The goal of incremental Few-shot Semantic Segmentation (iFSS) is to extend pre-trained segmentation models to new classes via few annotated images without access to old training data. During incrementally learning novel classes, the data distribution of old classes will be destroyed, leading to catastrophic forgetting. Meanwhile, the novel classes have only few samples, making models impossible to learn the satisfying representations of novel classes. For the iFSS problem, we propose a network called OINet, i.e., the background embedding space Organization and prototype Inherit Network. Specifically, when training base classes, OINet uses multiple classification heads for the background and sets multiple sub-class prototypes to reserve embedding space for the latent novel classes. During incrementally learning novel classes, we propose a strategy to select the sub-class prototypes that best match the current learning novel classes and make the novel classes inherit the selected prototypes' embedding space. This operation allows the novel classes to be registered in the embedding space using few samples without affecting the distribution of the base classes. Results on Pascal-VOC and COCO show that OINet achieves a new state of the art.

Lifelong learning aims to train a model with good performance for new tasks while retaining the capacity of previous tasks. However, some practical scenarios require the system to forget undesirable knowledge due to privacy issues, which is called selective forgetting. The joint task of the two is dubbed Learning with Selective Forgetting (LSF). In this paper, we propose a new framework based on contrastive strategy for LSF. Specifically, for the preserved classes (tasks), we make features extracted from different samples within a same class compacted. And for the deleted classes, we make the features from different samples of a same class dispersed and irregular, i.e., the network does not have any regular response to samples from a specific deleted class as if the network has no training at all. Through maintaining or disturbing the feature distribution, the forgetting and memory of different classes can be or independent of each other. Experiments are conducted on four benchmark datasets, and our method acieves new state-of-the-art.

Deep learning-based information processing consumes long time and requires huge computing resources, especially for dense prediction tasks which require an output for each pixel, like semantic segmentation and salient object detection. There are mainly two challenges for quantization of dense prediction tasks. Firstly, directly applying the upsampling operation that dense prediction tasks require is extremely crude and causes unacceptable accuracy reduction. Secondly, the complex structure of dense prediction networks means it is difficult to maintain a fast speed as well as a high accuracy when performing quantization. In this paper, we propose an effective upsampling method and an efficient attention computation strategy to transfer the success of the binary neural networks (BNN) from single prediction tasks to dense prediction tasks. Firstly, we design a simple and robust multi-branch parallel upsampling structure to achieve the high accuracy. Then we further optimize the attention method which plays an important role in segmentation but has huge computation complexity. Our attention method can reduce the computational complexity by a factor of one hundred times but retain the original effect. Experiments on Cityscapes, KITTI road, and ECSSD fully show the effectiveness of our work.

AbstractPhotocatalytic hydrogen production via solar energy is considered one of the most strategic ways to produce renewable energy. However, expensive Pt is normally used as the cocatalyst during photocatalysis, which prevents commercialization. Therefore, extensive research has been performed to seek abundant and low‐cost alternative catalysts. In this work, MoS2 quantum dots (QDs) synthesized by a hydrothermal method are incorporated with graphitic carbon nitride to form a heterostructure for photocatalytic hydrogen evolution. MoS2 QDs/g‐C3N4 heterostructure containing 5% and 10% MoS2 QDs exhibited a high hydrogen production of 140 and 152 µmol h−1g−1, respectively, demonstrating the potential of MoS2 as an effective economic cocatalyst. Detailed investigations indicate that incorporating MoS2 QDs with carbon nitride to form heterostructure reduces the bandgap, suppresses the recombination, and enhances electron kinetic energy resulting from the anti‐Stoke effect, thus leading to better performance for hydrogen evolution.

Network embedding has facilitated lots of network analytical tasks by representing nodes as low-dimensional vectors. As an extension of convolutional neural networks from Euclidean data to irregular data, Graph Convolutional Networks (GCNs) provide a novel way to learn network representations and have attracted widespread attention currently. Most of the existing GCNs are only applied to unsigned networks. However, networks could have both positive and negative links in the real world, to which the unsigned algorithms are no longer applicable. In this paper, we propose a novel Signed Directed Attention Network model to capture the structural and social theoretical information of signed directed networks comprehensively through an auto-encoder framework. In the encoding block, the information of sign, direction, social theory, and "bridge" edges are encoded into node embeddings by a node fine-grained classification aggregation layer. Besides, a direction parameterization layer is also introduced to convert directions into direction-specific convolutional filters to enhance the node embeddings. In the decoding block, loss functions are designed to model sign, direction, "bridge" edge, and social theory information accordingly and make them complementary to each other to capture the network information fully. Experimental results for the signed link prediction task on several real-world signed directed graphs show that the proposed framework can achieve state-of-the-art performance.

The growth of flexible semiconductor thin films and membranes is highly desirable for the fabrication of next-generation wearable devices. In this work, we have developed a one-step, surface tension-driven method for facile and scalable growth of silver sulfide (Ag2S) membranes with a nanomesh structure. The nanomesh membrane can in principle reach infinite size but only limited by the reactor size, while the thickness is self-limited to ca. 50 nm. In particular, the membrane can be continuously regenerated at the water surface after being transferred for mechanical and electronic tests. The free-standing membrane demonstrates exceptional flexibility and strength, resulting from the nanomesh structure and the intrinsic plasticity of the Ag2S ligaments, as revealed by robust manipulation, nanoindentation tests and a pseudo-in situ tensile test under scanning electron microscope. Bendable electronic resistance-switching devices are fabricated based on the nanomesh membrane.

Rechargeable aqueous zinc-ion batteries (ZIBs) are promising in stationary grid energy storage due to their advantages in safety and cost-effectiveness, and the search for competent cathode materials is one core task in the development of ZIBs. Herein, the authors design a 2D heterostructure combining amorphous vanadium pentoxide and electrochemically produced graphene oxide (EGO) using a fast and scalable spray drying technique. The unique 2D heterostructured xerogel is achieved by controlling the concentration of EGO in the precursor solution. Driven by the improved electrochemical kinetics, the resultant xerogel can deliver an excellent rate capability (334 mAh g-1 at 5 A g-1 ) as well as a high specific capacity (462 mAh g-1 at 0.2 A g-1 ) as the cathode material in ZIB. It is also shown that the coin cell constructed based on spray-dried xerogel can output steady, high energy densities over a broad power density window. This work provides a scalable and cost-effective approach for making high performance electrode materials from cheap sources through existing industrialized materials processing.

Molybdenum trioxide (MoO3) has attracted considerable research interest due to its unique structural and electronic properties. Herein, we report a monolithic and scalable NO2 gas sensor based on centimeter-sized single-crystalline alpha-MoO3 synthesized via vapor phase transport method. A combination of advanced characterization probes were utilized to study morphology, composition and crystalline structures of the as-synthesized crystals. At the optimal operating temperature of 100 ?degrees C, the centimeter-sized alpha-MoO3 single crystal based sensor shows an outstanding sensitivity towards NO2 with a limit of detection as low as 5 ppb, as well as superior selectivity and reversibility. The scalability and the intrinsic sensing response are further demonstrated by microscale sensor devices fabricated on individual exfoliated alpha-MoO3 nanoribbons. Our study presents promising opportunities to develop a high-performance gas sensing platform based on crystalline alpha-MoO3 to enable monolithic, scalable and integrable sensing technologies.

Recently, two-dimensional (2D) materials and their heterostructures have attracted considerable attention in gas sensing applications. In this work, we synthesized 2D MoS2@MoO3 heterostructures through post-sulfurization of α-MoO3 nanoribbons grown via vapor phase transport (VPT) and demonstrated highly sensitive NO2 gas sensors based on the hybrid heterostructures. The morphological, structural, and compositional properties of the MoS2@MoO3 hybrids were studied by a combination of advanced characterization techniques revealing a core-shell structure with the coexistence of 2H-MoS2 multilayers and intermediate molybdenum oxysulfides on the surface of α-MoO3. The MoS2@MoO3 hybrids also exhibit room-temperature ferromagnetism, revealed by vibrating sample magnetometry (VSM), as a result of the sulfurization process. The MoS2@MoO3 gas sensors display a p-type-like response towards NO2 with a detection limit of 0.15 ppm at a working temperature of 125 °C, as well as superb selectivity and reversibility. This p-type-like sensing behavior is attributed to the heterointerface of MoS2-MoO3 where interfacial charge transfer leads to a p-type inversion layer in MoS2, and is enhanced by magnetic dipole interactions between the paramagnetic NO2 and the ferromagnetic sensing layer. Our study demonstrates the promising application of 2D molybdenum hybrid compounds in gas sensing applications with a unique combination of electronic and magnetic properties.

The detection and monitoring of nitrogen dioxide (NO 2 ) plays a vital role in the environmental, healthcare, farming, and industrial sectors. However, the development of NO 2 gas sensors with simultaneously high sensitivity, reversibility, low detection limit, and excellent selectivity remains challenging. In this work, an ultrasensitive NO 2 gas sensor with superb selectivity and reversibility is demonstrated based on α‐phase molybdenum trioxide (α‐MoO 3 ). Nanoribbons of α‐MoO 3 are synthesized via vapor phase transport (VPT) and systematically characterized using a combination of advanced characterization probes. At an optimal operating temperature of 125 °C, the α‐MoO 3 ‐based sensor shows a very high sensitivity toward NO 2 with a detection limit as low as 24 ppb, while also exhibiting excellent selectivity and reversibility. Such impressive performance originates from the layered nature of the α‐MoO 3 nanoribbons as well as the hierarchical assembly of the nanoribbons as the sensing layer. The study demonstrates a facile sensing platform based on α‐MoO 3 for ultrasensitive and selective NO 2 gas sensing.

Nanoceria has evolved as a promising nanomaterial due to its unique enzyme-like properties, including excellent oxidase mimetic activity, which significantly increases in the presence of fluoride ions. However, this significant increase in oxidase activity has never been utilised as a signal enhancer for the detection of biological analytes partly because of the lack of understanding of the mechanism involved in this process. In this study, we show that the surface oxidation state of cerium ions plays a very crucial role in different enzymatic activities, especially the oxidase mimetic activity by engineering nanoceria with three different surface Ce4+/Ce3+ compositions. Using DFT calculations combined with Bader charge analysis, it is demonstrated that stoichiometric ceria registers a higher oxidase mimetic activity than oxygen-deficient ceria with a low Ce4+/Ce3+ ratio due to a higher charge transfer from a substrate, 3,3 ',5,5 ' tetramethylbenzidine (TMB), to the ceria surface. We also show that the fluoride ions can significantly increase the charge transfer from the TMB surface to ceria irrespective of the surface Ce4+/Ce3+ ratio. Using this knowledge, we first compare the fluoride sensing properties of nanoceria with high Ce4+ and mixed Ce4+/Ce3+ oxidation states and further demonstrate that the linear detection range of fluoride ions can be extended to 1-10 ppm for nanoceria with mixed oxidation states. Then, we also demonstrate an assay for fluoride assisted detection of glutathione, an antioxidant with elevated levels during cancer, using nanoceria with a high surface Ce4+/Ce3+ ratio. The addition of fluoride ions in this assay allows the detection of glutathione in the linear range of 2.5-50 ppm with a limit of detection (LOD) of 3.8 ppm. These studies not only underpin the role of the surface Ce4+/Ce3+ ratio in tuning the fluoride assisted boost in the oxidase mimetic activity of nanoceria but also its strategic application in designing better colourimetric assays.

The p-type surface conductivity of hydrogen-terminated diamond (H-diamond) provides a viable approach toward diamond-based wide-bandgap metal-oxide-semiconductor field-effect transistors (MOSFETs) for high-power and high-frequency electronics. A facile, low-cost, and low-temperature method to form gate dielectrics on diamond that also preserves the integrity of hydrogen-termination is highly desirable for high-performance diamond surface electronics with process flexibility and high yield. In this work, we demonstrate a p-channel diamond MOSFET with an ultrathin glassy Ga2O3 dielectric layer derived from liquid metal. A liquid metal printing method was employed to transfer an amorphous Ga2O3 layer over the desired active p-channel region of H-diamond at low temperature, allowing the protection and preservation the hydrogen-terminated surface while also forming an efficient gate dielectric. The results of this work suggest that the liquid metal method can provide an efficient, low-cost, and high-yield pathway to form high-quality dielectrics for diamond-based transistors.

Hierarchical structures are very common in knowledge graphs, and semantic hierarchy-preserved knowledge graph embeddings have achieved promising results in the knowledge graph link prediction task. However, handling one-to-many, many-to-one, and many-to-many relations that can provide hierarchical information is challenging and brings entity indistinguishability issues. To address this limitation, this paper proposes a novel knowledge graph embedding model, namely Entity-context and Relation-context combined Knowledge Graph Embeddings (ERKE), in which each relation is defined as a rotation with variable moduli from the source entity to the target entity in the polar coordinate system. It can be seen as a combination of two spaces-modulus space and phase space. In the modulus space, modulus information is used to model semantic hierarchies, and entity-context information is adopted to make node representations more expressive. Besides, based on the design of the propagation rule of Graph Convolution Network (GCN), a new GCN model suitable for processing semantic hierarchies in knowledge graphs is proposed. In the phase space, relation-context information is used to make entities easier to distinguish. Specifically, a rotation operation in the polar coordinate system is transformed to the addition operation in the rectangular coordinate system, and relations between entities are mapped into their entity-specific hyperplanes. The proposed method is verified by the experiments on three benchmark datasets, and experimental results demonstrate that the proposed method can learn the semantic hierarchies in knowledge graphs and improve the prediction accuracy of complex one-to-many, many-to-one, and many-to-many cases simultaneously.

Sodium-ion batteries (SIBs) are emerging as an inexpensive and more sustainable alternative to lithiumion batteries in the energy storage market. To advance their commercialization, one major scientific undertaking is to develop low-cost, reliable anode materials from abundant resources, like the success of graphite in the lithium-ion batteries. However, graphite is chemically inactive in storing sodium ions and, to render it viable in sodium-ion batteries, additional modification of graphite is required. Herein, we demonstrate a green and facile method to prepare cost-effective and stable graphitic SIB anodes. The modification process started with the electrochemical oxidation of expanded graphite to widen the interlayer and functionalize graphite layers, followed by a fast (20 min) thermal treatment at 150 degrees C to achieve controlled deoxygenation. The thermally processed electrochemical graphite oxide could provide a high reversible capacity of 268 mAh g(-1) at 100 mA g(-1) and 163 mAh g(-1) at 500 mA g(-1) as well as low fading in capacity (in average 0.0198% loss per cycle) over 2000 cycles. The electrochemical route eliminates the need for the harsh chemical oxidation of graphite, offering a promising approach for industrial production of low-cost anodes for sodium-ion batteries. (c) 2021 Elsevier Ltd. All rights reserved.

The surface of hydrogen-terminated diamond (H-terminated diamond) supports a p-type surface conductivity when interfaced with high electron-affinity surface acceptors through the surface transfer doping process. High electron-affinity transition metal oxides (TMOs), such as MoO3, have been regarded as superior candidates in surface transfer doping of diamond, holding great promise for enabling practical diamond electronic devices. In this work, a simple, solution-processed method is demonstrated to deposit a molybdenum trioxide (MoO3) layer on H-terminated diamond surface to achieve surface transfer doping of diamond. The surface of diamond following the deposition of solution-processed MoO3 (sp-MoO3) experienced significant reduction in sheet resistivity, corresponding to an increase in hole density, compared with the pristine air-doped diamond. This hole accumulation layer induced by sp-MoO3 is thermally stable up to 450 degrees C, and demonstrates metallic conductivity down to 250 mK with a strong spin-orbit interaction (SOI) in the induced two-dimensional (2D) surface conducting layer. Our study demonstrates that sp-MoO3 is comparable with thermally-evaporated MoO3 in doping performance but with great advantage as a simple, inexpensive and highly flexible fabrication method for developing diamond electronics. (C) 2020 Elsevier Ltd. All rights reserved.

Solar energy harvesting and conversion into useful chemical energy with the aid of semiconductor photocatalysts is a promising technique to solve both energy and environmental issues. This work reports a successful synthesis of CdSe quantum dots (QDs) modified phosphorus doped g-C3N4 (P-CN) for advanced photocatalytic applications. Phosphorus doping and structural coupling with CdSe QDs are shown to significantly extend visible-light response of g-C3N4 up to 700 nm. The optimized sample 4CdSe/P-CN demonstrates enhanced visible-light driven overall water splitting activities for H-2 and O-2 evolution i.e. 113 and 55.5 mu mol.h(-1).g(-1), respectively, as well as very high photocatalytic CO2 to CH4 conversion efficiency (47 mu mol.h(-1). g(-1)). It also exhibit higher activity (78 %) for 2,4-dichlorophenol degradation as compared to pristine CN-sample. Combined photoluminescence, transient/single wavelength photocurrent, photoelectrochemical, and coumarin fluorescence spectroscopy demonstrate that 4CdSe/P-CN nanocomposite exhibit enhanced charge separation efficiency which is responsible for improved visible light catalytic activities. Our work thus provide a new strategy to design low-cost and sustainable photocatalysis with wide visible-light activity for practical overall water splitting and CO2 reduction applications.

Graph convolutional networks (GCNs) have been developed as a general and powerful tool to handle various tasks related to graph data. However, current methods mainly consider homogeneous networks and ignore the rich semantics and multiple types of objects that are common in heterogeneous information networks (HINs). In this paper, we present a Heterogeneous Hyperedge Convolutional Network (HHCN), a novel graph convolutional network architecture that operates on HINs. Specifically, we extract the rich semantics by different metastructures and adopt hyperedge to model the interactions among metastructure-based neighbors. Due to the powerful information extraction capabilities of metastructure and hyperedge, HHCN has the flexibility to model the complex relationships in HINs by setting different combinations of metastructures and hyperedges. Moreover, a metastructure attention layer is also designed to allow each node to select the metastructures based on their importance and provide potential interpretability for graph analysis. As a result, HHCN can encode node features, metastructure-based semantics and hyperedge information simultaneously by aggregating features from metastructure-based neighbors in a hierarchical manner. We evaluate HHCN by applying it to the semi-supervised node classification task. Experimental results show that HHCN outperforms state-of-the-art graph embedding models and recently proposed graph convolutional network models.

Graphene oxide (GO) is well known as a key material to the commercialization of graphene-based applications due to its excellent processability and abundant starting materials. However, producing high quality GO with excellent structural intactness still remains a challenge despite the significant research effort over the past decade. Herein, we demonstrate an effective approach to achieving well-oxidized GO within only 5 h of oxidation time at 5 degrees C or 25 degrees C from expanded graphite as a starting material. Our finding reveals that the well-oxidized GO synthesized at 25 degrees C can regain more graphitic sp(2) networks and thus becomes as conductive as the GO prepared at 5 degrees C by means of thermal annealing or green chemical reduction. We also found that green chemical reduction using vitamin C (VC) is more efficient in restoring the electrical conductivity and reducing the defects in the GO produced at room temperature. The protocol reported in this paper offers a promising, sustainable way to fabricate high quality GO with minimal energy input, while being cost-effective. (C) 2020 Elsevier B.V. All rights reserved.

As an extension of convolutional neural networks from Euclidean data to irregular data, Graph Convolutional Networks (GCNs) have attracted much attention and applied in a variety of tasks. However, current methods mainly consider unsigned graphs and ignore the positive and negative links in graphs that are common in our daily lives. The inherent social theories (balance theory and status theory) and the different semantic information of positive and negative links of a signed graph bring great challenges for modeling signed graphs by GCNs. As the latest signed graph convolutional network, Signed Graph Attention Networks (SiGAT) generalizes Graph Attention Networks (GATs) to signed graph by adopting motifs to model the social theories and shows superior performance. However, there are information loss and redundancy in the designed motifs and the different effects of positive and negative links are also not considered. In this paper, we propose a Hierarchical Attention Signed Network (HASN) which includes a hierarchical motif framework and a hierarchical attention framework. Specifically, the hierarchical motif framework divides 68 motifs into two layers according to if they can reflect the balance theory. The hierarchical attention framework contains a node-level attention layer and a sign-level attention layer. The node-level attention layer learns the importance of motif-based neighbors and performs weighted aggregations to update node embeddings. The sign-level attention layer can distinguish the different importance of positive and negative links. Based on the hierarchical motif framework and hierarchical attention framework, the social theories information, the importance of different motif-based neighbors, and the importance of signs can be preserved comprehensively. Experimental results on three real-world signed graphs for the signed link prediction task show that HASN outperforms state-of-the-art graph embedding models and recently proposed signed graph convolutional network models.

Mucosal-like ZnS nanostructures were synthesized on surface of ST-cut quartz surface acoustic wave (SAW) device using a chemical bath deposition method for ammonia sensing applications. Results showed the SAW device with ZnS mucosal nanostructures achieved a good sensitivity, e.g., with a frequency shift of similar to 190 Hz upon exposure to 1 ppm ammonia gas. This is mainly attributed to the large specific surface areas and more active sites on the surfaces of the ZnS mucosal nanostructures. Selectivity of the SAW device with the ZnS mucosal nanostructure for ammonia sensing was excellent in comparisons with those for other types of gases such as hydrogen sulfide, hydrogen, nitrogen dioxide, carbon monoxide and ethanol.