Recently, with the advent of deep convolutional neural networks (DCNN), the improvements in visual saliency prediction research are impressive. One possible direction to approach the next improvement is to fully characterize the multiscale saliency-influential factors with a computationally-friendly module in DCNN architectures. In this work, we proposed an end-to-end dilated inception network (DINet) for visual saliency prediction. It captures multi-scale contextual features effectively with very limited extra parameters. Instead of utilizing parallel standard convolutions with different kernel sizes as the existing inception module, our proposed dilated inception module (DIM) uses parallel dilated convolutions with different dilation rates which can significantly reduce the computation load while enriching the diversity of receptive fields in feature maps. Moreover, the performance of our saliency model is further improved by using a set of linear normalization-based probability distribution distance metrics as loss functions. As such, we can formulate saliency prediction as a probability distribution prediction task for global saliency inference instead of a typical pixel-wise regression problem. Experimental results on several challenging saliency benchmark datasets demonstrate that our DINet with proposed loss functions can achieve state-of-the-art performance with shorter inference time.

The hallmark of mechanosensory hair cells is the stereocilia, where mechanical stimuli are converted into electrical signals. These delicate stereocilia are susceptible to acoustic trauma and ototoxic drugs. While hair cells in lower vertebrates and the mammalian vestibular system can spontaneously regenerate lost stereocilia, mammalian cochlear hair cells no longer retain this capability. We explored the possibility of regenerating stereocilia in the noise-deafened guinea pig cochlea by cochlear inoculation of a viral vector carrying Atoh1, a gene critical for hair cell differentiation. Exposure to simulated gunfire resulted in a 60–70 dB hearing loss and extensive damage and loss of stereocilia bundles of both inner and outer hair cells along the entire cochlear length. However, most injured hair cells remained in the organ of Corti for up to 10 days after the trauma. A viral vector carrying an EGFP-labeled Atoh1 gene was inoculated into the cochlea through the round window on the seventh day after noise exposure. Auditory brainstem response measured one month after inoculation showed that hearing thresholds were substantially improved. Scanning electron microscopy revealed that the damaged/lost stereocilia bundles were repaired or regenerated after Atoh1 treatment, suggesting that Atoh1 was able to induce repair/regeneration of the damaged or lost stereocilia. Therefore, our studies revealed a new role of Atoh1 as a gene critical for promoting repair/regeneration of stereocilia and maintaining injured hair cells in the adult mammal cochlea. Atoh1-based gene therapy, therefore, has the potential to treat noise-induced hearing loss if the treatment is carried out before hair cells die.

We reported a scalable and modular method to prepare a new type of sandwich-structured graphene-based nanohybrid paper and explore its practical application as high-performance electrode in flexible supercapacitor. The freestanding and flexible graphene paper was firstly fabricated by highly reproducible printing technique and bubbling delamination method, by which the area and thickness of the graphene paper can be freely adjusted in a wide range. The as-prepared graphene paper possesses a collection of unique properties of highly electrical conductivity (340 S cm−1), light weight (1 mg cm−2) and excellent mechanical properties. In order to improve its supercapacitive properties, we have prepared a unique sandwich-structured graphene/polyaniline/graphene paper by in situ electropolymerization of porous polyaniline nanomaterials on graphene paper, followed by wrapping an ultrathin graphene layer on its surface. This unique design strategy not only circumvents the low energy storage capacity resulting from the double-layer capacitor of graphene paper, but also enhances the rate performance and cycling stability of porous polyaniline. The as-obtained all-solid-state symmetric supercapacitor exhibits high energy density, high power density, excellent cycling stability and exceptional mechanical flexibility, demonstrative of its extensive potential applications for flexible energy-related devices and wearable electronics.

Abstract-Design patterns are generic design solutions that can be applied and composed in different applications where patternrelated information is generally implicit in the Unified Modeling Language (UML) diagrams of the applications. It is unclear in which pattern instances each modeling element, such as class, attribute, and operation, participates. It is hard for a designer to find the design patterns used in an application design. Consequently, the benefits of design patterns are compromised because designers cannot communicate with each other in terms of the design patterns they used and their design decisions and trade-offs. In this paper, we present a UML profile that defines new stereotypes, tagged values, and constraints for tracing design patterns in UML diagrams. These new stereotypes and tagged values are attached to a modeling element to explicitly represent the role the modeling element plays in a design pattern so that the user can identify the pattern in a UML diagram. Based on this profile, we also develop a Web service (tool) for explicitly visualizing design patterns in UML diagrams. With this service, users are able to visualize design patterns in their applications and compositions because pattern-related information can be dynamically displayed. A real-world case study and a comparative experiment with existing approaches are conducted to evaluate our approach.

A novel framework named Markov Clustering Network (MCN) is proposed for fast and robust scene text detection. MCN predicts instance-level bounding boxes by firstly converting an image into a Stochastic Flow Graph (SFG) and then performing Markov Clustering on this graph. Our method can detect text objects with arbitrary size and orientation without prior knowledge of object size. The stochastic flow graph encode objects' local correlation and semantic information. An object is modeled as strongly connected nodes, which allows flexible bottom-up detection for scale-varying and rotated objects. MCN generates bounding boxes without using Non-Maximum Suppression, and it can be fully parallelized on GPUs. The evaluation on public benchmarks shows that our method outperforms the existing methods by a large margin in detecting multioriented text objects. MCN achieves new state-of-art performance on challenging MSRA-TD500 dataset with precision of 0.88, recall of 0.79 and F-score of 0.83. Also, MCN achieves realtime inference with frame rate of 34 FPS, which is 1.5× speedup when compared with the fastest scene text detection algorithm.