Search

Spiking neural networks (SNNs) transmit information via low-power binary spikes and have received widespread attention in areas such as computer vision and reinforcement learning. However, there have been very few explorations of SNNs in more practical industrial scenarios. In this paper, we focus on the application of SNNs in bearing fault diagnosis to facilitate the integration of high-performance AI algorithms and real-world industries. In particular, we identify two key limitations of existing SNN fault diagnosis methods: inadequate encoding capacity that necessitates cumbersome data preprocessing, and non-spike-oriented architectures that constrain the performance of SNNs. To alleviate these problems, we propose a Multi-scale Residual Attention SNN (MRA-SNN) to simultaneously improve the efficiency, performance, and robustness of SNN methods. By incorporating a lightweight attention mechanism, we have designed a multi-scale attention encoding module to extract multiscale fault features from vibration signals and encode them as spatio-temporal spikes, eliminating the need for complicated preprocessing. Then, the spike residual attention block extracts high-dimensional fault features and enhances the expressiveness of sparse spikes with the attention mechanism for end-to-end diagnosis. In addition, the performance and robustness of MRA-SNN is further enhanced by introducing the lightweight attention mechanism within the spiking neurons to simulate the biological dendritic filtering effect. Extensive experiments on MFPT and JNU benchmark datasets demonstrate that MRA-SNN significantly outperforms existing methods in terms of accuracy, energy consumption and noise robustness, and is more feasible for deployment in real-world industrial scenarios., Comment: 13 pages, 10 figures

Spiking neural networks (SNNs) have received widespread attention as an ultra-low energy computing paradigm. Recent studies have focused on improving the feature extraction capability of SNNs, but they suffer from inefficient inference and suboptimal performance. In this paper, we propose a simple yet effective temporal reversed training (TRT) method to optimize the spatio-temporal performance of SNNs and circumvent these problems. We perturb the input temporal data by temporal reversal, prompting the SNN to produce original-reversed consistent output logits and to learn perturbation-invariant representations. For static data without temporal dimension, we generalize this strategy by exploiting the inherent temporal property of spiking neurons for spike feature temporal reversal. In addition, we utilize the lightweight ``star operation" (element-wise multiplication) to hybridize the original and temporally reversed spike firing rates and expand the implicit dimensions, which serves as spatio-temporal regularization to further enhance the generalization of the SNN. Our method involves only an additional temporal reversal operation and element-wise multiplication during training, thus incurring negligible training overhead and not affecting the inference efficiency at all. Extensive experiments on static/neuromorphic object/action recognition, and 3D point cloud classification tasks demonstrate the effectiveness and generalizability of our method. In particular, with only two timesteps, our method achieves 74.77\% and 90.57\% accuracy on ImageNet and ModelNet40, respectively., Comment: 15 pages, 8 figures

Since proposed, spiking neural networks (SNNs) gain recognition for their high performance, low power consumption and enhanced biological interpretability. However, while bringing these advantages, the binary nature of spikes also leads to considerable information loss in SNNs, ultimately causing performance degradation. We claim that the limited expressiveness of current binary spikes, resulting in substantial information loss, is the fundamental issue behind these challenges. To alleviate this, our research introduces a multi-bit information transmission mechanism for SNNs. This mechanism expands the output of spiking neurons from the original single bit to multiple bits, enhancing the expressiveness of the spikes and reducing information loss during the forward process, while still maintaining the low energy consumption advantage of SNNs. For SNNs, this represents a new paradigm of information transmission. Moreover, to further utilize the limited spikes, we extract effective signals from the previous layer to re-stimulate the neurons, thus encouraging full spikes emission across various bit levels. We conducted extensive experiments with our proposed method using both direct training method and ANN-SNN conversion method, and the results show consistent performance improvements., Comment: Under review

Spiking neural networks (SNNs) have attracted considerable attention for their event-driven, low-power characteristics and high biological interpretability. Inspired by knowledge distillation (KD), recent research has improved the performance of the SNN model with a pre-trained teacher model. However, additional teacher models require significant computational resources, and it is tedious to manually define the appropriate teacher network architecture. In this paper, we explore cost-effective self-distillation learning of SNNs to circumvent these concerns. Without an explicit defined teacher, the SNN generates pseudo-labels and learns consistency during training. On the one hand, we extend the timestep of the SNN during training to create an implicit temporal ``teacher" that guides the learning of the original ``student", i.e., the temporal self-distillation. On the other hand, we guide the output of the weak classifier at the intermediate stage by the final output of the SNN, i.e., the spatial self-distillation. Our temporal-spatial self-distillation (TSSD) learning method does not introduce any inference overhead and has excellent generalization ability. Extensive experiments on the static image datasets CIFAR10/100 and ImageNet as well as the neuromorphic datasets CIFAR10-DVS and DVS-Gesture validate the superior performance of the TSSD method. This paper presents a novel manner of fusing SNNs with KD, providing insights into high-performance SNN learning methods., Comment: 17 pages, 6 figures

Existing computer vision methods mainly focus on the recognition of rigid objects, whereas the recognition of flexible objects remains unexplored. Recognizing flexible objects poses significant challenges due to their inherently diverse shapes and sizes, translucent attributes, ambiguous boundaries, and subtle inter-class differences. In this paper, we claim that these problems primarily arise from the lack of object saliency. To this end, we propose the Flexible Vision Graph Neural Network (FViG) to optimize the self-saliency and thereby improve the discrimination of the representations for flexible objects. Specifically, on one hand, we propose to maximize the channel-aware saliency by extracting the weight of neighboring nodes, which adapts to the shape and size variations in flexible objects. On the other hand, we maximize the spatial-aware saliency based on clustering to aggregate neighborhood information for the centroid nodes, which introduces local context information for the representation learning. To verify the performance of flexible objects recognition thoroughly, for the first time we propose the Flexible Dataset (FDA), which consists of various images of flexible objects collected from real-world scenarios or online. Extensive experiments evaluated on our Flexible Dataset demonstrate the effectiveness of our method on enhancing the discrimination of flexible objects., Comment: under review

Neuromorphic object recognition with spiking neural networks (SNNs) is the cornerstone of low-power neuromorphic computing. However, existing SNNs suffer from significant latency, utilizing 10 to 40 timesteps or more, to recognize neuromorphic objects. At low latencies, the performance of existing SNNs is drastically degraded. In this work, we propose the Shrinking SNN (SSNN) to achieve low-latency neuromorphic object recognition without reducing performance. Concretely, we alleviate the temporal redundancy in SNNs by dividing SNNs into multiple stages with progressively shrinking timesteps, which significantly reduces the inference latency. During timestep shrinkage, the temporal transformer smoothly transforms the temporal scale and preserves the information maximally. Moreover, we add multiple early classifiers to the SNN during training to mitigate the mismatch between the surrogate gradient and the true gradient, as well as the gradient vanishing/exploding, thus eliminating the performance degradation at low latency. Extensive experiments on neuromorphic datasets, CIFAR10-DVS, N-Caltech101, and DVS-Gesture have revealed that SSNN is able to improve the baseline accuracy by 6.55% ~ 21.41%. With only 5 average timesteps and without any data augmentation, SSNN is able to achieve an accuracy of 73.63% on CIFAR10-DVS. This work presents a heterogeneous temporal scale SNN and provides valuable insights into the development of high-performance, low-latency SNNs., Comment: Accepted by AAAI 2024

In this work, six convolutional neural networks (CNNs) have been trained based on %different feature images and arrays from the database including 15,638 superflare candidates on solar-type stars, which are collected from the three-years observations of Transiting Exoplanet Survey Satellite ({\em TESS}). These networks are used to replace the artificially visual inspection, which was a direct way to search for superflares, and exclude false positive events in recent years. Unlike other methods, which only used stellar light curves to search superflare signals, we try to identify superflares through {\em TESS} pixel-level data with lower risks of mixing false positive events, and give more reliable identification results for statistical analysis. The evaluated accuracy of each network is around 95.57\%. After applying ensemble learning to these networks, stacking method promotes accuracy to 97.62\% with 100\% classification rate, and voting method promotes accuracy to 99.42\% with relatively lower classification rate at 92.19\%. We find that superflare candidates with short duration and low peak amplitude have lower identification precision, as their superflare-features are hard to be identified. The database including 71,732 solar-type stars and 15,638 superflare candidates from {\em TESS} with corresponding feature images and arrays, and trained CNNs in this work are public available., Comment: Published in ApJ, 27 pages, 11 figures, 3 tables

Magnetars are highly magnetized neutron stars that are characterized by recurrent emission of short-duration bursts in soft gamma-rays/hard X-rays. Recently, FRB 200428 were found to be associated with an X-ray burst from a Galactic magnetar. Two fast radio bursts (FRBs) show mysterious periodic activity. However, whether magnetar X-ray bursts are periodic phenomena is unclear. In this paper, we investigate the period of SGR 1806-20 activity. More than 3000 short bursts observed by different telescopes are collected, including the observation of RXTE, HETE-2, ICE and Konus. We consider the observation windows and divide the data into two sub-samples to alleviate the effect of unevenly sample. The epoch folding and Lomb-Scargle methods are used to derive the period of short bursts. We find a possible period about $ 398.20 \pm 25.45 $ days. While other peaks exist in the periodograms. If the period is real, the connection between short bursts of magnetars and FRBs should be extensively investigated., Comment: 14 pages, 4 figures, accepted for publication in ApJ

In this work, 1272 superflares on 311 stars are collected from 22,539 solar-type stars from the second-year observation of Transiting Exoplanet Survey Satellite (TESS), which almost covered the northern hemisphere of the sky. Three superflare stars contain hot Jupiter candidates or ultrashort-period planet candidates. We obtain $\gamma = -1.76\pm 0.11$ of the correlation between flare frequency and flare energy ($dN/dE\propto E^{-\gamma}$) for all superflares and get $\beta=0.42\pm0.01$ of the correlation between superflare duration and energy ($T_{\text {duration }} \propto E^{\beta}$), which supports that a similar mechanism is shared by stellar superflares and solar flares. Stellar photometric variability ($R_{\rm var}$) is estimated for all solar-type stars, and the relation of $E\propto {R_{\rm var}}^{3/2}$ is included. An indicator of chromospheric activity ($S$-index) is obtained by using data from the Large Sky Area Multi-Object Fiber Spectroscopic Telescope (LAMOST) for 7454 solar-type stars. Distributions of these two properties indicate that the Sun is generally less active than superflare stars. We find that saturation-like feature of $R_{\rm var}\sim 0.1$ may be the reason for superflare energy saturating around $10^{36}$ erg. Object TIC 93277807 was captured by the TESS first-year mission and generated the most energetic superflare. This superflare is valuable and unique that can be treated as an extreme event, which may be generated by different mechanisms rather than other superflares., Comment: Accepted for Publication in ApJS. Comments are welcome. 35 pages, 15 figures, 7 tables

Accompanied by an X-ray burst, the fast radio burst (FRB) FRB 200428 was recently confirmed as originating from the Galactic magnetar soft gamma repeater (SGR) SGR J1935+2154. Just before and after FRB 200428 was detected, the Five-hundred-meter Aperture Spherical radio Telescope (FAST) had been monitoring SGR J1935+2154 for eight hours. From UTC 2020 April 27 23:55:00 to 2020 April 28 00:50:37, FAST detected no pulsed radio emission from SGR J1935+2154, while Fermi/Gamma-ray Burst Monitor registered 34 bursts in the X/soft $\gamma$-ray band, forming a unique sample of X-ray bursts in the absence of FRBs. After a comprehensive analysis on light curves, time-integrated, and time-resolved spectral properties of these FRB-absent X-ray bursts, we compare this sample with the FRB-associated X-ray burst detected by Insight-HXMT, INTEGRAL, and Konus-Wind. The FRB-associated burst distinguishes itself from other X-ray bursts by its nonthermal spectrum and a higher spectral peak energy, but otherwise is not atypical. We also compare the cumulative energy distribution of our X-ray burst sample with that of first repeating FRB source, FRB 121102, with the calibration of FRB 200428-X-ray burst association. We find a similarity between the two, offering indirect support of the magnetar origin of cosmological FRBs. The event rate density of magnetar bursts is about $\sim 150$ times higher than the FRB event rate density at the energy of FRB 200428. This again suggests that, if all FRBs originate from magnetars, only a small fraction of X-ray bursts are associated with FRBs., Comment: 18 pages, 3 tables, 8 figures. Published in the Astrophysical Journal Letter on January 12, 2021

Timely, creditable, and fine-granular case information is vital for local communities and individual citizens to make rational and data-driven responses to the COVID-19 pandemic. This paper presents CovidNet, a COVID-19 tracking project associated with a large scale epidemic dataset, which was initiated by 1Point3Acres. To the best of our knowledge, the project is the only platform providing real-time global case information of more than 4,124 sub-divisions from over 27 countries worldwide with multi-language supports. The platform also offers interactive visualization tools to analyze the full historical case curves in each region. Initially launched as a voluntary project to bridge the data transparency gap in North America in January 2020, this project by far has become one of the major independent sources worldwide and has been consumed by many other tracking platforms. The accuracy and freshness of the dataset is a result of the painstaking efforts from our voluntary teamwork, crowd-sourcing channels, and automated data pipelines. As of May 18, 2020, the project website has been visited more than 200 million times and the CovidNet dataset has empowered over 522 institutions and organizations worldwide in policy-making and academic researches. All datasets are openly accessible for non-commercial purposes at https://coronavirus.1point3acres.com via a formal request through our APIs., Comment: 10 pages, 5 figures, 2 tables

Superflares, as strong explosions on stars, have been well studied with the progress of space time-domain astronomy. In this work, we present the study of superflares on solar-type stars using Transiting Exoplanet Survey Satellite ({\em{TESS}}) data. 13 sectors of observations during the first year of the {\em TESS} mission have covered the southern hemisphere of the sky, containing 25,734 solar-type stars. We verified 1,216 superflares on 400 solar-type stars through automatic search and visual inspection with 2-minute cadence data. Our result suggests a higher superflare frequency distribution than the result from {\em Kepler}. The reason may be that the majority of {\em TESS} solar-type stars in our dataset are rapidly rotating stars. The power-law index $\gamma$ of the superflare frequency distribution ($dN/dE\propto E^{-\gamma}$) is constrained to be $\gamma = 2.16\pm 0.10$, which is a little larger than that of solar flares but consistent with the results from {\em Kepler}. Because only 7 superflares of Sun-like stars are detected, we may not give a robust superflare occurrence frequency. And four stars are accompanied by unconfirmed hot planet candidates. Therefore, superflares are possibly caused by stellar magnetic activities instead of planet-star interactions. We also find an extraordinary star TIC43472154, which exhibits about 200 superflares per year. In addition, the correlation between energy and duration of superflares ($T_{\text {duration }} \propto E^{\beta}$) is analyzed. We derive the power-law index to be $\beta=0.42\pm0.01$, which is a little larger than $\beta=1/3$ from the prediction according to magnetic reconnection theory., Comment: Accepted for Publication in ApJ. Comments are welcome. 22 pages, 10 figures, 8 tables

We present HST/FOS spectra of the two bright images (A and B) of the gravitationally lensed QSO 0957+561 in the wavelength range 2200-3300 A. We find that the absorption system (Z(sub abs)) = 1.3911) near z(sub em) is a weak, damped Ly alpha system with strong Ly alpha absorption lines seen in both images. However, the H(I) column densities are different, with the line of sight to image A intersecting a larger column density. The continuum shapes of the two spectra differ in the sense that the flux level of image A increases more slowly toward shorter wavelengths than that of image B. We explain this as the result of differential reddening by dust grains in the damped Ly alpha absorber. A direct outcome of this explanation is a determination of the dust-to-gas ratio, k, in the damped Ly alpha system. We derive k = 0.55 + 0.18 for a simple 1/lambda extinction law and k = 0.31 + 0.10 for the Galactic extinction curve. For gravitationally lensed systems with damped Ly alpha absorbers, our method is a powerful tool for determining the values and dispersion of k, and the shapes of extinction curves, especially in the FUV and EUV regions. We compare our results with previous work.

We discuss the strength and frequency of broad Ne VIII lambda 774 emission from quasars measured in the Hubble Space Telescope Faint Object Spectrograph (HST-FOS) snapshot survey (Absnap). Five sources in the survey have suitable redshifts (0.86 less than or equal to Z(sub em) less than or equal to 1.31), signal-to-noise ratios and no Lyman limit absorptions. Three of the five sources have a strong broad emission line near 774 A (rest), and the remaining two sources have a less securely measured line near this wavelength. We identify these lines with Ne VIII lambda 774 based on the measured wavelengths and theoretical estimates of various line fluxes (Hamann et al. 1995a). Secure Ne VIII detections occur in both radio-loud and radio-quiet sources. We tentatively conclude that broad Ne VIII lambda 774 emission is common in quasars, with typical strengths between approximately 25% and approximately 200% of O VI lambda 1034. These Ne VIII lambda 774 measurements imply that the broad emission line regions have a much hotter and more highly ionized component than previously recognized. They also suggest that quasar continua have substantial ionizing flux out to energies greater than 207 eV (greater than 15.2 ryd, lambda less than 60 A). Photoionization calculations using standard incident spectra indicate that the Ne VIII emission requires ionization parameters U greater than or = 5, total column densities N(sub H) greater than or = 10(sub 22)/sq cm and covering factors greater than or = 25%. The temperatures could be as high as approximately 10(exp 5) K. If the gas is instead collisionally ionized, strong Ne VIII would imply equilibrium temperatures in the range approximately 400,000 less than or approximately = T(sub e) less than or approximately = 10(exp 6) K. In either case, the highly ionized Ne VIII emission regions would appear as X-ray 'warm absorbers' if they lie along our line of sight to the X-ray continuum source.

The possibility of using D(sub A), the mean depression of a quasar spectrum due to Ly-alpha forest absorption, to study the number density evolution of the Ly-alpha forest clouds is examined in some detail. Current D(sub A) measurements are made against a continuum that is a power-law extrapolation from the continuum longward of Ly-alpha emission. Compared to the line-counting approach, the D(sub A)-method has the advantage that the D(sub A) measurements are not affected by line-blending effects. However, we find using low-redshift quasar spectra obtained with the Hubble Space Telescope (HST), where the true continuum in the Ly-alpha forest can be estimated fairly reliably because of the much lower density of the Ly-alpha forest lines, that the extrapolated continuum often deviates systematically from the true continuum in the forest region. Such systematic continuum errors introduce large errors in the D(sub A) measurements. The current D(sub A) measurements may also be significantly biased by the possible presence of the Gunn-Peterson absorption. We propose a modification to the existing D(sub A)-method, namely, to measure D(sub A) against a locally established continuum in the Ly-alpha forest. Under conditions that the quasar spectrum has good resolution and S/N to allow for a reliable estimate of the local continuum in the Ly-alpha forest, the modified D(sub A) measurements should be largely free of the systematic uncertainties suffered by the existing D(sub A) measurements. We also introduce a formalism based on the work of Zuo (1993) to simplify the application of the D(sub A)-method(s) to real data. We discuss the merits and limitations of the modified D(sub A)-method, and conclude that it is a useful alternative. Our findings that the extrapolated continuum from longward of Ly-alpha emission often deviates systematically from the true continuum in the Ly-alpha forest present a major problem in the study of the Gunn-Peterson absorption.

Intensity correlation of ionizing background at high redshifts is discussed. The intensity correlation function xi(sub j) and the absorption line equivalent width correlation xi(sub 1/W) are discussed.