Adversarial robustness has attracted extensive studies recently by revealing the vulnerability and intrinsic characteristics of deep networks. However, existing works on adversarial mainly focus balanced datasets, while real-world data usually exhibits a long-tailed distribution. To push towards more realistic scenarios, in this work we investigate as well defense under distributions. In particular, first reveal negative impacts induced imbalanced both recognition performance robustness,...

With wearing masks becoming a new cultural norm, facial expression recognition (FER) while taking into account has become significant challenge. In this paper, we propose unified multi-branch vision transformer for and mask classification tasks. Our approach extracts shared features both tasks using dual-branch architecture that obtains multi-scale feature representations. Furthermore, cross-task fusion phase processes tokens each task with separate branches, exchanging information cross...

Object detection has achieved substantial progress in the last decade. However, detecting novel classes with only few samples remains challenging, since deep learning under low data regime usually leads to a degraded feature space. Existing works employ holistic fine-tuning paradigm tackle this problem, where model is first pre-trained on all base abundant samples, and then it used carve class Nonetheless, still imperfect. Durning fine-tuning, may implicitly leverage knowledge of multiple...

Objective: The effectiveness and safety of belimumab in Chinese lupus patients with different disease activities were investigated a real-world setting. Method: Patients who received 10 mg/kg intravenously on weeks 0, 2, 4, then every 4 background standard-of-care (SoC) therapy had follow-up more than 6 months enrolled from four centers China. They stratified according to the Safety Estrogens Lupus Erythematosus National Assessment-SLE Disease Activity Index (SELENA-SLEDAI) score at baseline...

Transformer are widely used in various fields such as natural language processing and computer vision. However, the training time for large models can be challenging due to Multi-Head Attention (MHA) mechanism. Especially become larger, becomes more costly. So it is crucial utilize resources efficient model training. Currently, NVIDIA Volta GPU still used. because computational shapes supported by Tensor Core Units (TCU) of differ from other architectures, most efforts have not focused on...

Recent works have demonstrated that deep learning models are vulnerable to backdoor poisoning attacks, where these attacks instill spurious correlations external trigger patterns or objects (e.g., stickers, sunglasses, etc.). We find such signals not necessary, as highly effective backdoors can be easily inserted using rotation-based image transformation. Our method constructs the poisoned dataset by rotating a limited amount of and labeling them incorrectly; once trained with it, victim's...

3D content creation from a single image is long-standing yet highly desirable task. Recent advances introduce 2D diffusion priors, yielding reasonable results. However, existing methods are not hyper-realistic enough for post-generation usage, as users cannot view, render and edit the resulting full range. To address these challenges, we HyperDreamer with several key designs appealing properties: 1) Full-range viewable: 360° mesh modeling high-resolution textures enables of visually...

Lidar is an active remote sensing technology that has many advantages, but the echo lidar signal extremely susceptible to noise and complex atmospheric environment, which affects effective detection range retrieval accuracy. In this paper, a wavelet transform (WT) locally weighted scatterplot smoothing (LOWESS) based on ensemble empirical mode decomposition (EEMD) for Rayleigh denoising was proposed. The WT method used remove in with signal-to-noise ratio (SNR) higher than 16 dB. EEMD...

We present a new loss function called Distribution-Balanced Loss for the multi-label recognition problems that exhibit long-tailed class distributions. Compared to conventional single-label classification problem, are often more challenging due two significant issues, namely co-occurrence of labels and dominance negative (when treated as multiple binary problems). The tackles these issues through key modifications standard cross-entropy loss: 1) way re-balance weights takes into account...

Cogan syndrome (CS) is a rare systemic vasculitis characterized primarily by nonsyphilitic interstitial keratitis and vestibular auditory dysfunction. In this article, we report the case of 31-year-old male diagnosed with CS for 1 year. He was admitted to hospital fever, dizziness, headache, tinnitus, hearing loss. After being treated glucocorticoids, cellular immunosuppressants, infliximab therapy, his symptoms were greatly relieved except Then, he attempted use tocilizumab (TCZ) which...

Metamorphic testing is an effective technique for systems that do not have test oracles, which it practically impossible to know the correct output of arbitrary input. In metamorphic testing, instead checking correctness a output, satisfaction relation among outputs checked. If violation found, system implementation must some defects. However, randomly or accidently generated incorrect may satisfy as well. Therefore, only relations good enough ensure quality. this paper, we propose...

A typological overview is given of the syntax prenominal relative clauses, based on a large number languages different families and areas presented in theory-neutral way. On one hand, previous assumptions are tested against new data. other question addressed to what extent clauses ordinary or unusual, compared types especially postnominal ones.

Photon propagation in biological tissue can be equivalently modeled with Monte Carlo simulations numerically or by the Radiative Transfer Equation (RTE) analytically. However, testing of a program modeling photon is difficult due to unknown character test oracles. Although approaches based on Beer-Lambert law, van de Hulst's table RTE used for program, these are only applied that designed homogeneous media. A rigorous way heterogeneous media needed. In this paper, we investigate...

Crowdsourcing activities, carrying out large-scale tasks via wisdoms of crowds, are widely used in practice. However, it is hard for users to find that suitable them. Thus, many participate tasks, and they not good at or interested in, give answers carelessly randomly. This phenomenon causes heavy astroturfing problem crowdsourcing systems, which only hurts the quality completing but also influences user experience. Therefore, recommendation mechanisms can optimize match between demand....

The diagnosis of blast-induced traumatic brain injury (bTBI) is paramount importance for early care and clinical therapy. Therefore, the rapid bTBI vital to treatment prognosis in clinic. In this paper, we reported a new strategy label-free through serum-based Raman spectroscopy. spectral characteristics serum rat were investigated at 3 h, 24 48 h 72 after mild moderate bTBIs. It has been demonstrated that both position intensity characteristic peaks exhibited apparent differences range...

Recent works have shown that interval bound propagation (IBP) can be used to train verifiably robust neural networks. Reseachers observe an intriguing phenomenon on these IBP trained networks: CROWN, a bounding method based tight linear relaxation, often gives very loose bounds We also most neurons become dead during the training process, which could hurt representation capability of network. In this paper, we study relationship between and prove CROWN is always tighter than when choosing...

Aimed at addressing the problem that subjective selection of start prediction time (SPT) in rolling bearing remaining useful life (RUL) will lead to excessive noise signal, a linear-regression-based SPT point determination was proposed. The sliding window linear regression method used establish windows root mean square (RMS) range obtain RMS gradient domain. threshold for set, and continuous trigger mechanism determine used. experimental results show can adaptively improve accuracy prediction.

Objectives To investigate the effectiveness of belimumab on active lupus nephritis (LN) and explore predictors, including serological biomarkers, renal response to in a real-world setting. Methods This multicentre, observational study enrolled patients with LN receiving intravenous as an add-on therapy 24-hour urine protein≥1 g estimated glomerular filtration rate≥30 mL/min/1.73 m 2 at baseline. Complete (CRR), partial (PRR), no (NRR) primary efficacy (PERR) were evaluated. Multivariable...