arXiv subjects

Xing Wang

Publications and source records attributed to Xing Wang.

At least 145 records · Page 8Linked to original sources

arXiv2023

Taming Calabi-Yau Feynman integrals: The four-loop equal-mass banana integral

Certain Feynman integrals are associated to Calabi-Yau geometries. We demonstrate how these integrals can be computed with the method of differential equations. The four-loop equal-mass banana integral is the simplest Feynman integral whose geometry is a non-trivial Calabi-Yau manifold. We show that its differential equation can be cast into an $\varepsilon$-factorised form. This allows us to obtain the solution to any desired order in the dimensional regularisation parameter $\varepsilon$. The method generalises to other Calabi-Yau Feynman integrals. Our calculation also shows that the four-loop banana integral is only minimally more complicated than the corresponding Feynman integrals at two or three loops.

hep-th↗

arXiv2022

Sharp $L^p$ estimates and size of nodal sets for generalized Steklov eigenfunctions

We prove sharp $L^p$ estimates for the Steklov eigenfunctions on compact manifolds with boundary in terms of their $L^2$ norms on the boundary. We prove it by establishing $L^p$ bounds for the harmonic extension operators as well as the spectral projection operators on the boundary. Moreover, we derive lower bounds on the size of nodal sets for a variation of the Steklov spectral problem. We consider a generalized version of the Steklov problem by adding a non-smooth potential on the boundary but some of our results are new even without potential.

math.AP↗

arXiv2022

Factorization at Next-to-Leading Power and Endpoint Divergences in $gg\to h$ Production

We derive a factorization theorem for the Higgs-boson production amplitude in gluon-gluon fusion induced by a light-quark loop, working at next-to-leading power in soft-collinear effective theory. The factorization is structurally similar to that obtained for the $h\toγγ$ decay amplitude induced by a light-quark loop, but additional complications arise because of external color charges. We show how the refactorization-based subtraction scheme developed in previous work leads to a factorization theorem free of endpoint divergences. We use renormalization-group techniques to predict the logarithmically enhanced terms in the three-loop $gg\to h$ form factor of order $α_s^3\ln^k(-M_h^2/m_b^2)$ with $k=6,5,4,3$. We also resum the first three towers of leading logarithms, $α_s^n\ln^{2n-k}(-M_h^2/m_b^2)$ with $k=0,1,2$, to all orders of perturbation theory.

hep-ph↗

arXiv2022

Hyperbolic Cosine Transformer for LiDAR 3D Object Detection

Recently, Transformer has achieved great success in computer vision. However, it is constrained because the spatial and temporal complexity grows quadratically with the number of large points in 3D object detection applications. Previous point-wise methods are suffering from time consumption and limited receptive fields to capture information among points. In this paper, we propose a two-stage hyperbolic cosine transformer (ChTR3D) for 3D object detection from LiDAR point clouds. The proposed ChTR3D refines proposals by applying cosh-attention in linear computation complexity to encode rich contextual relationships among points. The cosh-attention module reduces the space and time complexity of the attention operation. The traditional softmax operation is replaced by non-negative ReLU activation and hyperbolic-cosine-based operator with re-weighting mechanism. Extensive experiments on the widely used KITTI dataset demonstrate that, compared with vanilla attention, the cosh-attention significantly improves the inference speed with competitive performance. Experiment results show that, among two-stage state-of-the-art methods using point-level features, the proposed ChTR3D is the fastest one.

cs.CV↗

arXiv2022

Vector-Like top quark production via a chromo-magnetic moment at the LHC

Theories which provide a dynamical explanation for the large top-quark mass often include TeV-scale vector-like top-quark and bottom-quark partner states which can be potentially discovered at the LHC. These states are currently probed through model-independent searches for pair-production via gluon fusion, as well as through model-dependent complementary electroweak single production. In this paper we study the potential to extend those searches for the partners of the third-generation Standard Model quarks on the basis of their expected chromomagnetic interactions. We discuss how current searches for "excited" bottom-quarks produced via $b$-gluon fusion through chromomagnetic interactions are relevant, and provide significant constraints. We then explore the region of the parameter space in which the bottom-quark partner is heavier than the top-quark partner, in which case the top-partner can be primarily produced via the decay of the bottom-partner. Next, we probe the potential of the production of a single top-quark partner in association with an ordinary top-quark by gluon-fusion. Kinematically these two new processes are similar, and they yield the production of a heavy top partner and a lighter Standard Model state, a pattern which allows for the rejection of the associated dominant Standard Model backgrounds. We examine the sensitivity of these modes in the case where the top-partner subsequently decays to a Higgs boson and an ordinary top-quark, and we demonstrate that these new channels have the potential of extending and complementing the conventional strategies at LHC run III and at the high-luminosity phase of the LHC. In this last case, we find that partner masses that range up to about 3 TeV can be reached. This substantially expands the expected mass reach for these new states, including regions of parameter space that are inaccessible by traditional searches.

hep-ph↗

arXiv2022

Tencent AI Lab - Shanghai Jiao Tong University Low-Resource Translation System for the WMT22 Translation Task

This paper describes Tencent AI Lab - Shanghai Jiao Tong University (TAL-SJTU) Low-Resource Translation systems for the WMT22 shared task. We participate in the general translation task on English$\Leftrightarrow$Livonian. Our system is based on M2M100 with novel techniques that adapt it to the target language pair. (1) Cross-model word embedding alignment: inspired by cross-lingual word embedding alignment, we successfully transfer a pre-trained word embedding to M2M100, enabling it to support Livonian. (2) Gradual adaptation strategy: we exploit Estonian and Latvian as auxiliary languages for many-to-many translation training and then adapt to English-Livonian. (3) Data augmentation: to enlarge the parallel data for English-Livonian, we construct pseudo-parallel data with Estonian and Latvian as pivot languages. (4) Fine-tuning: to make the most of all available data, we fine-tune the model with the validation set and online back-translation, further boosting the performance. In model evaluation: (1) We find that previous work underestimated the translation performance of Livonian due to inconsistent Unicode normalization, which may cause a discrepancy of up to 14.9 BLEU score. (2) In addition to the standard validation set, we also employ round-trip BLEU to evaluate the models, which we find more appropriate for this task. Finally, our unconstrained system achieves BLEU scores of 17.0 and 30.4 for English to/from Livonian.

cs.CL↗

arXiv2022

Single Vector-Like top quark production via chromomagnetic interactions at present and future hadron colliders $-$A Snowmass 2021 White Paper

In our recent paper, we have investigated the potential for the LHC to discover vector-like quark partner states singly produced via their chromomagnetic moment interactions. These production mechanisms extend traditional searches which rely on pair-production of top-quark partner states or on the single production of these states through electroweak interactions, in the sense of providing greatly increased reach in parameter space regions where traditional searches are insensitive. In this study we determine the potential of both the 14 TeV high-luminosity LHC (HL-LHC) and a 100 TeV proton-proton collider to probe new vector-like quarks produced in this mode. We focus on the single production of a top-quark partner in association with an ordinary top-quark, as well as on the resonant production of the bottom-quark partner with its subsequent decay to a top-quark partner and a $W$ boson. For both cases we consider a top-partner decay to the Higgs boson and an ordinary top-quark. We find that HL-LHC and a future 100 TeV proton collider can probe vector-like partner masses up to about 3 TeV and 15-20 TeV respectively, visibly extending the range of the traditional vector like quark partner searches.

hep-ph↗

arXiv2022

STAD: Self-Training with Ambiguous Data for Low-Resource Relation Extraction

We present a simple yet effective self-training approach, named as STAD, for low-resource relation extraction. The approach first classifies the auto-annotated instances into two groups: confident instances and uncertain instances, according to the probabilities predicted by a teacher model. In contrast to most previous studies, which mainly only use the confident instances for self-training, we make use of the uncertain instances. To this end, we propose a method to identify ambiguous but useful instances from the uncertain instances and then divide the relations into candidate-label set and negative-label set for each ambiguous instance. Next, we propose a set-negative training method on the negative-label sets for the ambiguous instances and a positive training method for the confident instances. Finally, a joint-training method is proposed to build the final relation extraction system on all data. Experimental results on two widely used datasets SemEval2010 Task-8 and Re-TACRED with low-resource settings demonstrate that this new self-training approach indeed achieves significant and consistent improvements when comparing to several competitive self-training systems. Code is publicly available at https://github.com/jjyunlp/STAD

cs.CL↗

arXiv2022

Energy dissipation from confined states in nanoporous molecular networks

Crystalline nanoporous molecular networks are assembled on the Ag(111) surface, where the pores confine electrons originating from the surface state of the metal. Depending on the pore sizes and their coupling, an antibonding level is shifted upwards by 0.1 to 0.3 eV as measured by scanning tunneling microscopy. On molecular sites, a down-shifted bonding state is observed, which is occupied under equilibrium conditions. Low-temperature force spectroscopy reveals energy dissipation peaks and jumps of frequency shifts at bias voltages, which are related to the confined states. The dissipation maps show delocalization on the supra-molecular assembly and a weak distance-dependence of the dissipation peaks. These observations indicate that two-dimensional arrays of coupled quantum dots are formed, which are quantitatively characterized by their quantum capacitances and resonant tunneling rates. Our work provides a method for studying the capacitive and dissipative response of quantum materials with nanomechanical oscillators.

cond-mat.mtrl-sci↗

arXiv2022

Next-ViT: Next Generation Vision Transformer for Efficient Deployment in Realistic Industrial Scenarios

Due to the complex attention mechanisms and model design, most existing vision Transformers (ViTs) can not perform as efficiently as convolutional neural networks (CNNs) in realistic industrial deployment scenarios, e.g. TensorRT and CoreML. This poses a distinct challenge: Can a visual neural network be designed to infer as fast as CNNs and perform as powerful as ViTs? Recent works have tried to design CNN-Transformer hybrid architectures to address this issue, yet the overall performance of these works is far away from satisfactory. To end these, we propose a next generation vision Transformer for efficient deployment in realistic industrial scenarios, namely Next-ViT, which dominates both CNNs and ViTs from the perspective of latency/accuracy trade-off. In this work, the Next Convolution Block (NCB) and Next Transformer Block (NTB) are respectively developed to capture local and global information with deployment-friendly mechanisms. Then, Next Hybrid Strategy (NHS) is designed to stack NCB and NTB in an efficient hybrid paradigm, which boosts performance in various downstream tasks. Extensive experiments show that Next-ViT significantly outperforms existing CNNs, ViTs and CNN-Transformer hybrid architectures with respect to the latency/accuracy trade-off across various vision tasks. On TensorRT, Next-ViT surpasses ResNet by 5.5 mAP (from 40.4 to 45.9) on COCO detection and 7.7% mIoU (from 38.8% to 46.5%) on ADE20K segmentation under similar latency. Meanwhile, it achieves comparable performance with CSWin, while the inference speed is accelerated by 3.6x. On CoreML, Next-ViT surpasses EfficientFormer by 4.6 mAP (from 42.6 to 47.2) on COCO detection and 3.5% mIoU (from 45.1% to 48.6%) on ADE20K segmentation under similar latency. Our code and models are made public at: https://github.com/bytedance/Next-ViT

cs.CV↗

arXiv2022

The three-loop equal-mass banana integral in $\varepsilon$-factorised form with meromorphic modular forms

We show that the differential equation for the three-loop equal-mass banana integral can be cast into an $\varepsilon$-factorised form with entries constructed from (meromorphic) modular forms and one special function, which can be given as an iterated integral of meromorphic modular forms. The $\varepsilon$-factorised form of the differential equation allows for a systematic solution to any order in the dimensional regularisation parameter $\varepsilon$. The alphabet of the iterated integrals contains six letters.

hep-th↗

arXiv2022

Polygonal tessellations as predictive models of molecular monolayers

Molecular self-assembly plays a very important role in various aspects of technology as well as in biological systems. Governed by the covalent, hydrogen or van der Waals interactions - self-assembly of alike molecules results in a large variety of complex patterns even in two dimensions (2D). Prediction of pattern formation for 2D molecular networks is extremely important, though very challenging, and so far, relied on computationally involved approaches such as density functional theory, classical molecular dynamics, Monte Carlo, or machine learning. Such methods, however, do not guarantee that all possible patterns will be considered and often rely on intuition. Here we introduce a much simpler, though rigorous, hierarchical geometric model founded on the mean-field theory of 2D polygonal tessellations to predict extended network patterns based on molecular-level information. Based on graph theory, this approach yields pattern classification and pattern prediction within well-defined ranges. When applied to existing experimental data, our model provides an entirely new view of self-assembled molecular patterns, leading to interesting predictions on admissible patterns and potential additional phases. While developed for hydrogen-bonded systems, an extension to covalently bonded graphene-derived materials or 3D structures such as fullerenes is possible, significantly opening the range of potential future applications.

cond-mat.mtrl-sci↗

arXiv2022

Target density effects on charge tansfer of laser-accelerated carbon ions in dense plasma

We report on charge state measurements of laser-accelerated carbon ions in the energy range of several MeV penetrating a dense partially ionized plasma. The plasma was generated by irradiation of a foam target with laser-induced hohlraum radiation in the soft X-ray regime. We used the tri-cellulose acetate (C$_{9}$H$_{16}$O$_{8}$) foam of 2 mg/cm$^{-3}$ density, and $1$-mm interaction length as target material. This kind of plasma is advantageous for high-precision measurements, due to good uniformity and long lifetime compared to the ion pulse length and the interaction duration. The plasma parameters were diagnosed to be T$_{e}$=17 eV and n$_{e}$=4 $\times$ 10$^{20}$ cm$^{-3}$. The average charge states passing through the plasma were observed to be higher than those predicted by the commonly-used semiempirical formula. Through solving the rate equations, we attribute the enhancement to the target density effects which will increase the ionization rates on one hand and reduce the electron capture rates on the other hand. In previsous measurement with partially ionized plasma from gas discharge and z-pinch to laser direct irradiation, no target density effects were ever demonstrated. For the first time, we were able to experimentally prove that target density effects start to play a significant role in plasma near the critical density of Nd-Glass laser radiation. The finding is important for heavy ion beam driven high energy density physics and fast ignitions.

physics.plasm-ph↗

arXiv2022

TRT-ViT: TensorRT-oriented Vision Transformer

We revisit the existing excellent Transformers from the perspective of practical application. Most of them are not even as efficient as the basic ResNets series and deviate from the realistic deployment scenario. It may be due to the current criterion to measure computation efficiency, such as FLOPs or parameters is one-sided, sub-optimal, and hardware-insensitive. Thus, this paper directly treats the TensorRT latency on the specific hardware as an efficiency metric, which provides more comprehensive feedback involving computational capacity, memory cost, and bandwidth. Based on a series of controlled experiments, this work derives four practical guidelines for TensorRT-oriented and deployment-friendly network design, e.g., early CNN and late Transformer at stage-level, early Transformer and late CNN at block-level. Accordingly, a family of TensortRT-oriented Transformers is presented, abbreviated as TRT-ViT. Extensive experiments demonstrate that TRT-ViT significantly outperforms existing ConvNets and vision Transformers with respect to the latency/accuracy trade-off across diverse visual tasks, e.g., image classification, object detection and semantic segmentation. For example, at 82.7% ImageNet-1k top-1 accuracy, TRT-ViT is 2.7$\times$ faster than CSWin and 2.0$\times$ faster than Twins. On the MS-COCO object detection task, TRT-ViT achieves comparable performance with Twins, while the inference speed is increased by 2.8$\times$.

cs.CV↗

arXiv2022

Supersymmetry and Sum Rules in the Goldberger-Wise Model

In this work we demonstrate that the mixed gravitational and scalar sectors of the five-dimensional Goldberger-Wise (GW) model, in which the size of a warped extra dimension is dynamically determined, has a "hidden" dual $N=2$ supersymmetric structure. This symmetry structure, a generalization of one found in the unstabilized Randall-Sundrum model, is a result of the spontaneously broken five-dimensional diffeomorphism invariance of the underlying gravitational theory. The supersymmetries relate the properties of the spin-1 and spin-0 modes "eaten" by the massive spin-2 Kaluza-Klein states of the theory to the mode functions of the spin-2 modes. Because the symmetries relate the couplings and masses of the massive spin-2 states to those of the tower of physical spin-0 states of the GW model, they enable us to analytically prove the sum rule relations which ensure the tree-level scattering amplitudes of the massive spin-2 states will grow no faster than ${\cal O}(s)$. The analysis given here also explains the unconventional forms of the spin-0 mode equation, boundary condition(s), and normalization found in the GW model.

hep-ph↗

arXiv2022

Sepsis Prediction with Temporal Convolutional Networks

We design and implement a temporal convolutional network model to predict sepsis onset. Our model is trained on data extracted from MIMIC III database, based on a retrospective analysis of patients admitted to intensive care unit who did not fall under the definition of sepsis at the time of admission. Benchmarked with several machine learning models, our model is superior on this binary classification task, demonstrates the prediction power of convolutional networks for temporal patterns, also shows the significant impact of having longer look back time on sepsis prediction.

cs.LG↗

arXiv2022

Reliable and Efficient Broadcast Routing Using Multipoint Relays Over VANET For Vehicle Platooning

In this paper, we design and implement a reliable broadcast algorithm over a VANET for supporting multi-hop forwarding of vehicle sensor and control packets that will enable vehicles to platoon with each other in order to form a road train behind the lead truck. In particular, we use multipoint relays (MPRs) for packet transmission, which leads to more efficient communication in a VANET. We evaluate the performance based on simulation by running a platooning simulation application program, and show that with MPRs, the communication in the VANET to form a road train is more efficient and reliable.

cs.NI↗

arXiv2022

Asynchronous Hierarchical Federated Learning

Federated Learning is a rapidly growing area of research and with various benefits and industry applications. Typical federated patterns have some intrinsic issues such as heavy server traffic, long periods of convergence, and unreliable accuracy. In this paper, we address these issues by proposing asynchronous hierarchical federated learning, in which the central server uses either the network topology or some clustering algorithm to assign clusters for workers (i.e., client devices). In each cluster, a special aggregator device is selected to enable hierarchical learning, leads to efficient communication between server and workers, so that the burden of the server can be significantly reduced. In addition, asynchronous federated learning schema is used to tolerate heterogeneity of the system and achieve fast convergence, i.e., the server aggregates the gradients from the workers weighted by a staleness parameter to update the global model, and regularized stochastic gradient descent is performed in workers, so that the instability of asynchronous learning can be alleviated. We evaluate the proposed algorithm on CIFAR-10 image classification task, the experimental results demonstrate the effectiveness of asynchronous hierarchical federated learning.

cs.LG↗