Nanjing University of Science and Technology

Although convolutional neural networks (CNNs) have achieved great success in computer vision, this work investigates a simpler, convolution-free backbone network use-fid for many dense prediction tasks. Unlike the recently-proposed Vision Transformer (ViT) that was designed for image classification specifically, we introduce the Pyramid Vision Transformer (PVT), which overcomes the difficulties of porting Transformer to various dense prediction tasks. PVT has several merits compared to current state of the arts. (1) Different from ViT that typically yields low-resolution outputs and incurs high computational and memory costs, PVT not only can be trained on dense partitions of an image to achieve high output resolution, which is important for dense prediction, but also uses a progressive shrinking pyramid to reduce the computations of large feature maps. (2) PVT inherits the advantages of both CNN and Transformer, making it a unified backbone for various vision tasks without convolutions, where it can be used as a direct replacement for CNN backbones. (3) We validate PVT through extensive experiments, showing that it boosts the performance of many downstream tasks, including object detection, instance and semantic segmentation. For example, with a comparable number of parameters, PVT+RetinaNet achieves 40.4 AP on the COCO dataset, surpassing ResNet50+RetinNet (36.3 AP) by 4.1 absolute AP (see Figure 2). We hope that PVT could, serre as an alternative and useful backbone for pixel-level predictions and facilitate future research.

An experimental system was built to investigate convective heat transfer and flow features of the nanofluid in a tube. Both the convective heat transfer coefficient and friction factor of the sample nanofluids for the turbulent flow are measured, respectively. The effects of such factors as the volume fraction of suspended nanoparticles and the Reynolds number on the heat transfer and flow features are discussed in detail. A new type of convective heat transfer correlation is proposed to correlate experimental data of heat transfer for nanofluids.

In this paper, a new technique coined two-dimensional principal component analysis (2DPCA) is developed for image representation. As opposed to PCA, 2DPCA is based on 2D image matrices rather than 1D vectors so the image matrix does not need to be transformed into a vector prior to feature extraction. Instead, an image covariance matrix is constructed directly using the original image matrices, and its eigenvectors are derived for image feature extraction. To test 2DPCA and evaluate its performance, a series of experiments were performed on three face image databases: ORL, AR, and Yale face databases. The recognition rate across all trials was higher using 2DPCA than PCA. The experimental results also indicated that the extraction of image features is computationally more efficient using 2DPCA than PCA.

In standard Convolutional Neural Networks (CNNs), the receptive fields of artificial neurons in each layer are designed to share the same size. It is well-known in the neuroscience community that the receptive field size of visual cortical neurons are modulated by the stimulus, which has been rarely considered in constructing CNNs. We propose a dynamic selection mechanism in CNNs that allows each neuron to adaptively adjust its receptive field size based on multiple scales of input information. A building block called Selective Kernel (SK) unit is designed, in which multiple branches with different kernel sizes are fused using softmax attention that is guided by the information in these branches. Different attentions on these branches yield different sizes of the effective receptive fields of neurons in the fusion layer. Multiple SK units are stacked to a deep network termed Selective Kernel Networks (SKNets). On the ImageNet and CIFAR benchmarks, we empirically show that SKNet outperforms the existing state-of-the-art architectures with lower model complexity. Detailed analyses show that the neurons in SKNet can capture target objects with different scales, which verifies the capability of neurons for adaptively adjusting their receptive field sizes according to the input. The code and models are available at https://github.com/implus/SKNet.

Novel quantum-dot light-emitting diodes based on all-inorganic perovskite CsPbX3 (X = Cl, Br, I) nanocrystals are reported. The well-dispersed, single-crystal quantum dots (QDs) exhibit high quantum yields, and tunable light emission wavelength. The demonstration of these novel perovskite QDs opens a new avenue toward designing optoelectronic devices, such as displays, photodetectors, solar cells, and lasers. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.

Recently, Kovalenko and co‐workers and Li and co‐workers developed CsPbX 3 (X = Cl, Br, I) inorganic perovskite quantum dots (IPQDs), which exhibited ultrahigh photoluminescence (PL) quantum yields (QYs), low‐threshold lasing, and multicolor electroluminescence. However, the usual synthesis needs high temperature, inert gas protection, and localized injection operation, which are severely against applications. Moreover, the so unexpectedly high QYs are very confusing. Here, for the first time, the IPQDs' room‐temperature (RT) synthesis, superior PL, underlying origins and potentials in lighting and displays are reported. The synthesis is designed according to supersaturated recrystallization (SR), which is operated at RT, within few seconds, free from inert gas and injection operation. Although formed at RT, IPQDs' PLs have QYs of 80%, 95%, 70%, and FWHMs of 35, 20, and 18 nm for red, green, and blue emissions. As to the origins, the observed 40 meV exciton binding energy, halogen self‐passivation effect, and CsPbX 3 @X quantum‐well band alignment are proposed to guarantee the excitons generation and high‐rate radiative recombination at RT. Moreover, such superior optical merits endow them with promising potentials in lighting and displays, which are primarily demonstrated by the white light‐emitting diodes with tunable color temperature and wide color gamut.

Recently, Convolutional Neural Network (CNN) based models have achieved great success in Single Image Super-Resolution (SISR). Owing to the strength of deep networks, these CNN models learn an effective nonlinear mapping from the low-resolution input image to the high-resolution target image, at the cost of requiring enormous parameters. This paper proposes a very deep CNN model (up to 52 convolutional layers) named Deep Recursive Residual Network (DRRN) that strives for deep yet concise networks. Specifically, residual learning is adopted, both in global and local manners, to mitigate the difficulty of training very deep networks, recursive learning is used to control the model parameters while increasing the depth. Extensive benchmark evaluation shows that DRRN significantly outperforms state of the art in SISR, while utilizing far fewer parameters. Code is available at https://github.com/tyshiwo/DRRN_CVPR17.

Federated learning (FL), as a type of distributed machine learning, is capable of significantly preserving clients’ private data from being exposed to adversaries. Nevertheless, private information can still be divulged by analyzing uploaded parameters from clients, e.g., weights trained in deep neural networks. In this paper, to effectively prevent information leakage, we propose a novel framework based on the concept of differential privacy (DP), in which artificial noise is added to parameters at the clients’ side before aggregating, namely, noising before model aggregation FL (NbAFL). First, we prove that the NbAFL can satisfy DP under distinct protection levels by properly adapting different variances of artificial noise. Then we develop a theoretical convergence bound on the loss function of the trained FL model in the NbAFL. Specifically, the theoretical bound reveals the following three key properties: 1) there is a tradeoff between convergence performance and privacy protection levels, i.e., better convergence performance leads to a lower protection level; 2) given a fixed privacy protection level, increasing the number <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$N$ </tex-math></inline-formula> of overall clients participating in FL can improve the convergence performance; and 3) there is an optimal number aggregation times (communication rounds) in terms of convergence performance for a given protection level. Furthermore, we propose a <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$K$ </tex-math></inline-formula> -client random scheduling strategy, where <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$K$ </tex-math></inline-formula> ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$1\leq K&lt; N$ </tex-math></inline-formula> ) clients are randomly selected from the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$N$ </tex-math></inline-formula> overall clients to participate in each aggregation. We also develop a corresponding convergence bound for the loss function in this case and the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$K$ </tex-math></inline-formula> -client random scheduling strategy also retains the above three properties. Moreover, we find that there is an optimal <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$K$ </tex-math></inline-formula> that achieves the best convergence performance at a fixed privacy level. Evaluations demonstrate that our theoretical results are consistent with simulations, thereby facilitating the design of various privacy-preserving FL algorithms with different tradeoff requirements on convergence performance and privacy levels.

Transformers have recently lead to encouraging progress in computer vision. In this work, we present new baselines by improving the original Pyramid Vision Transformer (PVT v1) by adding three designs: (i) a linear complexity attention layer, (ii) an overlapping patch embedding, and (iii) a convolutional feed-forward network. With these modifications, PVT v2 reduces the computational complexity of PVT v1 to linearity and provides significant improvements on fundamental vision tasks such as classification, detection, and segmentation. In particular, PVT v2 achieves comparable or better performance than recent work such as the Swin transformer. We hope this work will facilitate state-of-the-art transformer research in computer vision. Code is available at https://github.com/whai362/PVT .

A composite of graphene oxide supported by needle-like MnO(2) nanocrystals (GO-MnO(2) nanocomposites) has been fabricated through a simple soft chemical route in a water-isopropyl alcohol system. The formation mechanism of these intriguing nanocomposites investigated by transmission electron microscopy and Raman and ultraviolet-visible absorption spectroscopy is proposed as intercalation and adsorption of manganese ions onto the GO sheets, followed by the nucleation and growth of the crystal species in a double solvent system via dissolution-crystallization and oriented attachment mechanisms, which in turn results in the exfoliation of GO sheets. Interestingly, it was found that the electrochemical performance of as-prepared nanocomposites could be enhanced by the chemical interaction between GO and MnO(2). This method provides a facile and straightforward approach to deposit MnO(2) nanoparticles onto the graphene oxide sheets (single layer of graphite oxide) and may be readily extended to the preparation of other classes of hybrids based on GO sheets for technological applications.

Grain refinement can make conventional metals several times stronger, but this comes at dramatic loss of ductility. Here we report a heterogeneous lamella structure in Ti produced by asymmetric rolling and partial recrystallization that can produce an unprecedented property combination: as strong as ultrafine-grained metal and at the same time as ductile as conventional coarse-grained metal. It also has higher strain hardening than coarse-grained Ti, which was hitherto believed impossible. The heterogeneous lamella structure is characterized with soft micrograined lamellae embedded in hard ultrafine-grained lamella matrix. The unusual high strength is obtained with the assistance of high back stress developed from heterogeneous yielding, whereas the high ductility is attributed to back-stress hardening and dislocation hardening. The process discovered here is amenable to large-scale industrial production at low cost, and might be applicable to other metal systems.

Recently, very deep convolutional neural networks (CNNs) have been attracting considerable attention in image restoration. However, as the depth grows, the longterm dependency problem is rarely realized for these very deep models, which results in the prior states/layers having little influence on the subsequent ones. Motivated by the fact that human thoughts have persistency, we propose a very deep persistent memory network (MemNet) that introduces a memory block, consisting of a recursive unit and a gate unit, to explicitly mine persistent memory through an adaptive learning process. The recursive unit learns multi-level representations of the current state under different receptive fields. The representations and the outputs from the previous memory blocks are concatenated and sent to the gate unit, which adaptively controls how much of the previous states should be reserved, and decides how much of the current state should be stored. We apply MemNet to three image restoration tasks, i.e., image denosing, super-resolution and JPEG deblocking. Comprehensive experiments demonstrate the necessity of the MemNet and its unanimous superiority on all three tasks over the state of the arts. Code is available at https://github.com/tyshiwo/MemNet.

Graphene sheets, which possess unique nanostructure and a variety of fascinating properties, can be considered as promising nanoscale building blocks of new composites, for example, a support material for the dispersion of nanoparticles. Here, we present a general approach for the preparation of graphene−metal particle nanocomposites in a water−ethylene glycol system using graphene oxide as a precursor and metal nanoparticles (Au, Pt and Pd) as building blocks. These metal nanoparticles are adsorbed on graphene oxide sheets and play a pivotal role in catalytic reduction of graphene oxide with ethylene glycol, leading to the formation of graphene−metal particle nanocomposites. The typical methanol oxidation of graphene−Pt composites in cyclic voltammograms analyses indicated its potential application in direct methanol fuel cells, bringing graphene−particle nanocomposites close to real technological applications.

The typical two-dimensional (2D) semiconductors MoS2, MoSe2, WS2, WSe2 and black phosphorus have garnered tremendous interest for their unique electronic, optical, and chemical properties. However, all 2D semiconductors reported thus far feature band gaps that are smaller than 2.0 eV, which has greatly restricted their applications, especially in optoelectronic devices with photoresponse in the blue and UV range. Novel 2D mono-elemental semiconductors, namely monolayered arsenene and antimonene, with wide band gaps and high stability were now developed based on first-principles calculations. Interestingly, although As and Sb are typically semimetals in the bulk, they are transformed into indirect semiconductors with band gaps of 2.49 and 2.28 eV when thinned to one atomic layer. Significantly, under small biaxial strain, these materials were transformed from indirect into direct band-gap semiconductors. Such dramatic changes in the electronic structure could pave the way for transistors with high on/off ratios, optoelectronic devices working under blue or UV light, and mechanical sensors based on new 2D crystals.

We report significant back stress strengthening and strain hardening in gradient structured (GS) interstitial-free (IF) steel. Back stress is long-range stress caused by the pileup of geometrically necessary dislocations (GNDs). A simple equation and a procedure are developed to calculate back stress basing on its formation physics from the tensile unloading-reloading hysteresis loop. The gradient structure has mechanical incompatibility due to its grain size gradient. This induces strain gradient, which needs to be accommodated by GNDs. Back stress not only raises the yield strength but also significantly enhances strain hardening to increase the ductility. [GRAPHICS] .

Heterostructured materials have been reported as a new class of materials with superior mechanical properties, which was attributed to the development of back stress. There are numerous reports on back stress theories and measurements with no consensus. Back stress is developed in soft domains to offset the applied stress, making them appear stronger, while forward stress is developed to make hard domains appear weaker. The extra hardening in heterostructured materials is resulted from interactions between back stresses and forward stresses, and should be described as hetero-deformation induced (HDI) hardening and the measured ‘back stress’ should be renamed HDI stress.

In this paper, a multichannel EEG emotion recognition method based on a novel dynamical graph convolutional neural networks (DGCNN) is proposed. The basic idea of the proposed EEG emotion recognition method is to use a graph to model the multichannel EEG features and then perform EEG emotion classification based on this model. Different from the traditional graph convolutional neural networks (GCNN) methods, the proposed DGCNN method can dynamically learn the intrinsic relationship between different electroencephalogram (EEG) channels, represented by an adjacency matrix, via training a neural network so as to benefit for more discriminative EEG feature extraction. Then, the learned adjacency matrix is used to learn more discriminative features for improving the EEG emotion recognition. We conduct extensive experiments on the SJTU emotion EEG dataset (SEED) and DREAMER dataset. The experimental results demonstrate that the proposed method achieves better recognition performance than the state-of-the-art methods, in which the average recognition accuracy of 90.4 percent is achieved for subject dependent experiment while 79.95 percent for subject independent cross-validation one on the SEED database, and the average accuracies of 86.23, 84.54 and 85.02 percent are respectively obtained for valence, arousal and dominance classifications on the DREAMER database.

Metal-halide perovskites have rapidly emerged as one of the most promising materials of the 21st century, with many exciting properties and great potential for a broad range of applications, from photovoltaics to optoelectronics and photocatalysis. The ease with which metal-halide perovskites can be synthesized in the form of brightly luminescent colloidal nanocrystals, as well as their tunable and intriguing optical and electronic properties, has attracted researchers from different disciplines of science and technology. In the last few years, there has been a significant progress in the shape-controlled synthesis of perovskite nanocrystals and understanding of their properties and applications. In this comprehensive review, researchers having expertise in different fields (chemistry, physics, and device engineering) of metal-halide perovskite nanocrystals have joined together to provide a state of the art overview and future prospects of metal-halide perovskite nanocrystal research.

Alginate is a natural polysaccharide exhibiting excellent biocompatibility and biodegradability, having many different applications in the field of biomedicine. Alginate is readily processable for applicable three-dimensional scaffolding materials such as hydrogels, microspheres, microcapsules, sponges, foams and fibers. Alginate-based biomaterials can be utilized as drug delivery systems and cell carriers for tissue engineering. Alginate can be easily modified via chemical and physical reactions to obtain derivatives having various structures, properties, functions and applications. Tuning the structure and properties such as biodegradability, mechanical strength, gelation property and cell affinity can be achieved through combination with other biomaterials, immobilization of specific ligands such as peptide and sugar molecules, and physical or chemical crosslinking. This review focuses on recent advances in the use of alginate and its derivatives in the field of biomedical applications, including wound healing, cartilage repair, bone regeneration and drug delivery, which have potential in tissue regeneration applications.

Here we present a perspective on heterogeneous materials, a new class of materials possessing superior combinations of strength and ductility that are not accessible to their homogeneous counterparts. Heterogeneous materials consist of domains with dramatic strength differences. The domain sizes may vary in the range of micrometers to millimeters. Large strain gradients near domain interfaces are produced during deformation, which produces a significant back-stress to strengthen the material and to produce high back-stress work hardening for good ductility. High interface density is required to maximize the back-stress, which is a new strengthening mechanism for improving mechanical properties.