arXiv subjects

Jin Wei

Publications and source records attributed to Jin Wei.

28 records · Page 2Linked to original sources

arXiv2024

RESURF Ga$_{2}$O$_{3}$-on-SiC Field Effect Transistors for Enhanced Breakdown Voltage

Heterosubstrates have been extensively studied as a method to improve the heat dissipation of Ga$_{2}$O$_{3}$ devices. In this simulation work, we propose a novel role for $p$-type available heterosubstrates, as a component of a reduced surface field (RESURF) structure in Ga$_{2}$O$_{3}$ lateral field-effect transistors (FETs). The RESURF structure can eliminate the electric field crowding and contribute to higher breakdown voltage. Using SiC as an example, the designing strategy for doping concentration and dimensions of the $p$-type region is systematically studied using TCAD modeling. To mimic realistic devices, the impacts of interface charge and binding interlayer at the Ga$_{2}$O$_{3}$/SiC interface are also explored. Additionally, the feasibility of the RESURF structure for high-frequency switching operation is supported by the short time constant ($\sim$0.5 ns) of charging/discharging the $p$-SiC depletion region. This study demonstrates the great potential of utilizing the electrical properties of heat-dissipating heterosubstrates to achieve a uniform electric field distribution in Ga$_{2}$O$_{3}$ FETs.

physics.app-ph↗

arXiv2023

Masked and Permuted Implicit Context Learning for Scene Text Recognition

Scene Text Recognition (STR) is difficult because of the variations in text styles, shapes, and backgrounds. Though the integration of linguistic information enhances models' performance, existing methods based on either permuted language modeling (PLM) or masked language modeling (MLM) have their pitfalls. PLM's autoregressive decoding lacks foresight into subsequent characters, while MLM overlooks inter-character dependencies. Addressing these problems, we propose a masked and permuted implicit context learning network for STR, which unifies PLM and MLM within a single decoder, inheriting the advantages of both approaches. We utilize the training procedure of PLM, and to integrate MLM, we incorporate word length information into the decoding process and replace the undetermined characters with mask tokens. Besides, perturbation training is employed to train a more robust model against potential length prediction errors. Our empirical evaluations demonstrate the performance of our model. It not only achieves superior performance on the common benchmarks but also achieves a substantial improvement of $9.1\%$ on the more challenging Union14M-Benchmark.

cs.CV↗

arXiv2023

Demonstration of a photonic lantern focal-plane wavefront sensor: measurement of atmospheric wavefront error modes and low wind effect in the non-linear regime

Here we present a laboratory analysis of the use of a 19-core photonic lantern (PL) in combination with neural network (NN) algorithms as an efficient focal plane wavefront sensor (FP-WFS) for adaptive optics (AO), measuring wavefront errors such as low wind effect (LWE), Zernike modes and Kolmogorov phase maps. The aberrated wavefronts were experimentally simulated using a Spatial Light Modulator (SLM) with combinations of different phase maps in both the linear regime (average incident RMS wavefront error (WFE) of 0.88 rad) and in the non-linear regime (average incident RMS WFE of 1.5 rad). Results were analysed using a NN to determine the transfer function of the relationship between the incident wavefront error (WFE) at the input modes at the multimode input of the PL and the intensity distribution output at the multicore fibre outputs end of the PL. The root mean square error (RMSE) of the reconstruction of petal and LWE modes were just $2.87\times10^{-2}$ rad and $2.07\times10^{-1}$ rad respectively, in the non-linear regime. The reconstruction RMSE for Zernike combinations ranged from $5.67\times10^{-2}$ rad to $8.43\times10^{-1}$ rad, depending on the number of Zernike terms and incident RMS WFE employed. These results demonstrate the promising potential of PLs as an innovative FP-WFS in conjunction with NNs.

physics.optics↗

arXiv2021

PIMNet: A Parallel, Iterative and Mimicking Network for Scene Text Recognition

Nowadays, scene text recognition has attracted more and more attention due to its various applications. Most state-of-the-art methods adopt an encoder-decoder framework with attention mechanism, which generates text autoregressively from left to right. Despite the convincing performance, the speed is limited because of the one-by-one decoding strategy. As opposed to autoregressive models, non-autoregressive models predict the results in parallel with a much shorter inference time, but the accuracy falls behind the autoregressive counterpart considerably. In this paper, we propose a Parallel, Iterative and Mimicking Network (PIMNet) to balance accuracy and efficiency. Specifically, PIMNet adopts a parallel attention mechanism to predict the text faster and an iterative generation mechanism to make the predictions more accurate. In each iteration, the context information is fully explored. To improve learning of the hidden layer, we exploit the mimicking learning in the training phase, where an additional autoregressive decoder is adopted and the parallel decoder mimics the autoregressive decoder with fitting outputs of the hidden layer. With the shared backbone between the two decoders, the proposed PIMNet can be trained end-to-end without pre-training. During inference, the branch of the autoregressive decoder is removed for a faster speed. Extensive experiments on public benchmarks demonstrate the effectiveness and efficiency of PIMNet. Our code will be available at https://github.com/Pay20Y/PIMNet.

cs.CV↗

arXiv2021

Learning the Lantern: Neural network applications to broadband photonic lantern modelling

Photonic lanterns allow the decomposition of highly multimodal light into a simplified modal basis such as single-moded and/or few-moded. They are increasingly finding uses in astronomy, optics and telecommunications. Calculating propagation through a photonic lantern using traditional algorithms takes $\sim 1$ hour per simulation on a modern CPU. This paper demonstrates that neural networks can bridge the disparate opto-electronic systems, and when trained can achieve a speed-up of over 5 orders of magnitude. We show that this approach can be used to model photonic lanterns with manufacturing defects as well as successfully generalising to polychromatic data. We demonstrate two uses of these neural network models, propagating seeing through the photonic lantern as well as performing global optimisation for purposes such as photonic lantern funnels and photonic lantern nullers.

physics.optics↗

arXiv2020

An all-photonic focal-plane wavefront sensor

Adaptive optics (AO) is critical in astronomy, optical communications and remote sensing to deal with the rapid blurring caused by the Earth's turbulent atmosphere. But current AO systems are limited by their wavefront sensors, which need to be in an optical plane non-common to the science image and are insensitive to certain wavefront-error modes. Here we present a wavefront sensor based on a photonic lantern fibre-mode-converter and deep learning, which can be placed at the same focal plane as the science image, and is optimal for single-mode fibre injection. By measuring the intensities of an array of single-mode outputs, both phase and amplitude information on the incident wavefront can be reconstructed. We demonstrate the concept with simulations and an experimental realisation wherein Zernike wavefront errors are recovered from focal-plane measurements to a precision of $5.1\times10^{-3}\;π$ radians root-mean-squared-error.

astro-ph.IM↗

arXiv2019

Blockchain-Powered Software Defined Network-Enabled Networking Infrastructure for Cloud Management

Cloud architecture has become a valuable solution for different applications, such as big data analytics, due to its high-degree of availability, scalability and strategic value. However, there still remain challenges in managing cloud architecture, in areas such as cloud security. In this paper, we exploit software-defined networking (SDN) and blockchain technologies to secure cloud management platforms from a networking perspective. We develop a blockchain-powered SDN-enabled networking infrastructure in which the integration between blockchain-based security and autonomy management layer and multi-controller SDN networking layer is defined to enhance the integrity of the control and management messages. Furthermore, our proposed networking infrastructure also enables the autonomous bandwidth provisioning to enhance the availability of cloud architecture. In the simulation section, we evaluate the performance of our proposed blockchain-powered SDN-enabled networking infrastructure by considering different scenarios.

cs.CR↗

arXiv2019

Blockchain as a Service: A Decentralized and Secure Computing Paradigm

Thanks to the advances in machine learning, data-driven analysis tools have become valuable solutions for various applications. However, there still remain essential challenges to develop effective data-driven methods because of the need to acquire a large amount of data and to have sufficient computing power to handle the data. In many instances these challenges are addressed by relying on a dominant cloud computing vendor, but, although commercial cloud vendors provide valuable platforms for data analytics, they can suffer from a lack of transparency, security, and privacy-perservation. Furthermore, reliance on cloud servers prevents applying big data analytics in environments where the computing power is scattered. To address these challenges, a decentralize, secure, and privacy-preserving computing paradigm is proposed to enable an asynchronized cooperative computing process amongst scattered and untrustworthy computing nodes that may have limited computing power and computing intelligence. This paradigm is designed by exploring blockchain, decentralized learning, homomorphic encryption, and software defined networking(SDN) techniques. The performance of the proposed paradigm is evaluated via different scenarios in the simulation section.

cs.CR↗

arXiv2019

Domain Constraint Approximation based Semi Supervision

Deep learning for supervised learning has achieved astonishing performance in various machine learning applications. However, annotated data is expensive and rare. In practice, only a small portion of data samples are annotated. Pseudo-ensembling-based approaches have achieved state-of-the-art results in computer vision related tasks. However, it still relies on the quality of an initial model built by labeled data. Less labeled data may degrade model performance a lot. Domain constraint is another way regularize the posterior but has some limitation. In this paper, we proposed a fuzzy domain-constraint-based framework which loses the requirement of traditional constraint learning and enhances the model quality for semi supervision. Simulations results show the effectiveness of our design.

cs.LG↗

arXiv2019

Spectral Attention-Driven Intelligent Target Signal Identification on a Wideband Spectrum

This paper presents a spectral attention-driven reinforcement learning based intelligent method for effective and efficient detection of important signals in a wideband spectrum. In the work presented in this paper, it is assumed that the modulation technique used is available as a priori knowledge of the targeted important signal. The proposed spectral attention-driven intelligent method is consists of two main components, a spectral correlation function (SCF) based spectral visualization scheme and a spectral attention-driven reinforcement learning mechanism that adaptively selects the spectrum range and implements the intelligent signal detection. Simulations illustrate that the proposed method can achieve high accuracy of signal detection while observation of spectrum is limited to few ranges via effectively selecting the spectrum ranges to be observed. Furthermore, the proposed spectral attention-driven machine learning method can lead to an efficient adaptive intelligent spectrum sensor designs in cognitive radio (CR) receivers.

eess.SP↗