arXiv subjects

Jun Lin

Publications and source records attributed to Jun Lin.

At least 91 records · Page 5Linked to original sources

arXiv2020

Magnetic Reconnection Invoked by Sweeping of the CME-Driven Fast-Mode Shock

Coronal waves exist ubiquitously in the solar atmosphere. They are important not only in their own rich physics but also essential candidates of triggering magnetic eruptions in the remote. However, the later mechanism has never been directly confirmed. By revisiting the successive eruptions on 2012 March 7, fast-mode shocks are identified to account for the X5.4 flare-related EUV wave with a velocity of 550 km/s, and appeared faster than 2060$\pm$270 km/s at the front of the corresponding coronal mass ejection in the slow-rising phase. They not only propagated much faster than the local Alfven speed of about 260 km/s, but also simultaneously accompanied by type II radio burst, i.e., a typical feature of shock wave. The observations show that the shock wave disturbs the coronal loops C1 connecting active regions (ARs) 11429 and 11430, which is neighboring a null point region. Following a 40-min-oscillation, an external magnetic reconnection (EMR) occurred in the null point region. About 10 min later, a large-scale magnetic flux rope (MFR) overlaid by the C1 became unstable and erupted quickly. It is thought that the fast-mode shock triggered EMR in the null point region and caused the subsequent eruptions. This scenario is observed directly for the first time, and provides new hint for understanding the physics of solar activities and eruptions.

astro-ph.SR↗

arXiv2020

Hardware Accelerator for Multi-Head Attention and Position-Wise Feed-Forward in the Transformer

Designing hardware accelerators for deep neural networks (DNNs) has been much desired. Nonetheless, most of these existing accelerators are built for either convolutional neural networks (CNNs) or recurrent neural networks (RNNs). Recently, the Transformer model is replacing the RNN in the natural language processing (NLP) area. However, because of intensive matrix computations and complicated data flow being involved, the hardware design for the Transformer model has never been reported. In this paper, we propose the first hardware accelerator for two key components, i.e., the multi-head attention (MHA) ResBlock and the position-wise feed-forward network (FFN) ResBlock, which are the two most complex layers in the Transformer. Firstly, an efficient method is introduced to partition the huge matrices in the Transformer, allowing the two ResBlocks to share most of the hardware resources. Secondly, the computation flow is well designed to ensure the high hardware utilization of the systolic array, which is the biggest module in our design. Thirdly, complicated nonlinear functions are highly optimized to further reduce the hardware complexity and also the latency of the entire system. Our design is coded using hardware description language (HDL) and evaluated on a Xilinx FPGA. Compared with the implementation on GPU with the same setting, the proposed design demonstrates a speed-up of 14.6x in the MHA ResBlock, and 3.4x in the FFN ResBlock, respectively. Therefore, this work lays a good foundation for building efficient hardware accelerators for multiple Transformer networks.

eess.SP↗

arXiv2020

Measurement of magnetic field and relativistic electrons along a solar flare current sheet

In the standard model of solar flares, a large-scale reconnection current sheet is postulated as the central engine for powering the flare energy release and accelerating particles. However, where and how the energy release and particle acceleration occur remain unclear due to the lack of measurements for the magnetic properties of the current sheet. Here we report the measurement of spatially-resolved magnetic field and flare-accelerated relativistic electrons along a current-sheet feature in a solar flare. The measured magnetic field profile shows a local maximum where the reconnecting field lines of opposite polarities closely approach each other, known as the reconnection X point. The measurements also reveal a local minimum near the bottom of the current sheet above the flare loop-top, referred to as a "magnetic bottle". This spatial structure agrees with theoretical predictions and numerical modeling results. A strong reconnection electric field of ~4000 V/m is inferred near the X point. This location, however, shows a local depletion of microwave-emitting relativistic electrons. These electrons concentrate instead at or near the magnetic bottle structure, where more than 99% of them reside at each instant. Our observations suggest that the loop-top magnetic bottle is likely the primary site for accelerating and/or confining the relativistic electrons.

astro-ph.SR↗

arXiv2019

Training Deep Neural Networks Using Posit Number System

With the increasing size of Deep Neural Network (DNN) models, the high memory space requirements and computational complexity have become an obstacle for efficient DNN implementations. To ease this problem, using reduced-precision representations for DNN training and inference has attracted many interests from researchers. This paper first proposes a methodology for training DNNs with the posit arithmetic, a type- 3 universal number (Unum) format that is similar to the floating point(FP) but has reduced precision. A warm-up training strategy and layer-wise scaling factors are adopted to stabilize training and fit the dynamic range of DNN parameters. With the proposed training methodology, we demonstrate the first successful training of DNN models on ImageNet image classification task in 16 bits posit with no accuracy loss. Then, an efficient hardware architecture for the posit multiply-and-accumulate operation is also proposed, which can achieve significant improvement in energy efficiency than traditional floating-point implementations. The proposed design is helpful for future low-power DNN training accelerators.

cs.LG↗

arXiv2019

Design Light-weight 3D Convolutional Networks for Video Recognition Temporal Residual, Fully Separable Block, and Fast Algorithm

Deep 3-dimensional (3D) Convolutional Network (ConvNet) has shown promising performance on video recognition tasks because of its powerful spatio-temporal information fusion ability. However, the extremely intensive requirements on memory access and computing power prohibit it from being used in resource-constrained scenarios, such as portable and edge devices. So in this paper, we first propose a two-stage Fully Separable Block (FSB) to significantly compress the model sizes of 3D ConvNets. Then a feature enhancement approach named Temporal Residual Gradient (TRG) is developed to improve the performance of compressed model on video tasks, which provides higher accuracy, faster convergency and better robustness. Moreover, in order to further decrease the computing workload, we propose a hybrid Fast Algorithm (hFA) to drastically reduce the computation complexity of convolutions. These methods are effectively combined to design a light-weight and efficient ConvNet for video recognition tasks. Experiments on the popular dataset report 2.3x compression rate, 3.6x workload reduction, and 6.3% top-1 accuracy gain, over the state-of-the-art SlowFast model, which is already a highly compact model. The proposed methods also show good adaptability on traditional 3D ConvNet, demonstrating 7.4x more compact model, 11.0x less workload, and 3.0% higher accuracy

cs.CV↗

arXiv2019

A Hardware-Oriented and Memory-Efficient Method for CTC Decoding

The Connectionist Temporal Classification (CTC) has achieved great success in sequence to sequence analysis tasks such as automatic speech recognition (ASR) and scene text recognition (STR). These applications can use the CTC objective function to train the recurrent neural networks (RNNs), and decode the outputs of RNNs during inference. While hardware architectures for RNNs have been studied, hardware-based CTCdecoders are desired for high-speed CTC-based inference systems. This paper, for the first time, provides a low-complexity and memory-efficient approach to build a CTC-decoder based on the beam search decoding. Firstly, we improve the beam search decoding algorithm to save the storage space. Secondly, we compress a dictionary (reduced from 26.02MB to 1.12MB) and use it as the language model. Meanwhile searching this dictionary is trivial. Finally, a fixed-point CTC-decoder for an English ASR and an STR task using the proposed method is implemented with C++ language. It is shown that the proposed method has little precision loss compared with its floating-point counterpart. Our experiments demonstrate the compression ratio of the storage required by the proposed beam search decoding algorithm are 29.49 (ASR) and 17.95 (STR).

eess.SP↗

arXiv2018

Magnetic reconnection in the low solar chromosphere with a more realistic radiative cooling model

Magnetic reconnection is the most likely mechanism responsible for the high temperature events that are observed in strongly magnetized locations around the temperature minimum in the low solar chromosphere. This work improves upon our previous work ["Magnetic Reconnection in Strongly Magnetized Regions of the Low Solar Chromosphere", The Astrophysical Journal 852, 95 (2018) ] by using a more realistic radiative cooling model computed from the OPACITY project and the CHIANTI database. We find that the rate of ionization of the neutral component of the plasma is still faster than recombination within the current sheet region. For low $β$ plasmas, the ionized and neutral fluid flows are well-coupled throughout the reconnection region resembling the single-fluid Sweet-Parker model dynamics. Decoupling of the ion and neutral inflows appears in the higher $β$ case with $β_0=1.46$, which leads to a reconnection rate about three times faster than the rate predicted by the Sweet-Parker model. In all cases, the plasma temperature increases with time inside the current sheet, and the maximum value is above $2\times10^4$~K when the reconnection magnetic field strength is greater than $500$~G. While the more realistic radiative cooling model does not result in qualitative changes of the characteristics of magnetic reconnection, it is necessary for studying the variations of the plasma temperature and ionization fraction inside current sheets in strongly magnetized regions of the low solar atmosphere. It is also important for studying energy conversion during the magnetic reconnection process when the hydrogen-dominated plasma approaches full ionization.

astro-ph.SR↗

arXiv2018

SGAD: Soft-Guided Adaptively-Dropped Neural Network

Deep neural networks (DNNs) have been proven to have many redundancies. Hence, many efforts have been made to compress DNNs. However, the existing model compression methods treat all the input samples equally while ignoring the fact that the difficulties of various input samples being correctly classified are different. To address this problem, DNNs with adaptive dropping mechanism are well explored in this work. To inform the DNNs how difficult the input samples can be classified, a guideline that contains the information of input samples is introduced to improve the performance. Based on the developed guideline and adaptive dropping mechanism, an innovative soft-guided adaptively-dropped (SGAD) neural network is proposed in this paper. Compared with the 32 layers residual neural networks, the presented SGAD can reduce the FLOPs by 77% with less than 1% drop in accuracy on CIFAR-10.

cs.LG↗

arXiv2018

Parameter-free $\ell_p$-Box Decoding of LDPC Codes

The Alternating Direction Method of Multipliers (ADMM) decoding of Low Density Parity Check (LDPC) codes has received many attentions due to its excellent performance at the error floor region. In this paper, we develop a parameter-free decoder based on Linear Program (LP) decoding by replacing the binary constraint with the intersection of a box and an $\ell_p$ sphere. An efficient $\ell_2$-box ADMM is designed to handle this model in a distributed fashion. Numerical experiments demonstrate that our decoder attains better adaptability to different Signal-to-Noise Ratio and channels.

cs.IT↗

arXiv2018

Numerical Studies of the Kelvin-Hemholtz Instability in the Coronal Jet

The Kelvin-Hemholtz (K-H) instability in the corona EUV jet is studied via 2.5 MHD numerical simulations. The jet results from magnetic reconnection due to the interation of the new emerging magnetic field and the pre-existing magnetic field in the corona. Our results show that the Alfven Mach number along the jet is about 5-14 just before the instability occurs, and it is even higher than 14 at some local areas. During the K-H instability process, several vortex-like plasma blobs of high temperature and high density appear along the jet, and magnetic fields have also been rolled up and the magnetic configuration including anti-parallel magnetic fields forms, which leads to magnetic reconnection at many X-points and current sheet fragments inside the vortex-like blob. After magnetic islands appear inside the main current sheet, the total kinetic energy of the reconnection outflows decreases, and cannot support the formation of the vortex-like blob along the jet any longer, then the K-H instability eventually disappears. We also present the results about how the guide field and the flux emerging speed affect the K-H instability. We find that the strong guide field inhibits the shock formation in the reconnection upward outflow regions but helps secondary magnetic islands appear earlier in the main current sheet, and then apparently suppresses the K-H instability. As the speed of the emerging magnetic field decreases, the K-H instability appears later, the highest temperature inside the vortex blob gets lower and the vortex structure gets smaller.

astro-ph.SR↗

arXiv2017

Magnetic reconnection in strongly magnetized regions of the low solar chromosphere

Magnetic reconnection in strongly magnetized regions around the temperature minimum region of the low solar atmosphere is studied by employing MHD-based simulations of a partially ionized plasma within a reactive 2.5D multi-fluid model. It is shown that in the absence of magnetic nulls in a low $β$ plasma the ionized and neutral fluid flows are well-coupled throughout the reconnection region. However, non-equilibrium ionization-recombination dynamics play a critical role in determining the structure of the reconnection region, lead to much lower temperature increases and a faster magnetic reconnection rate as compared to simulations that assume plasma to be in ionization-recombination equilibrium. The rate of ionization of the neutral component of the plasma is always faster than recombination within the current sheet region even when the initial plasma $β$ is as high as $β_0=1.46$. When the reconnecting magnetic field is in excess of a kilogauss and the plasma $β$ is lower than 0.0145, the initially weakly ionized plasmas can become fully ionized within the reconnection region and the current sheet can be strongly heated to above $2.5\times10^4$~K, even as most of the collisionally dissipated magnetic energy is radiated away. The Hall effect increases the reconnection rate slightly, but in the absence of magnetic nulls it does not result in significant asymmetries or change the characteristics of the reconnection current sheet down to meter scales.

astro-ph.SR↗

arXiv2017

Blob formation and ejection in coronal jets due to the plasmoid and Kelvin-Helmholtz instabilities

We perform two-dimensional resistive magnetohydrodynamic simulations of coronal jets driven by flux emergence along the lower boundary. The reconnection layers are susceptible the formation of blobs that are ejected in the jet. Our simulation with low plasma $β$ (Case I) shows that magnetic islands form easily and propagate upwards in the jet. These islands are multithermal and thus are predicted to show up in hot channels (335 Å\, and 211 Å) and the cool channel (304 Å) in observations by the Atmospheric Imaging Assembly (AIA) on the \emph{Solar Dynamics Observatory}. The islands have maximum temperatures of 8 MK, lifetimes of 120 s, diameters of 6 Mm, and velocities of 200 km s$^{-1}$. These parameters are similar to the properties of blobs observed in EUV jets by AIA\@. The Kelvin-Helmholtz instability develops in our simulation with moderately high plasma $β$ (Case II), and leads to the formation of bright vortex-like blobs above the multiple high magneto-sonic Mach number regions that appear along the jet. These vortex-like blobs can also be identified in the AIA channels. However, they eventually move downward and disappear after the high magneto-sonic Mach number regions disappear. In the lower plasma $β$ case, the lifetime for the jet is shorter, the jet and magnetic islands are formed with higher velocities and temperatures, the current sheet fragments are more chaotic, and more magnetic islands are generated. Our results show that the plasmoid instability and Kelvin-Helmholtz instability along the jet are both possible causes of the formation of blobs observed at extreme ultraviolet (EUV) wavelengths.

astro-ph.SR↗

arXiv2016

The Broken Lane of a Type II Radio Burst Caused by Collision of a Coronal Shock with a Flare Current Sheet

We investigated a peculiar metric type II solar radio burst with a broken lane structure, which was observed on November 13, 2012. In addition to the radio data, we also studied the data in the other wavelengths. The bursts were associated with two CMEs and two flares that originated from active region AR 11613. A long current sheet was developed in the first CME, and the second CME collided with the current sheet first and then merged with the first one. Combing information revealed by the multi-wavelength data indicated that a coronal shock accounting for the type II radio burst, and that the collision of this shock with the current sheet resulted in the broken lane of the type II radio burst. The type II burst lane resumed after the shock passed through the current sheet. We further estimated the thickness of the current sheet according to the gap on the lane of the type II burst, and found that the result is consistent with previous ones obtained for various events observed in different wavelengths by different instruments. In addition, the regular type II burst associated with the first CME/flare was also studied, and the magnetic field in each source region of the two type II bursts was further deduced in different way.

astro-ph.SR↗

arXiv2016

Heating mechanisms in the low solar atmosphere through magnetic reconnection in current sheets

We simulate several magnetic reconnection processes in the low solar chromosphere/photosphere, the radiation cooling, heat conduction and ambipolar diffusion are all included. Our numerical results indicate that both the high temperature($ \gtrsim 8\times10^4$~K) and low temperature($\sim 10^4$~K) magnetic reconnection events can happen in the low solar atmosphere ($100\sim600$~km above the solar surface). The plasma $β$ controlled by plasma density and magnetic fields is one important factor to decide how much the plasma can be heated up. The low temperature event is formed in a high $β$ magnetic reconnection process, Joule heating is the main mechanism to heat plasma and the maximum temperature increase is only several thousand Kelvin. The high temperature explosions can be generated in a low $β$ magnetic reconnection process, slow and fast-mode shocks attached at the edges of the well developed plasmoids are the main physical mechanisms to heat the plasma from several thousand Kelvin to over $8\times10^4$~K. Gravity in the low chromosphere can strongly hinder the plasmoind instability and the formation of slow-mode shocks in a vertical current sheet. Only small secondary islands are formed; these islands, however, are not well developed as those in the horizontal current sheets. This work can be applied for understanding the heating mechanism in the low solar atmosphere and could possibly be extended to explain the formation of common low temperature EBs ($\sim10^4$~K) and the high tenperature IRIS bombs ($\gtrsim 8\times10^4$) in the future.

astro-ph.SR↗

arXiv2016

Intra-layer Nonuniform Quantization for Deep Convolutional Neural Network

Deep convolutional neural network (DCNN) has achieved remarkable performance on object detection and speech recognition in recent years. However, the excellent performance of a DCNN incurs high computational complexity and large memory requirement. In this paper, an equal distance nonuniform quantization (ENQ) scheme and a K-means clustering nonuniform quantization (KNQ) scheme are proposed to reduce the required memory storage when low complexity hardware or software implementations are considered. For the VGG-16 and the AlexNet, the proposed nonuniform quantization schemes reduce the number of required memory storage by approximately 50\% while achieving almost the same or even better classification accuracy compared to the state-of-the-art quantization method. Compared to the ENQ scheme, the proposed KNQ scheme provides a better tradeoff when higher accuracy is required.

cs.CV↗

arXiv2015

Numerical experiments on the detailed energy conversion and spectrum studies in a corona current sheet

In this paper, we study the energy conversion and spectra in a corona current sheet by 2.5-dimensional MHD numerical simulations. Numerical results show that many Petschek-like fine structures with slow-mode shocks mediated by plasmoid instabilities develop during the magnetic reconnection process. The termination shocks can also be formed above the primary magnetic island and at the head of secondary islands. These shocks play important roles in generating thermal energy in a corona current sheet. For a numerical simulation with initial conditions close to the solar corona environment, the ratio of the generated thermal energy to the total dissipated magnetic energy is around $1/5$ before secondary islands appear. After secondary islands appear, the generated thermal energy starts to increase sharply and this ratio can reach a value about $3/5$. In an environment with a relatively lower plasma density and plasma $β$, the plasma can be heated to a much higher temperature. After secondary islands appear, the one dimensional energy spectra along the current sheet do not behave as a simple power law and the spectrum index increases with the wave number. The average spectrum index for the magnetic energy spectrum along the current sheet is about $1.8$. The two dimensional spectra intuitively show that part of the high energy is cascaded to large $kx$ and $ky$ space after secondary islands appear. The plasmoid distribution function calculated from numerical simulations behaves as a power law closer to $f(ψ) \sim ψ^{-1}$ in the intermediate $ψ$ regime. By using $η_{eff} = v_{inflow}\cdot L$, the effective magnetic diffusivity is estimated about $10^{11}\sim10^{12}$~m$^2$\,s$^{-1}$.

astro-ph.SR↗

arXiv2015

A multi-mode area-efficient SCL polar decoder

Polar codes are of great interest since they are the first provably capacity-achieving forward error correction codes. To improve throughput and to reduce decoding latency of polar decoders, maximum likelihood (ML) decoding units are used by successive cancellation list (SCL) decoders as well as successive cancellation (SC) decoders. This paper proposes an approximate ML (AML) decoding unit for SCL decoders first. In particular, we investigate the distribution of frozen bits of polar codes designed for both the binary erasure and additive white Gaussian noise channels, and take advantage of the distribution to reduce the complexity of the AML decoding unit, improving the area efficiency of SCL decoders. Furthermore, a multi-mode SCL decoder with variable list sizes and parallelism is proposed. If high throughput or small latency is required, the decoder decodes multiple received codewords in parallel with a small list size. However, if error performance is of higher priority, the multi-mode decoder switches to a serial mode with a bigger list size. Therefore, the multi-mode SCL decoder provides a flexible tradeoff between latency, throughput and error performance, and adapts to different throughput and latency requirements at the expense of small overhead. Hardware implementation and synthesis results show that our polar decoders not only have a better area efficiency but also easily adapt to different communication channels and applications.

cs.IT↗

arXiv2015

Reduced Complexity Belief Propagation Decoders for Polar Codes

Polar codes are newly discovered capacity-achieving codes, which have attracted lots of research efforts. Polar codes can be efficiently decoded by the low-complexity successive cancelation (SC) algorithm and the SC list (SCL) decoding algorithm. The belief propagation (BP) decoding algorithm not only is an alternative to the SC and SCL decoders, but also provides soft outputs that are necessary for joint detection and decoding. Both the BP decoder and the soft cancelation (SCAN) decoder were proposed for polar codes to output soft information about the coded bits. In this paper, first a belief propagation decoding algorithm, called reduced complexity soft cancelation (RCSC) decoding algorithm, is proposed. Let $N$ denote the block length. Our RCSC decoding algorithm needs to store only $5N-3$ log-likelihood ratios (LLRs), significantly less than $4N-2+\frac{N\log_2N}{2}$ and $N(\log_2N+1)$ LLRs needed by the BP and SCAN decoders, respectively, when $N\geqslant 64$. Besides, compared to the SCAN decoding algorithm, our RCSC decoding algorithm eliminates unnecessary additions over the real field. Then the simplified SC (SSC) principle is applied to our RCSC decoding algorithm, and the resulting SSC-aided RCSC (S-RCSC) decoding algorithm further reduces the computational complexity. Finally, based on the S-RCSC decoding algorithm, we propose a corresponding memory efficient decoder architecture, which has better error performance than existing architectures. Besides, our decoder architecture consumes less energy on updating LLRs.

cs.IT↗