ترغب بنشر مسار تعليمي؟ اضغط هنا

Massively Parallel Amplitude-Only Fourier Neural Network

76   0   0.0 ( 0 )
 نشر من قبل Volker Sorger
 تاريخ النشر 2020
والبحث باللغة English




اسأل ChatGPT حول البحث

Machine-intelligence has become a driving factor in modern society. However, its demand outpaces the underlying electronic technology due to limitations given by fundamental physics such as capacitive charging of wires, but also by system architecture of storing and handling data, both driving recent trends towards processor heterogeneity. Here we introduce a novel amplitude-only Fourier-optical processor paradigm capable of processing large-scale ~(1,000 x 1,000) matrices in a single time-step and 100 microsecond-short latency. Conceptually, the information-flow direction is orthogonal to the two-dimensional programmable-network, which leverages 10^6-parallel channels of display technology, and enables a prototype demonstration performing convolutions as pixel-wise multiplications in the Fourier domain reaching peta operations per second throughputs. The required real-to-Fourier domain transformations are performed passively by optical lenses at zero-static power. We exemplary realize a convolutional neural network (CNN) performing classification tasks on 2-Megapixel large matrices at 10 kHz rates, which latency-outperforms current GPU and phase-based display technology by one and two orders of magnitude, respectively. Training this optical convolutional layer on image classification tasks and utilizing it in a hybrid optical-electronic CNN, shows classification accuracy of 98% (MNIST) and 54% (CIFAR-10). Interestingly, the amplitude-only CNN is inherently robust against coherence noise in contrast to phase-based paradigms and features an over 2 orders of magnitude lower delay than liquid crystal-based systems. Beyond contributing to novel accelerator technology, scientifically this amplitude-only massively-parallel optical compute-paradigm can be far-reaching as it de-validates the assumption that phase-information outweighs amplitude in optical processors for machine-intelligence.

قيم البحث

اقرأ أيضاً

For this work, we introduced the use of binary amplitude for our proposed complex amplitude encoding of a phase-only hologram. By principle, a complex amplitude in a hologram plane can be represented by the amplitude and its phase. However, a phase-o nly hologram contains only phase information of the complex amplitude, which results in degradation of reconstruction quality from the hologram. In our method, by approximating the amplitude in the hologram plane using a binary amplitude, we can finally record the complex amplitude of an original light in the phase-only hologram. We validated the effectiveness of our method with two examples, hologram reconstruction and generation of Hermite-Gaussian beams.
Light field microscopy (LFM) uses a microlens array (MLA) near the sensor plane of a microscope to achieve single-shot 3D imaging of a sample without any moving parts. Unfortunately, the 3D capability of LFM comes with a significant loss of lateral r esolution at the focal plane. Placing the MLA near the pupil plane of the microscope, instead of the image plane, can mitigate the artifacts and provide an efficient forward model, at the expense of field-of-view (FOV). Here, we demonstrate improved resolution across a large volume with Fourier DiffuserScope, which uses a diffuser in the pupil plane to encode 3D information, then computationally reconstructs the volume by solving a sparsity-constrained inverse problem. Our diffuser consists of randomly placed microlenses with varying focal lengths; the random positions provide a larger FOV compared to a conventional MLA, and the diverse focal lengths improve the axial depth range. To predict system performance based on diffuser parameters, we for the first time establish a theoretical framework and design guidelines, which are verified by numerical simulations, then build an experimental system that achieves $< 3$ um lateral and $4$ um axial resolution over a $1000 times 1000 times 280$ um$^3$ volume. Our diffuser design outperforms the MLA used in LFM, providing more uniform resolution over a larger volume, both laterally and axially.
The GooFit Framework is designed to perform maximum-likelihood fits for arbitrary functions on various parallel back ends, for example a GPU. We present an extension to GooFit which adds the functionality to perform time-dependent amplitude analyses of pseudoscalar mesons decaying into four pseudoscalar final states. Benchmarks of this functionality show a significant performance increase when utilizing a GPU compared to a CPU. Furthermore, this extension is employed to study the sensitivity on the $D^0 - bar{D}^0$ mixing parameters $x$ and $y$ in a time-dependent amplitude analysis of the decay $D^0 rightarrow K^+pi^-pi^+pi^-$. Studying a sample of 50 000 events and setting the central values to the world average of $x = (0.49 pm0.15) %$ and $y = (0.61 pm0.08) %$, the statistical sensitivities of $x$ and $y$ are determined to be $sigma(x) = 0.019 %$ and $sigma(y) = 0.019 %$.
Formal verification of neural networks (NNs) is a challenging and important problem. Existing efficient complete solvers typically require the branch-and-bound (BaB) process, which splits the problem domain into sub-domains and solves each sub-domain using faster but weaker incomplete verifiers, such as Linear Programming (LP) on linearly relaxed sub-domains. In this paper, we propose to use the backward mode linear relaxation based perturbation analysis (LiRPA) to replace LP during the BaB process, which can be efficiently implemented on the typical machine learning accelerators such as GPUs and TPUs. However, unlike LP, LiRPA when applied naively can produce much weaker bounds and even cannot check certain conflicts of sub-domains during splitting, making the entire procedure incomplete after BaB. To address these challenges, we apply a fast gradient based bound tightening procedure combined with batch splits and the design of minimal usage of LP bound procedure, enabling us to effectively use LiRPA on the accelerator hardware for the challenging complete NN verification problem and significantly outperform LP-based approaches. On a single GPU, we demonstrate an order of magnitude speedup compared to existing LP-based approaches.
Deep neural networks have emerged as effective tools for computational imaging including quantitative phase microscopy of transparent samples. To reconstruct phase from intensity, current approaches rely on supervised learning with training examples; consequently, their performance is sensitive to a match of training and imaging settings. Here we propose a new approach to phase microscopy by using an untrained deep neural network for measurement formation, encapsulating the image prior and imaging physics. Our approach does not require any training data and simultaneously reconstructs the sought phase and pupil-plane aberrations by fitting the weights of the network to the captured images. To demonstrate experimentally, we reconstruct quantitative phase from through-focus images blindly (i.e. no explicit knowledge of the aberrations).
التعليقات
جاري جلب التعليقات جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
mircosoft-partner

هل ترغب بارسال اشعارات عن اخر التحديثات في شمرا-اكاديميا