Deep neural networks are vulnerable to adversarial examples, which poses security concerns on these algorithms due to the potentially severe consequences. Adversarial attacks serve as an important surrogate to evaluate the robustness of deep learning models before they are deployed. However, most of existing adversarial attacks can only fool a black-box model with a low success rate. To address this issue, we propose a broad class of momentum-based iterative algorithms to boost adversarial attacks. By integrating the momentum term into the iterative process for attacks, our methods can stabilize update directions and escape from poor local maxima during the iterations, resulting in more transferable adversarial examples. To further improve the success rates for black-box attacks, we apply momentum iterative algorithms to an ensemble of models, and show that the adversarially trained models with a strong defense ability are also vulnerable to our black-box attacks. We hope that the proposed methods will serve as a benchmark for evaluating the robustness of various deep models and defense methods. With this method, we won the first places in NIPS 2017 Non-targeted Adversarial Attack and Targeted Adversarial Attack competitions. * Corresponding author. Alps: 94.39%Dog: 99.99%Puffer: 97.99% Crab: 100.00%

Background: In late December 2019, an outbreak of acute respiratory illness, coronavirus disease 2019 , emerged in Wuhan, China. We aimed to study the epidemiology, clinical features and short-term outcomes of patients with COVID-19 in Wuhan, China. Methods: We performed a single center, retrospective case series study in 221 patients with laboratory confirmed SARS-CoV-2 pneumonia at a university hospital, including 55 severe patients and 166 non-severe patients, from January 2, 2020 to February 10, 2020. Results: Of the 221 patients with COVID-19, the median age was 55.0 years and 48.9% were male and only 8 (3.6%) patients had a history of exposure to the Huanan Seafood Market. Compared to the non-severe pneumonia patients, the median age of the severe patients was significantly older, and they were more likely to have chronic comorbidities. Most common symptoms in severe patients were high fever, anorexia and dyspnea. On admission, 33.0% patients showed leukopenia and 73.8% showed lymphopenia. In addition, the severe patients suffered a higher rate of co-infections with bacteria or fungus and they were more likely to developing complications. As of February 15, 2020, 19.0% patients had been discharged and 5.4% patients died. 80% of severe cases received ICU (intensive care unit) care, and 52.3% of them transferred to the general wards due to relieved symptoms, and the mortality rate of severe patients in ICU was 20.5%. Conclusions: Patients with elder age, chronic comorbidities, blood leukocyte/lymphocyte count, procalcitonin level, co-infection and severe complications might increase the risk of poor clinical outcomes.

Obstructive sleep apnoea, a syndrome that leads to recurrent intermittent hypoxia, is associated with insulin resistance in obese individuals, but the mechanisms underlying this association remain unknown. We utilized a mouse model to examine the effects of intermittent hypoxia on insulin resistance in lean C57BL/6J mice and leptin‐deficient obese (C57BL/6J−Lepob) mice. In lean mice, exposure to intermittent hypoxia for 5 days (short term) resulted in a decrease in fasting blood glucose levels (from 173 ± 11 mg dl−1 on day 0 to 138 ± 10 mg dl−1 on day 5, P < 0.01), improvement in glucose tolerance without a change in serum insulin levels and an increase in serum leptin levels in comparison with control (2.6 ± 0.3 vs. 1.7 ± 0.2 ng ml−1, P < 0.05). Microarray mRNA analysis of adipose tissue revealed that leptin was the only upregulated gene affecting glucose uptake. In obese mice, short‐term intermittent hypoxia led to a decrease in blood glucose levels accompanied by a 607 ± 136 % (P < 0.01) increase in serum insulin levels. This increase in insulin secretion after 5 days of intermittent hypoxia was completely abolished by prior leptin infusion. Obese mice exposed to intermittent hypoxia for 12 weeks (long term) developed a time‐dependent increase in fasting serum insulin levels (from 3.6 ± 1.1 ng ml−1 at baseline to 9.8 ± 1.8 ng ml−1 at week 12, P < 0.001) and worsening glucose tolerance, consistent with an increase in insulin resistance. We conclude that the increase in insulin resistance in response to intermittent hypoxia is dependent on the disruption of leptin pathways.

We present Deeply Supervised Object Detector (DSOD), a framework that can learn object detectors from scratch. State-of-the-art object objectors rely heavily on the offthe-shelf networks pre-trained on large-scale classification datasets like ImageNet, which incurs learning bias due to the difference on both the loss functions and the category distributions between classification and detection tasks. Model fine-tuning for the detection task could alleviate this bias to some extent but not fundamentally. Besides, transferring pre-trained models from classification to detection between discrepant domains is even more difficult (e.g. RGB to depth images). A better solution to tackle these two critical problems is to train object detectors from scratch, which motivates our proposed DSOD. Previous efforts in this direction mostly failed due to much more complicated loss functions and limited training data in object detection. In DSOD, we contribute a set of design principles for training object detectors from scratch. One of the key findings is that deep supervision, enabled by dense layer-wise connections, plays a critical role in learning a good detector. Combining with several other principles, we develop DSOD following the single-shot detection (SSD) framework. Experiments on PASCAL VOC 2007, 2012 and MS COCO datasets demonstrate that DSOD can achieve better results than the state-of-the-art solutions with much more compact models. For instance, DSOD outperforms SSD on all three benchmarks with real-time detection speed, while requires only 1/2 parameters to SSD and 1/10 parameters to Faster RCNN. Our code and models are available at: https://github.com/szq0214/DSOD.