arXiv subjects

Chen Xu

Publications and source records attributed to Chen Xu.

At least 217 records · Page 12Linked to original sources

arXiv2021

Cooling in poor air quality environments -- Impact of fan operation on particle deposition

Environmental pollutants are a source for reliability issues across data center and telecommunications equipment. A primary driver of this is the transport and deposition of particle matter (PM$_{2.5}$, PM$_{10}$) on printed circuit boards, electronic components and heat exchange surfaces. This process is enhanced by turbulent air flows generated from cooling fans. Particle pollutants can persist after contemporary filtering, highlighting the importance of elucidating particle transport mechanisms and utilising this information to design robust equipment. This study investigates particle transport behaviour arising from axial fans operating under varied aerodynamic conditions. Transient, multi-phase numerical simulations were performed to model the flow of millions of microscale particles in air and determine their fate. Across a comprehensive range of fan operation conditions, from aerodynamic stall to free delivery, non-dimensional deposition velocities spanned an order of magnitude. Deposition profiles vary from monotonic to non-monotonic behaviour, influenced by local flow impingement, blade tip vortices, and shear velocity. A simple flow control solution that mitigates the factors influencing deposition has been demonstrated for equipment already deployed. The findings and numerical methods can be applied for the optimization of fan-cooled equipment intended for indoor and outdoor environments where air quality is poor, or pollution levels are high.

physics.flu-dyn↗

arXiv2021

Estimating Adsorption Isotherm Parameters in Chromatography via a Virtual Injection Promoting Feed-forward Neural Network

The means to obtain the adsorption isotherms is a fundamental open problem in competitive chromatography. A modern technique of estimating adsorption isotherms is to solve an inverse problem so that the simulated batch separation coincides with actual experimental results. However, this identification process is usually ill-posed in the sense that the small noise in the measured response can lead to a large fluctuation in the estimated quantity of adsorption isotherms. The conventional mathematical method of solving this problem is the variational regularization, which is formulated as a non-convex minimization problem with a regularized objective functional. However, in this method, the choice of regularization parameter and the design of a convergent solution algorithm are quite difficult in practice. Moreover, due to the restricted number of injection profiles in experiments, the types of measured data are extremely limited, which may lead to a biased estimation. In order to overcome these difficulties, in this paper, we develop a new inversion method -- the Virtual Injection Promoting Feed-forward Neural Network (VIP-FNN). In this approach, the training data contain various types of artificial injections and synthetic noisy measurement at outlet, generated by a conventional physics model -- a time-dependent convection-diffusion system. Numerical experiments with both artificial and real data from laboratory experiments show that the proposed VIP-FNN is an efficient and robust algorithm.

eess.SP↗

arXiv2021

HAVANA: Hierarchical and Variation-Normalized Autoencoder for Person Re-identification

Person Re-Identification (Re-ID) is of great importance to the many video surveillance systems. Learning discriminative features for Re-ID remains a challenge due to the large variations in the image space, e.g., continuously changing human poses, illuminations and point of views. In this paper, we propose HAVANA, a novel extensible, light-weight HierArchical and VAriation-Normalized Autoencoder that learns features robust to intra-class variations. In contrast to existing generative approaches that prune the variations with heavy extra supervised signals, HAVANA suppresses the intra-class variations with a Variation-Normalized Autoencoder trained with no additional supervision. We also introduce a novel Jensen-Shannon triplet loss for contrastive distribution learning in Re-ID. In addition, we present Hierarchical Variation Distiller, a hierarchical VAE to factorize the latent representation and explicitly model the variations. To the best of our knowledge, HAVANA is the first VAE-based framework for person ReID.

cs.CV↗

arXiv2020

Dynamic Curriculum Learning for Low-Resource Neural Machine Translation

Large amounts of data has made neural machine translation (NMT) a big success in recent years. But it is still a challenge if we train these models on small-scale corpora. In this case, the way of using data appears to be more important. Here, we investigate the effective use of training data for low-resource NMT. In particular, we propose a dynamic curriculum learning (DCL) method to reorder training samples in training. Unlike previous work, we do not use a static scoring function for reordering. Instead, the order of training samples is dynamically determined in two ways - loss decline and model competence. This eases training by highlighting easy samples that the current model has enough competence to learn. We test our DCL method in a Transformer-based system. Experimental results show that DCL outperforms several strong baselines on three low-resource machine translation benchmarks and different sized data of WMT' 16 En-De.

cs.CL↗

arXiv2020

Predictive Inference Is Free with the Jackknife+-after-Bootstrap

Ensemble learning is widely used in applications to make predictions in complex decision problems---for example, averaging models fitted to a sequence of samples bootstrapped from the available training data. While such methods offer more accurate, stable, and robust predictions and model estimates, much less is known about how to perform valid, assumption-lean inference on the output of these types of procedures. In this paper, we propose the jackknife+-after-bootstrap (J+aB), a procedure for constructing a predictive interval, which uses only the available bootstrapped samples and their corresponding fitted models, and is therefore "free" in terms of the cost of model fitting. The J+aB offers a predictive coverage guarantee that holds with no assumptions on the distribution of the data, the nature of the fitted model, or the way in which the ensemble of models are aggregated---at worst, the failure rate of the predictive interval is inflated by a factor of 2. Our numerical experiments verify the coverage and accuracy of the resulting predictive intervals on real data.

stat.ME↗

arXiv2020

Wasserstein Distance Regularized Sequence Representation for Text Matching in Asymmetrical Domains

One approach to matching texts from asymmetrical domains is projecting the input sequences into a common semantic space as feature vectors upon which the matching function can be readily defined and learned. In real-world matching practices, it is often observed that with the training goes on, the feature vectors projected from different domains tend to be indistinguishable. The phenomenon, however, is often overlooked in existing matching models. As a result, the feature vectors are constructed without any regularization, which inevitably increases the difficulty of learning the downstream matching functions. In this paper, we propose a novel match method tailored for text matching in asymmetrical domains, called WD-Match. In WD-Match, a Wasserstein distance-based regularizer is defined to regularize the features vectors projected from different domains. As a result, the method enforces the feature projection function to generate vectors such that those correspond to different domains cannot be easily discriminated. The training process of WD-Match amounts to a game that minimizes the matching loss regularized by the Wasserstein distance. WD-Match can be used to improve different text matching methods, by using the method as its underlying matching model. Four popular text matching methods have been exploited in the paper. Experimental results based on four publicly available benchmarks showed that WD-Match consistently outperformed the underlying methods and the baselines.

cs.CL↗

arXiv2020

Non-Markovian recovery makes complex networks more resilient against large-scale failures

Non-Markovian spontaneous recovery processes with a time delay (memory) are ubiquitous in the real world. How does the non-Markovian characteristic affect failure propagation in complex networks? We consider failures due to internal causes at the nodal level and external failures due to an adverse environment, and develop a pair approximation analysis taking into account the two-node correlation. In general, a high failure stationary state can arise, corresponding to large-scale failures that can significantly compromise the functioning of the network. We uncover a striking phenomenon: memory associated with nodal recovery can counter-intuitively make the network more resilient against large-scale failures. In natural systems, the intrinsic non-Markovian characteristic of nodal recovery may thus be one reason for their resilience. In engineering design, incorporating certain non-Markovian features into the network may be beneficial to equipping it with a strong resilient capability to resist catastrophic failures.

physics.soc-ph↗

arXiv2020

Deposition of particle pollution in turbulent forced-air cooling

Rotating fans are the prevalent forced cooling method for heat generating equipment and buildings. As the concentration of atmospheric pollutants has increased, the accumulation of microscale and nanoscale particles on surfaces due to advection-diffusion has led to adverse mechanical, chemical and electrical effects that increase cooling demands and reduce the reliability of electronic equipment. Here, we uncover the mechanisms leading to enhanced deposition of particle matter (PM$_{10}$ and PM$_{2.5}$) on surfaces due to turbulent axial fan flows operating at Reynolds numbers, $Re \sim 10^5$. Qualitative observations of long-term particle deposition from the field were combined with \textit{in situ} particle image velocimetry on a telecommunications base station, revealing the dominant role of impingement velocity and angle. Near-wall momentum transport for $10 < y^+ < 50$ were explored using a quadrant analysis to uncover the contributions of turbulent events that promote particle deposition through turbulent diffusion and eddy impaction. By decomposing these events, the local transport behaviour of fine particles from the bulk flow to the surface has been categorised. The transition from deposition to clean surfaces was accompanied by a decrease in shear velocity, turbulent stresses, and particle sweep motions with lower flux in the wall-normal direction. Finally, using these insights, selective filtering of coarse particles was found to promote the conditions that enhance the deposition of fine particle matter.

physics.flu-dyn↗

arXiv2020

Privileged Features Distillation at Taobao Recommendations

Features play an important role in the prediction tasks of e-commerce recommendations. To guarantee the consistency of off-line training and on-line serving, we usually utilize the same features that are both available. However, the consistency in turn neglects some discriminative features. For example, when estimating the conversion rate (CVR), i.e., the probability that a user would purchase the item if she clicked it, features like dwell time on the item detailed page are informative. However, CVR prediction should be conducted for on-line ranking before the click happens. Thus we cannot get such post-event features during serving. We define the features that are discriminative but only available during training as the privileged features. Inspired by the distillation techniques which bridge the gap between training and inference, in this work, we propose privileged features distillation (PFD). We train two models, i.e., a student model that is the same as the original one and a teacher model that additionally utilizes the privileged features. Knowledge distilled from the more accurate teacher is transferred to the student to improve its accuracy. During serving, only the student part is extracted and it relies on no privileged features. We conduct experiments on two fundamental prediction tasks at Taobao recommendations, i.e., click-through rate (CTR) at coarse-grained ranking and CVR at fine-grained ranking. By distilling the interacted features that are prohibited during serving for CTR and the post-event features for CVR, we achieve significant improvements over their strong baselines. During the on-line A/B tests, the click metric is improved by +5.0% in the CTR task. And the conversion metric is improved by +2.3% in the CVR task. Besides, by addressing several issues of training PFD, we obtain comparable training speed as the baselines without any distillation.

cs.IR↗

arXiv2020

Multiplicative Noise Removal: Nonlocal Low-Rank Model and Its Proximal Alternating Reweighted Minimization Algorithm

The goal of this paper is to develop a novel numerical method for efficient multiplicative noise removal. The nonlocal self-similarity of natural images implies that the matrices formed by their nonlocal similar patches are low-rank. By exploiting this low-rank prior with application to multiplicative noise removal, we propose a nonlocal low-rank model for this task and develop a proximal alternating reweighted minimization (PARM) algorithm to solve the optimization problem resulting from the model. Specifically, we utilize a generalized nonconvex surrogate of the rank function to regularize the patch matrices and develop a new nonlocal low-rank model, which is a nonconvex nonsmooth optimization problem having a patchwise data fidelity and a generalized nonlocal low-rank regularization term. To solve this optimization problem, we propose the PARM algorithm, which has a proximal alternating scheme with a reweighted approximation of its subproblem. A theoretical analysis of the proposed PARM algorithm is conducted to guarantee its global convergence to a critical point. Numerical experiments demonstrate that the proposed method for multiplicative noise removal significantly outperforms existing methods such as the benchmark SAR-BM3D method in terms of the visual quality of the denoised images, and the PSNR (the peak-signal-to-noise ratio) and SSIM (the structural similarity index measure) values.

math.OC↗

arXiv2020

Metasurface integrated Vertical Cavity Surface Emitting Lasers for programmable directional lasing emissions

Vertical cavity surface emitting lasers (VCSELs) have made indispensable contributions to the development of modern optoelectronic technologies. However, arbitrary beam shaping of VCSELs within a compact system still remains inaccessible till now. The emerging ultra-thin flat optical structures, namely metasurfaces, offer a powerful technique to manipulate electromagnetic fields with subwavelength spatial resolution. Here, we show that the monolithic integration of dielectric metasurfaces with VCSELs enables remarkable arbitrary control of the laser beam profiles, including self-collimation, Bessel and Vortex lasers, with high efficiency. Such wafer-level integration of metasurface through VCSELs-compatible technology simplifies the assembling process and preserves the high performance of the VCSELs. We envision that our approach can be implemented in various wide-field applications, such as optical fibre communications, laser printing, smartphones, optical sensing, face recognition, directional displays and ultra-compact light detection and ranging (LiDAR).

physics.optics↗

arXiv2019

Megahertz single-particle imaging at the European XFEL

The emergence of high repetition-rate X-ray free-electron lasers (XFELs) powered by superconducting accelerator technology enables the measurement of significantly more experimental data per day than was previously possible. The European XFEL will soon provide 27,000 pulses per second, more than two orders of magnitude more than any other XFEL. The increased pulse rate is a key enabling factor for single-particle X-ray diffractive imaging, which relies on averaging the weak diffraction signal from single biological particles. Taking full advantage of this new capability requires that all experimental steps, from sample preparation and delivery to the acquisition of diffraction patterns, are compatible with the increased pulse repetition rate. Here, we show that single-particle imaging can be performed using X-ray pulses at megahertz repetition rates. The obtained results pave the way towards exploiting high repetition-rate X-ray free-electron lasers for single-particle imaging at their full repetition rate.

physics.ins-det↗

arXiv2019

Graphene-assisted preparation of large-scale single crystal Ag(111) nanoparticle arrays

Surface plasmon resonance of metal nanostructures has broad application prospects in the fields of photocatalysis, optical sensing, biomarkers and surface-enhanced Raman scattering. This paper reports a graphene-assisted method for preparing large-scale single crystal Ag(111) nanoparticle arrays based on ion implantation technique. By surface periodic treatment and annealing of the implanted sample, regularly arranged Ag nanoparticles can be prepared on the sample surface. A new application for graphene is proposed, that is, as a perfect barrier layer to prevent metal atoms from evaporating or diffusing. All the Ag NPs show (111) crystal orientation. Besides, the Ag atoms are covered by graphene immediately when they precipitate from the substrate, which can prevent them from being oxidized. On the basis of this structure, as one of the applications of metal SPR, we measured the Raman enhancement effect, and found that the G peak of the Raman spectrum of graphene achieved about 20 times enhancement.

physics.app-ph↗

arXiv2019

SAIS: Single-stage Anchor-free Instance Segmentation

In this paper, we propose a simple yet efficientinstance segmentation approach based on the single-stage anchor-free detector, termed SAIS. In our approach, the instancesegmentation task consists of two parallel subtasks which re-spectively predict the mask coefficients and the mask prototypes.Then, instance masks are generated by linearly combining theprototypes with the mask coefficients. To enhance the quality ofinstance mask, the information from regression and classificationis fused to predict the mask coefficients. In addition, center-aware target is designed to preserve the center coordination ofeach instance, which achieves a stable improvement in instancesegmentation. The experiment on MS COCO shows that SAISachieves the performance of the exiting state-of-the-art single-stage methods with a much less memory footpr

cs.CV↗

arXiv2019

Buffer-aware Wireless Scheduling based on Deep Reinforcement Learning

In this paper, the downlink packet scheduling problem for cellular networks is modeled, which jointly optimizes throughput, fairness and packet drop rate. Two genie-aided heuristic search methods are employed to explore the solution space. A deep reinforcement learning (DRL) framework with A2C algorithm is proposed for the optimization problem. Several methods have been utilized in the framework to improve the sampling and training efficiency and to adapt the algorithm to a specific scheduling problem. Numerical results show that DRL outperforms the baseline algorithm and achieves similar performance as genie-aided methods without using the future information.

cs.IT↗

arXiv2019

Tunable anisotropic perfect absorption in hyperbolic metamaterials based on black phosphorous/dielectric multilayer structures

Black phosphorus (BP) as a promising two-dimensional material with extraordinary optical properties constitutes an excellent building block in multilayer hyperbolic metamaterials. In this work, we design a multilayer structure composed of BP/dielectric layer stacking unit cells patterned on a gold mirror and theoretically demonstrate the tunable anisotropic absorption in the infrared regime. The electric dipole resonance between the adjacent unit cells drives the structure in the critical coupling state, and impedance of the structure matches to that of free space, showing the perfect absorption for one polarization direction, while the impedance mismatch for the other polarization direction leads to only 8.2$\%$ absorption at the same wavelength. The anisotropic absorption response of the proposed structure can be attributed to the intrinsic anisotropy of BP, which exhibits few dependence on the incident angle. Furthermore, we investigate the tunable optical absorption of the proposed structure with the electron doping of BP and the geometric parameters. These results demonstrate great potentials of BP in constituting multilayer hyperbolic metamaterials, and open up avenues in designing anisotropic metadevices with tunable spectral and polarization selectivity in the infrared regime.

physics.optics↗

arXiv2019

PR Product: A Substitute for Inner Product in Neural Networks

In this paper, we analyze the inner product of weight vector w and data vector x in neural networks from the perspective of vector orthogonal decomposition and prove that the direction gradient of w decreases with the angle between them close to 0 or π. We propose the Projection and Rejection Product (PR Product) to make the direction gradient of w independent of the angle and consistently larger than the one in standard inner product while keeping the forward propagation identical. As a reliable substitute for standard inner product, the PR Product can be applied into many existing deep learning modules, so we develop the PR Product version of fully connected layer, convolutional layer and LSTM layer. In static image classification, the experiments on CIFAR10 and CIFAR100 datasets demonstrate that the PR Product can robustly enhance the ability of various state-of-the-art classification networks. On the task of image captioning, even without any bells and whistles, our PR Product version of captioning model can compete or outperform the state-of-the-art models on MS COCO dataset. Code has been made available at:https://github.com/wzn0828/PR_Product.

cs.CV↗

arXiv2019

Deep Reinforcement Learning for Scheduling in Cellular Networks

Integrating artificial intelligence (AI) into wireless networks has drawn significant interest in both industry and academia. A common solution is to replace partial or even all modules in the conventional systems, which is often lack of efficiency and robustness due to their ignoring of expert knowledge. In this paper, we take deep reinforcement learning (DRL) based scheduling as an example to investigate how expert knowledge can help with AI module in cellular networks. A simulation platform, which has considered link adaption, feedback and other practical mechanisms, is developed to facilitate the investigation. Besides the traditional way, which is learning directly from the environment, for training DRL agent, we propose two novel methods, i.e., learning from a dual AI module and learning from the expert solution. The results show that, for the considering scheduling problem, DRL training procedure can be improved on both performance and convergence speed by involving the expert knowledge. Hence, instead of replacing conventional scheduling module in the system, adding a newly introduced AI module, which is capable to interact with the conventional module and provide more flexibility, is a more feasible solution.

cs.IT↗