Search

Traditional online advertising systems for sponsored search follow a cascade paradigm with retrieval, pre-ranking,ranking, respectively. Constrained by strict requirements on online inference efficiency, it tend to be difficult to deploy useful but computationally intensive modules in the ranking stage. Moreover, ranking models currently used in the industry assume the user click only relies on the advertisements itself, which results in the ranking stage overlooking the impact of organic search results on the predicted advertisements (ads). In this work, we propose a novel framework PCDF(Parallel-Computing Distributed Framework), allowing to split the computation cost into three parts and to deploy them in the pre-module in parallel with the retrieval stage, the middle-module for ranking ads, and the post-module for re-ranking ads with external items. Our PCDF effectively reduces the overall inference latency compared with the classic framework. The whole module is end-to-end offline training and adapt for the online learning paradigm. To our knowledge, we are the first to propose an end-to-end solution for online training and deployment on complex CTR models from the system framework side.

Through infrared spectroscopy, we systematically study the pressure effect on electronic structures of few-layer black phosphorus (BP) with layer number ranging from 2 to 13. We reveal that the pressure-induced shift of optical transitions exhibits strong layer-dependence. In sharp contrast to the bulk counterpart which undergoes a semiconductor to semimetal transition under ~1.8 GPa, the bandgap of 2 L increases with increasing pressure until beyond 2 GPa. Meanwhile, for a sample with a given layer number, the pressure-induced shift also differs for transitions with different indices. Through the tight-binding model in conjunction with a Morse potential for the interlayer coupling, this layer- and transition-index-dependent pressure effect can be fully accounted. Our study paves a way for versatile van der Waals engineering of two-dimensional BP., Comment: 16 pages, 4 figures

Accurate layout analysis without subsequent text-line segmentation remains an ongoing challenge, especially when facing the Kangyur, a kind of historical Tibetan document featuring considerable touching components and mottled background. Aiming at identifying different regions in document images, layout analysis is indispensable for subsequent procedures such as character recognition. However, there was only a little research being carried out to perform line-level layout analysis which failed to deal with the Kangyur. To obtain the optimal results, a fine-grained sub-line level layout analysis approach is presented. Firstly, we introduced an accelerated method to build the dataset which is dynamic and reliable. Secondly, enhancement had been made to the SOLOv2 according to the characteristics of the Kangyur. Then, we fed the enhanced SOLOv2 with the prepared annotation file during the training phase. Once the network is trained, instances of the text line, sentence, and titles can be segmented and identified during the inference stage. The experimental results show that the proposed method delivers a decent 72.7% average precision on our dataset. In general, this preliminary research provides insights into the fine-grained sub-line level layout analysis and testifies the SOLOv2-based approaches. We also believe that the proposed methods can be adopted on other language documents with various layouts., Comment: This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible

Hyperbolic polaritons exhibit large photonic density of states and can be collimated in certain propagation directions. The majority of hyperbolic polaritons are sustained in man-made metamaterials. However, natural-occurring hyperbolic materials also exist. Particularly, natural in-plane hyperbolic polaritons in layered materials have been demonstrated in MoO3 and WTe2, which are based on phonon and plasmon resonances respectively. Here, by determining the anisotropic optical conductivity (dielectric function) through optical spectroscopy, we predict that monolayer black phosphorus naturally hosts hyperbolic exciton-polaritons due to the pronounced in-plane anisotropy and strong exciton resonances. We simultaneously observe a strong and sharp ground state exciton peak and weaker excited states in high quality monolayer samples in the reflection spectrum, which enables us to determine the exciton binding energy of ~452 meV. Our work provides another appealing platform for the in-plane natural hyperbolic polaritons, which is based on excitons rather than phonons or plasmons.

Point clouds and images could provide complementary information when representing 3D objects. Fusing the two kinds of data usually helps to improve the detection results. However, it is challenging to fuse the two data modalities, due to their different characteristics and the interference from the non-interest areas. To solve this problem, we propose a Multi-Branch Deep Fusion Network (MBDF-Net) for 3D object detection. The proposed detector has two stages. In the first stage, our multi-branch feature extraction network utilizes Adaptive Attention Fusion (AAF) modules to produce cross-modal fusion features from single-modal semantic features. In the second stage, we use a region of interest (RoI) -pooled fusion module to generate enhanced local features for refinement. A novel attention-based hybrid sampling strategy is also proposed for selecting key points in the downsampling process. We evaluate our approach on two widely used benchmark datasets including KITTI and SUN-RGBD. The experimental results demonstrate the advantages of our method over state-of-the-art approaches.

Graph neural networks (GNNs) have been popularly used in analyzing graph-structured data, showing promising results in various applications such as node classification, link prediction and network recommendation. In this paper, we present a new graph attention neural network, namely GIPA, for attributed graph data learning. GIPA consists of three key components: attention, feature propagation and aggregation. Specifically, the attention component introduces a new multi-layer perceptron based multi-head to generate better non-linear feature mapping and representation than conventional implementations such as dot-product. The propagation component considers not only node features but also edge features, which differs from existing GNNs that merely consider node features. The aggregation component uses a residual connection to generate the final embedding. We evaluate the performance of GIPA using the Open Graph Benchmark proteins (ogbn-proteins for short) dataset. The experimental results reveal that GIPA can beat the state-of-the-art models in terms of prediction accuracy, e.g., GIPA achieves an average test ROC-AUC of $0.8700\pm 0.0010$ and outperforms all the previous methods listed in the ogbn-proteins leaderboard., Comment: 5 pages, 2 figures, 2 tables; Technical report

Graph neural networks (GNNs) have been demonstrated as a powerful tool for analyzing non-Euclidean graph data. However, the lack of efficient distributed graph learning systems severely hinders applications of GNNs, especially when graphs are big and GNNs are relatively deep. Herein, we present GraphTheta, the first distributed and scalable graph learning system built upon vertex-centric distributed graph processing with neural network operators implemented as user-defined functions. This system supports multiple training strategies and enables efficient and scalable big-graph learning on distributed (virtual) machines with low memory. To facilitate graph convolutions, GraphTheta puts forward a new graph learning abstraction named NN-TGAR to bridge the gap between graph processing and graph deep learning. A distributed graph engine is proposed to conduct the stochastic gradient descent optimization with a hybrid-parallel execution, and a new cluster-batched training strategy is supported. We evaluate GraphTheta using several datasets with network sizes ranging from small-, modest- to large-scale. Experimental results show that GraphTheta can scale well to 1,024 workers for training an in-house developed GNN on an industry-scale Alipay dataset of 1.4 billion nodes and 4.1 billion attributed edges, with a cluster of CPU virtual machines (dockers) of small memory each (5$\sim$12GB). Moreover, GraphTheta can outperform DistDGL by up to $2.02\times$, with better scalability, and GraphLearn by up to $30.56\times$. As for model accuracy, GraphTheta is capable of learning as good GNNs as existing frameworks. To the best of our knowledge, this work presents the largest edge-attributed GNN learning task in the literature., Comment: 19 pages, 13 figures, 8 tables

In 2D materials, the quantum confinement and van der Waals-type interlayer interactions largely govern the fundamental electronic and optical properties, and the dielectric screening plays a dominant role in the excitonic properties. This suggests strongly layer-dependent properties, and a central topic is to characterize and control the interlayer interactions in 2D materials and heterostructures. Black phosphorus is an emerging 2D semiconductor with unusually strong interlayer interactions and widely tunable direct bandgaps from the monolayer to the bulk, offering us an ideal platform to probe the layer-dependent properties and the crossover from 2D to 3D (i.e., the scaling effects). In this review, we present a comprehensive and thorough summary of the fundamental physical properties of black phosphorus, with a special focus on the layer-dependence characters, including the electronic band structures, optical absorption and photoluminescence, and excitonic properties, as well as the band structure engineering by means of electrical gating, strain, and electrochemical intercalation. Finally, we give an outlook for the future research., Comment: Review Article, 57 pages, 16 figures

The observation of the electrically tunable and highly confined plasmons in graphene has stimulated the exploration of interesting properties of plasmons in other two dimensional materials. Recently, hyperbolic plasmon resonance modes are observed in exfoliated WTe2 films, a type-II Weyl semimetal with layered structure, providing a platform for the assembly of plasmons with hyperbolicity and exotic topological properties. However, the plasmon modes were observed in relatively thick and small-area films, which restrict the tunability and application for plasmons. Here, large-area (~ cm) WTe2 films with different thickness are grown by chemical vapor deposition method, in which plasmon resonance modes are observed in films with different thickness down to about 8 nm. Hybridization of plasmon and surface polar phonons of the substrate is revealed by mapping the plasmon dispersion. The plasmon frequency is demonstrated to be tunable by changing the temperature and film thickness. Our results facilitate the development of a tunable and scalable WTe2 plasmonic system for revealing topological properties and towards various applications in sensing, imaging and light modulation., Comment: 18 pages, 5 figures

The temperature dependence of the band gap is crucial to a semiconductor. Bulk black phosphorus (BP) is known to exhibit an anomalous behavior. Through optical spectroscopy, here we show that the temperature effect on BP band gap gradually evolves with decreasing layer number, eventually turns into a normal one in the monolayer limit, rendering a crossover from the anomalous to the normal. Meanwhile, the temperature-induced shift in optical resonance also differs with different transition indices for the same thickness sample. A comprehensive analysis reveals that the temperature-tunable interlayer coupling is responsible for the observed diverse scenario. Our study provides a key to the apprehension of the anomalous temperature behavior in certain layered semiconductors.

The strength of light-matter interaction is of central importance in photonics and optoelectronics. For many widely studied two-dimensional semiconductors, such as MoS2, the optical absorption due to exciton resonances increases with thickness. However, here we will show, few-layer black phosphorus exhibits an opposite trend. We determine the optical conductivity of few-layer black phosphorus with thickness down to bilayer by infrared spectroscopy. On the contrary to our expectations, the frequency-integrated exciton absorption is found to be enhanced in thinner samples. Moreover, the continuum absorption near the band edge is almost a constant, independent of the thickness. We will show such scenario is related to the quanta of the universal optical conductivity of graphene, with a prefactor originating from the band anisotropy., Comment: 22 pages, 4 figures

A new variational mode decomposition (VMD) based deep learning approach is proposed in this paper for time series forecasting problem. Firstly, VMD is adopted to decompose the original time series into several sub-signals. Then, a convolutional neural network (CNN) is applied to learn the reconstruction patterns on the decomposed sub-signals to obtain several reconstructed sub-signals. Finally, a long short term memory (LSTM) network is employed to forecast the time series with the decomposed sub-signals and the reconstructed sub-signals as inputs. The proposed VMD-CNN-LSTM approach is originated from the decomposition-reconstruction-ensemble framework, and innovated by embedding the reconstruction, single forecasting, and ensemble steps in a unified deep learning approach. To verify the forecasting performance of the proposed approach, four typical time series datasets are introduced for empirical analysis. The empirical results demonstrate that the proposed approach outperforms consistently the benchmark approaches in terms of forecasting accuracy, and also indicate that the reconstructed sub-signals obtained by CNN is of importance for further improving the forecasting performance.

In the fast growing two-dimensional (2D) materials family, anisotropic 2D materials, with their intrinsic in-plane anisotropy, exhibit a great potential in optoelectronics. One such typical material is black phosphorus (BP), with a layer-dependent and highly tunable band gap. Such intrinsic anisotropy adds a new degree of freedom to the excitation, detection and control of light. Particularly, hyperbolic plasmons with hyperbolic q-space dispersion are predicted to exist in BP films, where highly directional propagating polaritons with divergent densities of states are hosted. Combined with a tunable electronic structure, such natural hyperbolic surfaces may enable a series of exotic applications in nanophotonics. In this review, the anisotropic optical properties and plasmons (especially hyperbolic plasmons) of BP are discussed. In addition, other possible 2D material candidates (especially anisotropic layered semimetals) for hyperbolic plasmons are examined. This review may stimulate further research interest in anisotropic 2D materials and fully unleash their potential in flatland photonics., Comment: 60 pages,10 figures

Interlayer interactions in 2D materials, also known as van der Waals (vdWs) interactions, play a critical role in the physical properties of layered materials. It is fascinating to manipulate the vdWs interaction, and hence to "redefine" the material properties. Here, we demonstrate that in-plane biaxial strain can effectively tune the vdWs interaction of few-layer black phosphorus with thickness of 2-10 layers, using infrared spectroscopy. Surprisingly, our results reveal that in-plane tensile strain efficiently weakens the interlayer coupling, even though the sample shrinks in the vertical direction due to the Poisson effect, in sharp contrast to one's intuition. Moreover, density functional theory (DFT) calculations further confirm our observations and indicate a dominant role of the puckered lattice structure. Our study highlights the important role played by vdWs interactions in 2D materials during external physical perturbations., Comment: 23 pages, 3 figures