Qihoo 360 (China)

To summarize and analyze the characteristics and treatment strategies of post-traumatic lower limb deformity based on QIN Sihe Orthopaedic Surgery Database. A clinical data of 837 patients with post-traumatic lower limb deformities treated by orthopaedic surgery between May 25, 1978 and December 31, 2020 in QIN Sihe Orthopaedic Surgery Database were analyzed retrospectively. The information of the patient's gender, age at the time of surgery, region of origin, cause of trauma, deformity side, orthopedic surgery related information (operation time, location, type, and fixation method after operation) were summarized and analyzed. All patients came from 32 provinces, municipalities, autonomous regions, and Taiwan in China. Among them, 551 cases (65.83%) were male and 286 cases (34.17%) were female. The age of the patients at the time of surgery was 3-84 years old, with an average of 27.6 years old, and the most patients were 16-45 years old (559 cases, 66.78%). The main cause of trauma was traffic accident injury (639 cases, 76.34%). The deformity mainly involved unilateral limbs, including 394 cases (47.07%) on the left side and 376 cases (44.92%) on the right side. The most patients were admitted between 2008 and 2017, accounting for 53.05% (444/837). All patients were operated on one or more sites (1 048 sites), among which ankle and toe surgery were the most, accounting for 48.38% (507/1 048). The patients received 1204 surgeries including tendon lengthening and soft tissue contracture release, et al. Orthopedic surgery combined with bone external fixation was used in 624 cases (467 cases of Ilizarov external fixation and 157 cases of combined external fixation), and plaster or brace external fixation was used in 213 cases. Post-traumatic lower extremity deformity patients have a large proportion of males, with a wide geographical distribution, involving various parts of the lower extremities, and most commonly in the foot and ankle. Orthopedic surgery combined with bone external fixation (Ilizarov technique) is the main methods for correction and functional reconstruction of post-traumatic lower limb deformity.

With The development of Internet, the globalization of information and the increase of digital multimedia communication services, satellite communication is developing rapidly because of its unique advantages. Low-orbit satellite communications are becoming more and more important. Most of the existing automatic tracking schemes drive the tracking system based on the beacon signal or forwarded pilot signal transmitted by the received satellite. The antenna is aligned with the automatic tracking satellite. This system combines satellite program tracking. Theory, proposed a low-orbit satellite automatic tracking system design based on the SGP4 model, and tried to calculate the satellite position through the SGP4 model. Then control the antenna to point to the target satellite to realize the automatic tracking of low-orbit satellites. Experimental results show that the system can achieve automatic tracking of low-orbit satellites within a small error range, and has good stability. Compared with traditional satellite ground stations, the system achieves low-cost and miniaturized automatic tracking of low-orbit satellites.

Human motion prediction, which aims to predict future human poses given past poses, has recently seen increased interest. Many recent approaches are based on Recurrent Neural Networks (RNN) which model human poses with exponential maps. These approaches neglect the pose velocity as well as temporal relation of different poses, and tend to converge to the mean pose or fail to generate natural-looking poses. We therefore propose a novel Position-Velocity Recurrent Encoder-Decoder (PVRED) for human motion prediction, which makes full use of pose velocities and temporal positional information. A temporal position embedding method is presented and a Position-Velocity RNN (PVRNN) is proposed. We also emphasize the benefits of quaternion parameterization of poses and design a novel trainable Quaternion Transformation (QT) layer, which is combined with a robust loss function during training. We provide quantitative results for both short-term prediction in the future 0.5 seconds and long-term prediction in the future 0.5 to 1 seconds. Experiments on several benchmarks show that our approach considerably outperforms the state-of-the-art methods. In addition, qualitative visualizations in the future 4 seconds show that our approach could predict future human-like and meaningful poses in very long time horizons. Code is publicly available on GitHub: https://github.com/hongsong-wang/PVRNN.

Deep learning is data-hungry, and generally its performance highly depends on the amount of training data. Multiple parties can obtain better models by sharing their data and train models collaboratively. To privacy concerns, sensitive raw data of each entity can not be shared directly. In this paper, we propose a data sharing mechanism called ALRS (for Adversarial Latent Representation Sharing) that shares data representations rather than raw data, and applies adversarial example noise to protect shared representations against model inversion attacks, and achieve a balance between privacy and utility. Compared with prior collaborative learning works, ALRS requires no centralized control. We evaluate ALRS in different contexts, and the results demonstrate that our mechanism is effective against reconstruction and feature extraction attacks, while maintaining the utility of models at the same time.

Human motion prediction, which aims to predict future human poses given past poses, has recently seen increased interest. Many recent approaches are based on Recurrent Neural Networks (RNN) which model human poses with exponential maps. These approaches neglect the pose velocity as well as temporal relation of different poses, and tend to converge to the mean pose or fail to generate natural-looking poses. We therefore propose a novel Position-Velocity Recurrent Encoder-Decoder (PVRED) for human motion prediction, which makes full use of pose velocities and temporal positional information. A temporal position embedding method is presented and a Position-Velocity RNN (PVRNN) is proposed. We also emphasize the benefits of quaternion parameterization of poses and design a novel trainable Quaternion Transformation (QT) layer, which is combined with a robust loss function during training. We provide quantitative results for both short-term prediction in the future 0.5 seconds and long-term prediction in the future 0.5 to 1 seconds. Experiments on several benchmarks show that our approach considerably outperforms the state-of-the-art methods. In addition, qualitative visualizations in the future 4 seconds show that our approach could predict future human-like and meaningful poses in very long time horizons. Code is publicly available on GitHub: \textcolor{red}{https://github.com/hongsong-wang/PVRNN}.

Traditional recommendation algorithms on Collaborative Filtering (CF) mainly focus on the rating prediction with explicit ratings, and cannot be applied to the top-N recommendation with implicit feedbacks. To tackle this problem, we propose a new collaborative filtering approach namely Maximize MAP with Matrix Factorization (MFMAP). In addition, in order to solve the problem of non-smoothing loss function in learning to rank (LTR) algorithm based on pairwise, we also propose a smooth MAP measure which can be easily implemented by standard optimization approaches. We perform experiments on three different datasets, and the experimental results show that the performance of MFMAP is significantly better than other recommendation approaches.

In this paper, we present a large-scale analysis about an emerging new type of domain-name fraud, which we call levelsquatting. Unlike existing frauds that impersonate well-known brand names (like google.com) by using similar second-level domain names, adversaries here embed brand name in the subdomain section, deceiving users especially mobile users who do not pay attention to the entire domain names.

Autonomous vehicles rely on sensors to measure road condition and make driving decisions, and their safety relies heavily on the reliability of these sensors. Out of all obstacle detection sensors, ultrasonic sensors have the largest market share and are expected to be increasingly installed on automobiles. Such sensors discover obstacles by emitting ultrasounds and analyzing their reflections. By exploiting the built-in vulnerabilities of sensors, we designed random spoofing, adaptive spoofing, and jamming attacks on ultrasonic sensors, and we managed to trick a vehicle to stop when it should keep moving, and let it fail to stop when it should. We validate our attacks on stand-alone sensors and moving vehicles, including a Tesla Model S with the “Autopilot” system. The results show that the attacks cause blindness and malfunction of not only sensors but also autonomous vehicles, which can lead to collisions. To enhance the security of ultrasonic sensors and autonomous vehicles, we propose two defense strategies, single-sensor-based physical shift authentication that verifies signals on the physical level, and multiple sensor consistency check that employs multiple sensors to verify signals on the system level. Our experiments on real sensors and MATLAB simulation reveal the validity of both schemes.

Style synthesis attracts great interests recently, while few works focus on its dual problem "style separation". In this paper, we propose the Style Separation and Synthesis Generative Adversarial Network (S3-GAN) to simultaneously implement style separation and style synthesis on object photographs of specific categories. Based on the assumption that the object photographs lie on a manifold, and the contents and styles are independent, we employ S3-GAN to build mappings between the manifold and a latent vector space for separating and synthesizing the contents and styles. The S3-GAN consists of an encoder network, a generator network, and an adversarial network. The encoder network performs style separation by mapping an object photograph to a latent vector. Two halves of the latent vector represent the content and style, respectively. The generator network performs style synthesis by taking a concatenated vector as input. The concatenated vector contains the style half vector of the style target image and the content half vector of the content target image. Once obtaining the images from the generator network, an adversarial network is imposed to generate more photo-realistic images. Experiments on CelebA and UT Zappos 50K datasets demonstrate that the S3-GAN has the capacity of style separation and synthesis simultaneously, and could capture various styles in a single model.