifier that achieves 93.1% accurate classification of handwriting digits and 93.8% accuracy in classifying both the digit and its polarization state. This approach could enable compact, high-speed, and low-power image and information processing systems for a wide range of applications in machine vision and artificial intelligence.

SUBMITTER: Zheng H

PROVIDER: S-EPMC9328681 | biostudies-literature | 2022 Jul

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Meta-optic accelerators for object classifiers.

Zheng Hanyu H Liu Quan Q Zhou You Y Kravchenko Ivan I II Huo Yuankai Y Valentine Jason J

Science advances 20220727 30

Rapid advances in deep learning have led to paradigm shifts in a number of fields, from medical image analysis to autonomous systems. These advances, however, have resulted in digital neural networks with large computational requirements, resulting in high energy consumption and limitations in real-time decision-making when computation resources are limited. Here, we demonstrate a meta-optic-based neural network accelerator that can off-load computationally expensive convolution operations into ...[more]

PMID: 35895828

Similar Datasets

Project description:PurposeTo better meet clinical needs and facilitate optimal treatment planning, we added two new electron energy beams (7 and 11 MeV) to two Varian TrueBeam linacs.MethodsWe worked with the vendor to create two additional customized electron energies without hardware modifications. For each beam, we set the bending magnet current and then optimized other beam-specific parameters to achieve depths of 50% ionization (I50 ) of 2.9 cm for 7 MeV and 4.2 cm for the 11 MeV beam with the 15 × 15 cm2 cone at 100 cm source-to-surface distance (SSD) by using an ionization chamber profiler (ICP) with a double-wedge (DW) phantom. Beams were steered and balanced to optimize symmetry with the ICP. After all parameters were set, full commissioning was done including measuring beam profiles, percent depth doses (PDDs), output factors (OFs) at standard, and extended SSDs. Measured data were compared between the two linacs and against the values calculated by our RayStation treatment planning system (TPS) following Medical Physics Practice Guideline 5.a (MPPG 5.a) guidelines.ResultsThe I50 values initially determined with the ICP/DW agreed with those from a PDD-scanned in-water phantom within 0.2 mm for the 7 and 11 MeV on both linacs. Comparison of the beam characteristics from the two linacs indicated that flatness and symmetry agreed within 0.4%, and point-by-point differences in PDD were within 0.01% ± 0.3% for the 7 MeV and 0.01% ± 0.3% for the 11 MeV. The OF ratios between the two linacs were 1.000 ± 0.007 for the 7 MeV and 1.004 ± 0.007 for the 11 MeV. Agreement between TPS-calculated outputs and measurements were -0.1% ± 1.0% for the 7 MeV and 0.2% ± 0.8% for the 11 MeV. All other parameters met the MPPG 5.a's 3%/3-mm criteria.ConclusionWe were able to add two new beam energies with no hardware modifications. Tuning of the new beams was facilitated by the ICP/DW system allowing us to have the procedures done in a few hours and achieve highly consistent results across two linacs. PACS numbers: 87.55.Qr, 87.56.Fc.

Dataset Information

Meta-optic accelerators for object classifiers.

Publications

Meta-optic accelerators for object classifiers.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets