Showing papers by "Hong Kong Polytechnic University published in 2018"

FFDNet: Toward a Fast and Flexible Solution for CNN-Based Image Denoising

Abstract: Due to the fast inference and good performance, discriminative learning methods have been widely studied in image denoising. However, these methods mostly learn a specific model for each noise level, and require multiple models for denoising images with different noise levels. They also lack flexibility to deal with spatially variant noise, limiting their applications in practical denoising. To address these issues, we present a fast and flexible denoising convolutional neural network, namely FFDNet, with a tunable noise level map as the input. The proposed FFDNet works on downsampled sub-images, achieving a good trade-off between inference speed and denoising performance. In contrast to the existing discriminative denoisers, FFDNet enjoys several desirable properties, including: 1) the ability to handle a wide range of noise levels (i.e., [0, 75]) effectively with a single network; 2) the ability to remove spatially variant noise by specifying a non-uniform noise level map; and 3) faster speed than benchmark BM3D even on CPU without sacrificing denoising performance. Extensive experiments on synthetic and real noisy images are conducted to evaluate FFDNet in comparison with state-of-the-art denoisers. The results show that FFDNet is effective and efficient, making it highly attractive for practical denoising applications.

Next-generation organic photovoltaics based on non-fullerene acceptors

Abstract: Over the past three years, a particularly exciting and active area of research within the field of organic photovoltaics has been the use of non-fullerene acceptors (NFAs). Compared with fullerene acceptors, NFAs possess significant advantages including tunability of bandgaps, energy levels, planarity and crystallinity. To date, NFA solar cells have not only achieved impressive power conversion efficiencies of ~13–14%, but have also shown excellent stability compared with traditional fullerene acceptor solar cells. This Review highlights recent progress on single-junction and tandem NFA solar cells and research directions to achieve even higher efficiencies of 15–20% using NFA-based organic photovoltaics are also proposed.

Deeper Insights into Graph Convolutional Networks for Semi-Supervised Learning

Abstract: Many interesting problems in machine learning are being revisited with new deep learning tools. For graph-based semi-supervised learning, a recent important development is graph convolutional networks (GCNs), which nicely integrate local vertex features and graph topology in the convolutional layers. Although the GCN model compares favorably with other state-of-the-art methods, its mechanisms are not clear and it still requires considerable amount of labeled data for validation and model selection. In this paper, we develop deeper insights into the GCN model and address its fundamental limits. First, we show that the graph convolution of the GCN model is actually a special form of Laplacian smoothing, which is the key reason why GCNs work, but it also brings potential concerns of over-smoothing with many convolutional layers. Second, to overcome the limits of the GCN model with shallow architectures, we propose both co-training and self-training approaches to train GCNs. Our approaches significantly improve GCNs in learning with very few labels, and exempt them from requiring additional labels for validation. Extensive experiments on benchmarks have verified our theory and proposals.

All-inorganic perovskite nanocrystal scintillators

Abstract: The rising demand for radiation detection materials in many applications has led to extensive research on scintillators1–3. The ability of a scintillator to absorb high-energy (kiloelectronvolt-scale) X-ray photons and convert the absorbed energy into low-energy visible photons is critical for applications in radiation exposure monitoring, security inspection, X-ray astronomy and medical radiography4,5. However, conventional scintillators are generally synthesized by crystallization at a high temperature and their radioluminescence is difficult to tune across the visible spectrum. Here we describe experimental investigations of a series of all-inorganic perovskite nanocrystals comprising caesium and lead atoms and their response to X-ray irradiation. These nanocrystal scintillators exhibit strong X-ray absorption and intense radioluminescence at visible wavelengths. Unlike bulk inorganic scintillators, these perovskite nanomaterials are solution-processable at a relatively low temperature and can generate X-ray-induced emissions that are easily tunable across the visible spectrum by tailoring the anionic component of colloidal precursors during their synthesis. These features allow the fabrication of flexible and highly sensitive X-ray detectors with a detection limit of 13 nanograys per second, which is about 400 times lower than typical medical imaging doses. We show that these colour-tunable perovskite nanocrystal scintillators can provide a convenient visualization tool for X-ray radiography, as the associated image can be directly recorded by standard digital cameras. We also demonstrate their direct integration with commercial flat-panel imagers and their utility in examining electronic circuit boards under low-dose X-ray illumination. All-inorganic perovskite nanocrystals containing caesium and lead provide low-cost, flexible and solution-processable scintillators that are highly sensitive to X-ray irradiation and emit radioluminescence that is colour-tunable across the visible spectrum.

Multicomponent intermetallic nanoparticles and superb mechanical behaviors of complex alloys

Abstract: Alloy design based on single-principal-element systems has approached its limit for performance enhancements. A substantial increase in strength up to gigapascal levels typically causes the premature failure of materials with reduced ductility. Here, we report a strategy to break this trade-off by controllably introducing high-density ductile multicomponent intermetallic nanoparticles (MCINPs) in complex alloy systems. Distinct from the intermetallic-induced embrittlement under conventional wisdom, such MCINP-strengthened alloys exhibit superior strengths of 1.5 gigapascals and ductility as high as 50% in tension at ambient temperature. The plastic instability, a major concern for high-strength materials, can be completely eliminated by generating a distinctive multistage work-hardening behavior, resulting from pronounced dislocation activities and deformation-induced microbands. This MCINP strategy offers a paradigm to develop next-generation materials for structural applications.

Anchoring zero valence single atoms of nickel and iron on graphdiyne for hydrogen evolution

Abstract: Electrocatalysis by atomic catalysts is a major focus of chemical and energy conversion effort. Although transition-metal-based bulk electrocatalysts for electrochemical application on energy conversion processes have been reported frequently, anchoring the stable transition-metal atoms (e.g. nickel and iron) still remains a practical challenge. Here we report a strategy for fabrication of ACs comprising only isolated nickel/iron atoms anchored on graphdiyne. Our findings identify the very narrow size distributions of both nickel (1.23 A) and iron (1.02 A), typical sizes of single-atom nickel and iron. The precision of this method motivates us to develop a general approach in the field of single-atom transition-metal catalysis. Such atomic catalysts have high catalytic activity and stability for hydrogen evolution reactions.

Learning a Deep Single Image Contrast Enhancer from Multi-Exposure Images

Abstract: Due to the poor lighting condition and limited dynamic range of digital imaging devices, the recorded images are often under-/over-exposed and with low contrast. Most of previous single image contrast enhancement (SICE) methods adjust the tone curve to correct the contrast of an input image. Those methods, however, often fail in revealing image details because of the limited information in a single image. On the other hand, the SICE task can be better accomplished if we can learn extra information from appropriately collected training data. In this paper, we propose to use the convolutional neural network (CNN) to train a SICE enhancer. One key issue is how to construct a training data set of low-contrast and high-contrast image pairs for end-to-end CNN learning. To this end, we build a large-scale multi-exposure image data set, which contains 589 elaborately selected high-resolution multi-exposure sequences with 4,413 images. Thirteen representative multi-exposure image fusion and stack-based high dynamic range imaging algorithms are employed to generate the contrast enhanced images for each sequence, and subjective experiments are conducted to screen the best quality one as the reference image of each scene. With the constructed data set, a CNN can be easily trained as the SICE enhancer to improve the contrast of an under-/over-exposure image. Experimental results demonstrate the advantages of our method over existing SICE methods with a significant margin.

RippleNet: Propagating User Preferences on the Knowledge Graph for Recommender Systems

Abstract: To address the sparsity and cold start problem of collaborative filtering, researchers usually make use of side information, such as social networks or item attributes, to improve recommendation performance. This paper considers the knowledge graph as the source of side information. To address the limitations of existing embedding-based and path-based methods for knowledge-graph-aware recommendation, we propose RippleNet, an end-to-end framework that naturally incorporates the knowledge graph into recommender systems. Similar to actual ripples propagating on the water, RippleNet stimulates the propagation of user preferences over the set of knowledge entities by automatically and iteratively extending a user's potential interests along links in the knowledge graph. The multiple "ripples" activated by a user's historically clicked items are thus superposed to form the preference distribution of the user with respect to a candidate item, which could be used for predicting the final clicking probability. Through extensive experiments on real-world datasets, we demonstrate that RippleNet achieves substantial gains in a variety of scenarios, including movie, book and news recommendation, over several state-of-the-art baselines.

A fatal outbreak of ST11 carbapenem-resistant hypervirulent Klebsiella pneumoniae in a Chinese hospital: a molecular epidemiological study

Abstract: Summary Background Hypervirulent Klebsiella pneumoniae strains often cause life-threatening community-acquired infections in young and healthy hosts, but are usually sensitive to antibiotics. In this study, we investigated a fatal outbreak of ventilator-associated pneumonia caused by a new emerging hypervirulent K pneumoniae strain. Methods The outbreak occurred in the integrated intensive care unit of a new branch of the Second Affiliated Hospital of Zhejiang University (Hangzhou, China). We collected 21 carbapenem-resistant K pneumoniae strains from five patients and characterised these strains for their antimicrobial susceptibility, multilocus sequence types, and genetic relatedness using VITEK-2 compact system, multilocus sequence typing, and whole genome sequencing. We selected one representative isolate from each patient to establish the virulence potential using a human neutrophil assay and Galleria mellonella model and to establish the genetic basis of their hypervirulence phenotype. Findings All five patients had undergone surgery for multiple trauma and subsequently received mechanical ventilation. The patients were aged 53–73 years and were admitted to the intensive care unit between late February and April, 2016. They all had severe pneumonia, carbapenem-resistant K pneumoniae infections, and poor responses to antibiotic treatment and died due to severe lung infection, multiorgan failure, or septic shock. All five representative carbapenem-resistant K pneumoniae strains belonged to the ST11 type, which is the most prevalent carbapenem-resistant K pneumoniae type in China, and originated from the same clone. The strains were positive on the string test, had survival of about 80% after 1 h incubation in human neutrophils, and killed 100% of wax moth larvae ( G mellonella ) inoculated with 1 × 10 6 colony-forming units of the specimens within 24 h, suggesting that they were hypervirulent K pneumoniae . Genomic analyses showed that the emergence of these ST11 carbapenem-resistant hypervirulent K pneumoniae strains was due to the acquisition of a roughly 170 kbp pLVPK-like virulence plasmid by classic ST11 carbapenem-resistant K pneumoniae strains. We also detected these strains in specimens collected in other regions of China. Interpretation The ST11 carbapenem-resistant hypervirulent K pneumoniae strains pose a substantial threat to human health because they are simultaneously hypervirulent, multidrug resistant, and highly transmissible. Control measures should be implemented to prevent further dissemination of such organisms in the hospital setting and the community. Funding Chinese National Key Basic Research and Development Program and Collaborative Research Fund of Hong Kong Research Grant Council.

Learning Spatial-Temporal Regularized Correlation Filters for Visual Tracking

Abstract: Discriminative Correlation Filters (DCF) are efficient in visual tracking but suffer from unwanted boundary effects. Spatially Regularized DCF (SRDCF) has been suggested to resolve this issue by enforcing spatial penalty on DCF coefficients, which, inevitably, improves the tracking performance at the price of increasing complexity. To tackle online updating, SRDCF formulates its model on multiple training images, further adding difficulties in improving efficiency. In this work, by introducing temporal regularization to SRDCF with single sample, we present our spatial-temporal regularized correlation filters (STRCF). The STRCF formulation can not only serve as a reasonable approximation to SRDCF with multiple training samples, but also provide a more robust appearance model than SRDCF in the case of large appearance variations. Besides, it can be efficiently solved via the alternating direction method of multipliers (ADMM). By incorporating both temporal and spatial regularization, our STRCF can handle boundary effects without much loss in efficiency and achieve superior performance over SRDCF in terms of accuracy and speed. Compared with SRDCF, STRCF with hand-crafted features provides a 5A— speedup and achieves a gain of 5.4% and 3.6% AUC score on OTB-2015 and Temple-Color, respectively. Moreover, STRCF with deep features also performs favorably against state-of-the-art trackers and achieves an AUC score of 68.3% on OTB-2015.

Severe Surface Ozone Pollution in China: A Global Perspective

Abstract: The nationwide extent of surface ozone pollution in China and its comparison to the global ozone distribution have not been recognized because of the scarcity of Chinese monitoring sites before 2012. Here we address this issue by using the latest 5 year (2013–2017) surface ozone measurements from the Chinese monitoring network, combined with the recent Tropospheric Ozone Assessment Report (TOAR) database for other industrialized regions such as Japan, South Korea, Europe, and the United States (JKEU). We use various human health and vegetation exposure metrics. We find that although the median ozone values are comparable between Chinese and JKEU cities, the magnitude and frequency of high-ozone events are much larger in China. The national warm-season (April–September) fourth highest daily maximum 8 h average (4MDA8) ozone level (86.0 ppb) and the number of days with MDA8 values of >70 ppb (NDGT70, 29.7 days) in China are 6.3–30% (range of regional mean differences) and 93–575% higher, respectively, than ...

Construction and demolition waste management in China through the 3R principle

Abstract: Construction and demolition waste (CDW) accounts for 30% to 40% of the total amount of waste in China. CDW is usually randomly dumped or disposed in landfills and the average recycling rate of CDW in China is only about 5%. Considering there is big challenge in adoption of circular economy in CDW industry in China while related research is still limited, we conduct the CDW management analysis through 3R principle. Existing policies and management situations were investigated and analyzed based on the reduction, reuse and recycle principles. Results reveal that primary barriers of reducing CDW in China include lack of building design standard for reducing CDW, low cost for CDW disposal and inappropriate urban planning. Barriers to reuse CDW include lack of guidance for effective CDW collection and sorting, lack of knowledge and standard for reused CDW, and an under-developed market for reused CDW. As for recycling of CDW, key challenges are identified as ineffective management system, immature recycling technology, under-developed market for recycled CDW products and immature recycling market operation. Proposals to improve the current situation based on 3R principle are also proposed, including designing effective circular economy model, reinforcing the source control of CDW, adopting innovative technologies and market models, and implementing targeted economic incentives.

The link between Industry 4.0 and lean manufacturing: mapping current research and establishing a research agenda

Abstract: In recent years, Industry 4.0 has emerged as one of the most discussed concepts and has gained significant popularity in both academia and the industrial sector. Both Industry 4.0 and lean manufact...

Big Data Analytics in Operations Management

Abstract: Big data analytics is critical in modern operations management (OM). In this study, we first explore the existing big data‐related analytics techniques, and identify their strengths, weaknesses as well as major functionalities. We then discuss various big data analytics strategies to overcome the respective computational and data challenges. After that, we examine the literature and reveal how different types of big data methods (techniques, strategies, and architectures) can be applied to different OM topical areas, namely forecasting, inventory management, revenue management and marketing, transportation management, supply chain management, and risk analysis. We also investigate via case studies the real‐world applications of big data analytics in top branded enterprises. Finally, we conclude the study with a discussion of future research.

Pyramid Dilated Deeper ConvLSTM for Video Salient Object Detection

Abstract: This paper proposes a fast video salient object detection model, based on a novel recurrent network architecture, named Pyramid Dilated Bidirectional ConvLSTM (PDB-ConvLSTM). A Pyramid Dilated Convolution (PDC) module is first designed for simultaneously extracting spatial features at multiple scales. These spatial features are then concatenated and fed into an extended Deeper Bidirectional ConvLSTM (DB-ConvLSTM) to learn spatiotemporal information. Forward and backward ConvLSTM units are placed in two layers and connected in a cascaded way, encouraging information flow between the bi-directional streams and leading to deeper feature extraction. We further augment DB-ConvLSTM with a PDC-like structure, by adopting several dilated DB-ConvLSTMs to extract multi-scale spatiotemporal information. Extensive experimental results show that our method outperforms previous video saliency models in a large margin, with a real-time speed of 20 fps on a single GPU. With unsupervised video object segmentation as an example application, the proposed model (with a CRF-based post-process) achieves state-of-the-art results on two popular benchmarks, well demonstrating its superior performance and high applicability.

Automatic Pixel-Level Crack Detection and Measurement Using Fully Convolutional Network

Abstract: The spatial characteristics of cracks are significant indicators to assess and evaluate the health of existing buildings and infrastructures However, the current manual crack description

Electronic synapses made of layered two-dimensional materials

Abstract: Neuromorphic computing systems, which use electronic synapses and neurons, could overcome the energy and throughput limitations of today’s computing architectures. However, electronic devices that can accurately emulate the short- and long-term plasticity learning rules of biological synapses remain limited. Here, we show that multilayer hexagonal boron nitride (h-BN) can be used as a resistive switching medium to fabricate high-performance electronic synapses. The devices can operate in a volatile or non-volatile regime, enabling the emulation of a range of synaptic-like behaviour, including both short- and long-term plasticity. The behaviour results from a resistive switching mechanism in the h-BN stack, based on the generation of boron vacancies that can be filled by metallic ions from the adjacent electrodes. The power consumption in standby and per transition can reach as low as 0.1 fW and 600 pW, respectively, and with switching times reaching less than 10 ns, demonstrating their potential for use in energy-efficient brain-like computing. Vertically structured electronic synapses, which exhibit both short- and long-term plasticity, can be created using layered two-dimensional hexagonal boron nitride.

R³Net: Recurrent Residual Refinement Network for Saliency Detection

Abstract: Saliency detection is a fundamental yet challenging task in computer vision, aiming at highlighting the most visually distinctive objects in an image. We propose a novel recurrent residual refinement network (R^3Net) equipped with residual refinement blocks (RRBs) to more accurately detect salient regions of an input image. Our RRBs learn the residual between the intermediate saliency prediction and the ground truth by alternatively leveraging the low-level integrated features and the high-level integrated features of a fully convolutional network (FCN). While the low-level integrated features are capable of capturing more saliency details, the high-level integrated features can reduce non-salient regions in the intermediate prediction. Furthermore, the RRBs can obtain complementary saliency information of the intermediate prediction, and add the residual into the intermediate prediction to refine the saliency maps. We evaluate the proposed R^3Net on five widely-used saliency detection benchmarks by comparing it with 16 state-of-the-art saliency detectors. Experimental results show that our network outperforms our competitors in all the benchmark datasets.

Many analysts, one dataset: Making transparent how variations in analytical choices affect results

Abstract: Twenty-nine teams involving 61 analysts used the same data set to address the same research question: whether soccer referees are more likely to give red cards to dark-skin-toned players than to light-skin-toned players. Analytic approaches varied widely across the teams, and the estimated effect sizes ranged from 0.89 to 2.93 (Mdn = 1.31) in odds-ratio units. Twenty teams (69%) found a statistically significant positive effect, and 9 teams (31%) did not observe a significant relationship. Overall, the 29 different analyses used 21 unique combinations of covariates. Neither analysts’ prior beliefs about the effect of interest nor their level of expertise readily explained the variation in the outcomes of the analyses. Peer ratings of the quality of the analyses also did not account for the variability. These findings suggest that significant variation in the results of analyses of complex data may be difficult to avoid, even by experts with honest intentions. Crowdsourcing data analysis, a strategy in which numerous research teams are recruited to simultaneously investigate the same research question, makes transparent how defensible, yet subjective, analytic choices influence research results.

Scalable 2D Hierarchical Porous Carbon Nanosheets for Flexible Supercapacitors with Ultrahigh Energy Density.

Abstract: 2D carbon nanomaterials such as graphene and its derivatives, have gained tremendous research interests in energy storage because of their high capacitance and chemical stability. However, scalable synthesis of ultrathin carbon nanosheets with well-defined pore architectures remains a great challenge. Herein, the first synthesis of 2D hierarchical porous carbon nanosheets (2D-HPCs) with rich nitrogen dopants is reported, which is prepared with high scalability through a rapid polymerization of a nitrogen-containing thermoset and a subsequent one-step pyrolysis and activation into 2D porous nanosheets. 2D-HPCs, which are typically 1.5 nm thick and 1-3 µm wide, show a high surface area (2406 m2 g-1 ) and with hierarchical micro-, meso-, and macropores. This 2D and hierarchical porous structure leads to robust flexibility and good energy-storage capability, being 139 Wh kg-1 for a symmetric supercapacitor. Flexible supercapacitor devices fabricated by these 2D-HPCs also present an ultrahigh volumetric energy density of 8.4 mWh cm-3 at a power density of 24.9 mW cm-3 , which is retained at 80% even when the power density is increased by 20-fold. The devices show very high electrochemical life (96% retention after 10000 charge/discharge cycles) and excellent mechanical flexibility.

A critical review of the use of virtual reality in construction engineering education and training

Abstract: Virtual Reality (VR) has been rapidly recognized and implemented in construction engineering education and training (CEET) in recent years due to its benefits of providing an engaging and immersive environment. The objective of this review is to critically collect and analyze the VR applications in CEET, aiming at all VR-related journal papers published from 1997 to 2017. The review follows a three-stage analysis on VR technologies, applications and future directions through a systematic analysis. It is found that the VR technologies adopted for CEET evolve over time, from desktop-based VR, immersive VR, 3D game-based VR, to Building Information Modelling (BIM)-enabled VR. A sibling technology, Augmented Reality (AR), for CEET adoptions has also emerged in recent years. These technologies have been applied in architecture and design visualization, construction health and safety training, equipment and operational task training, as well as structural analysis. Future research directions, including the integration of VR with emerging education paradigms and visualization technologies, have also been provided. The findings are useful for both researchers and educators to usefully integrate VR in their education and training programs to improve the training performance.