Search arXiv⌕ Search

arXiv subjects

Peng Zhang

Publications and source records attributed to Peng Zhang.

At least 703 records · Page 39Linked to original sources
arXiv2019

GraphNAS: Graph Neural Architecture Search with Reinforcement Learning

Graph Neural Networks (GNNs) have been popularly used for analyzing non-Euclidean data such as social network data and biological data. Despite their success, the design of graph neural networks requires a lot of manual work and domain knowledge. In this paper, we propose a Graph Neural Architecture Search method (GraphNAS for short) that enables automatic search of the best graph neural architecture based on reinforcement learning. Specifically, GraphNAS first uses a recurrent network to generate variable-length strings that describe the architectures of graph neural networks, and then trains the recurrent network with reinforcement learning to maximize the expected accuracy of the generated architectures on a validation data set. Extensive experimental results on node classification tasks in both transductive and inductive learning settings demonstrate that GraphNAS can achieve consistently better performance on the Cora, Citeseer, Pubmed citation network, and protein-protein interaction network. On node classification tasks, GraphNAS can design a novel network architecture that rivals the best human-invented architecture in terms of test set accuracy.

cs.LG↗
arXiv2019

Interaction Control of Ultracold Alkaline-Earth Atoms

Ultracold alkaline-earth atoms have now been widely explored for precision measurements and quantum simulation. Because of its unique atomic structure, alkaline earth atoms possess great advantages for quantum simulation and studying quantum many-body matters, such as simulating synthetic gauge field, Kondo physics and $SU(N)$ physics. To fully explore the potential of ultracold alkaline-earth atoms, these systems also need to be equipped with the capability of tuning the inter-atomic interaction to the strongly interacting regime. Recently several theoretical proposals and experimental demonstrations have shown that both spin-independent and spin-exchanging interaction can be tuned to resonance. In this perspective, we will review these progress and discuss the new opportunities brought by these interaction control tools for future quantum simulation studies with ultracold alkaline-earth atoms.

cond-mat.quant-gas↗
arXiv2019

Unsupervised Online Video Object Segmentation with Motion Property Understanding

Unsupervised video object segmentation aims to automatically segment moving objects over an unconstrained video without any user annotation. So far, only few unsupervised online methods have been reported in literature and their performance is still far from satisfactory, because the complementary information from future frames cannot be processed under online setting. To solve this challenging problem, in this paper, we propose a novel Unsupervised Online Video Object Segmentation (UOVOS) framework by construing the motion property to mean moving in concurrence with a generic object for segmented regions. By incorporating salient motion detection and object proposal, a pixel-wise fusion strategy is developed to effectively remove detection noise such as dynamic background and stationary objects. Furthermore, by leveraging the obtained segmentation from immediately preceding frames, a forward propagation algorithm is employed to deal with unreliable motion detection and object proposals. Experimental results on several benchmark datasets demonstrate the efficacy of the proposed method. Compared to the state-of-the-art unsupervised online segmentation algorithms, the proposed method achieves an absolute gain of 6.2%. Moreover, our method achieves better performance than the best unsupervised offline algorithm on the DAVIS-2016 benchmark dataset. Our code is available on the project website: https://github.com/visiontao/uovos.

cs.CV↗
arXiv2019

REAPS: Towards Better Recognition of Fine-grained Images by Region Attending and Part Sequencing

Fine-grained image recognition has been a hot research topic in computer vision due to its various applications. The-state-of-the-art is the part/region-based approaches that first localize discriminative parts/regions, and then learn their fine-grained features. However, these approaches have some inherent drawbacks: 1) the discriminative feature representation of an object is prone to be disturbed by complicated background; 2) it is unreasonable and inflexible to fix the number of salient parts, because the intended parts may be unavailable under certain circumstances due to occlusion or incompleteness, and 3) the spatial correlation among different salient parts has not been thoroughly exploited (if not completely neglected). To overcome these drawbacks, in this paper we propose a new, simple yet robust method by building part sequence model on the attended object region. Concretely, we first try to alleviate the background effect by using a region attention mechanism to generate the attended region from the original image. Then, instead of localizing different salient parts and extracting their features separately, we learn the part representation implicitly by applying a mapping function on the serialized features of the object. Finally, we combine the region attending network and the part sequence learning network into a unified framework that can be trained end-to-end with only image-level labels. Our extensive experiments on three fine-grained benchmarks show that the proposed method achieves the state of the art performance.

cs.CV↗
arXiv2019

Plate-like precipitate effects on plasticity of Al-Cu micro-pillar: {100}-interfacial slip

In this paper, we study the effects of $θ^\prime$-Al$_2$Cu plate-like precipitates on the plasticity of Al-Cu micro-pillars, with a sample size allowing the precipitates to cross the entire micro-pillar. {100}-slip traces are identified for the first time in Al and Al alloys at room temperature. We investigate the underlying mechanisms of this unusual {100}-slip, and show that it operates along the coherent $θ^\prime$-Al$_2$Cu precipitate/$α$-Al matrix interface. A combination of molecular dynamics simulations and stress analysis indicates that screw dislocations can cross-slip from the {111} plane onto the {100} $θ^\prime$-Al$_2$Cu/$α$-Al interface, then move on it through a kink-pair mechanism, providing a reasonable explanation to the observed {100}-slips. The roles of the $θ^\prime$-Al$_2$Cu/$α$-Al matrix interface on the properties of interfacial dislocations are studied within the Peierls-Nabarro framework, showing that the interface can stabilize the {100} screw dislocations from the spreading of the core, and increases the Peierls stress. These results improve our understanding of the mechanical behavior of Al-Cu micro-pillars at room temperature, and imply an enhanced role of interfacial slip in Al-Cu based alloys at elevated temperature in consideration of the underlying kink-pair mechanism.

cond-mat.mtrl-sci↗
arXiv2019

Plate-like precipitate effects on plasticity of Al-Cu alloys at micrometer to submicrometer scales

The continuous miniaturization of modern electromechanical systems calls for a comprehensive understanding of the mechanical properties of metallic materials specific to micrometer and sub-micrometer scales. At these scales, the nature of dislocation-mediated plasticity changes radically: sub-micrometer metallic samples exhibit high yield strengths, however accompanied by detrimental intermittent strain fluctuations compromising forming processes and endangering structural stability. In this paper, we studied the effects of plate-like $θ^\prime$-Al$_2$Cu precipitates on the strength, plastic fluctuations and deformation mechanisms of Al-Cu alloys from micro-pillar compression testing. The plate-like precipitates have diameters commensurate with the external size of the Al-Cu micro-pillars. Our results show that these plate-like precipitates can strengthen the materials and suppress plastic fluctuations efficiently at large sample sizes ($\geq 3 μm$). However, the breakdown of the mean-field pinning landscape at smaller scales weakens its taming effect on intermittency. Over an intermediate range of sample sizes allowing the precipitates to cross the entire pillar, an enhanced apparent strain hardening and a sharp decrease of jerkiness are observed, in association with the presence of {100}-slip traces along the coherent $θ^\prime$-Al$_2$Cu precipitate/$α$-Al matrix interface and precipitate shearing. These complex effects of plate-like precipitates on plasticity are analyzed, experimentally and theoretically, in view of the interferences between external and internal sizes, and the related modifications of the underlying plastic mechanisms.

cond-mat.mtrl-sci↗
arXiv2019

A new Majorana platform in an Fe-As bilayer superconductor

Recently, iron-chalcogenide superconductors have emerged as a new and promising platform for studying and manipulating Majorana zero mode (MZM). By combining topological band structure and superconductivity in a multiband material, they provide significant advantages such as higher superconducting transition temperature (Tc) and isolated Majorana mode. However, iron-chalcogenide superconductors, especially Fe(Te,Se), suffer from strong inhomogeneity which may hamper their practical application. On the other hand, some iron-pnictide (Fe-As) superconductors, such as LiFeAs, have been demonstrated to have a similar topological band structure, yet no MZM has been observed in its vortex cores, raising a question of universality of MZM presence in iron-based superconductors. In this work, by using high-resolution angle-resolved photoemission spectroscopy and scanning tunneling microscopy/spectroscopy, we identify the first Fe-As superconductor CaKFe4As4 (Tc = 35 K) which has both the superconducting Dirac surface states and the MZMs inside its vortex cores. The topological band inversion is largely due to the down-shift of the pz band caused by the bilayer band folding in this material. More strikingly, the energies and spatial line profiles of MZM and multiple quantized Caroli-de Gennes-Matricon bound states observed inside the topological vortex can be accurately reproduced by a simple theoretical model derived from a surface Dirac cone, firmly establishing Majorana nature of the zero mode.

cond-mat.supr-con↗
arXiv2019

Unified Viscous-to-inertial Scaling in Liquid Droplet Coalescence

This letter presents a theory on the coalescence of two spherical liquid droplets that are initially stationary. The evolution of the radius of a liquid neck formed upon coalescence was formulated as an initial value problem and then solved to yield an exact solution without free parameters, with its two asymptotic approximations reproducing the well-known scaling relations in the viscous and inertial regimes. The viscous-to-inertial crossover observed by Paulsen et al. [Phys. Rev. Lett. 106, 114501 (2011)] is also recovered by the theory, rendering the collapse of data of different viscosities onto a single curve.

physics.flu-dyn↗
arXiv2019

Using Restart Heuristics to Improve Agent Performance in Angry Birds

Over the past few years the Angry Birds AI competition has been held in an attempt to develop intelligent agents that can successfully and efficiently solve levels for the video game Angry Birds. Many different agents and strategies have been developed to solve the complex and challenging physical reasoning problems associated with such a game. However none of these agents attempt one of the key strategies which humans employ to solve Angry Birds levels, which is restarting levels. Restarting is important in Angry Birds because sometimes the level is no longer solvable or some given shot made has little to no benefit towards the ultimate goal of the game. This paper proposes a framework and experimental evaluation for when to restart levels in Angry Birds. We demonstrate that restarting is a viable strategy to improve agent performance in many cases.

cs.AI↗
arXiv2019

Two-body bound state of ultracold Fermi atoms with two-dimensional spin-orbit coupling

In a recent experiment, a two-dimensional spin-orbit coupling (SOC) was realized for fermions in the continuum [Nat. Phys. 12, 540 (2016)], which represents an important step forward in the study of synthetic gauge field using cold atoms. In the experiment, it was shown that a Raman-induced two-dimensional SOC exists in the dressed-state basis close to a Dirac point of the single-particle spectrum. By contrast, the short-range inter-atomic interactions of the system are typically expressed in the hyperfine-spin basis. The interplay between synthetic SOC and interactions can potentially lead to interesting few- and many-body phenomena but has so far eluded theoretical attention. Here we study in detail properties of two-body bound states of such a system. We find that, due to the competition between SOC and interaction, the stability region of the two-body bound state is in general reduced. Particularly, the threshold of the lowest two-body bound state is shifted to a positive, SOC-dependent scattering length. Furthermore, the center-of-mass momentum of the lowest two-body bound state becomes nonzero, suggesting the emergence of Fulde-Ferrell pairing states in a many-body setting. Our results reveal the critical difference between the experimentally realized two-dimensional SOC and the more symmetric Rashba or Dresselhaus SOCs in an interacting system, and paves the way for future characterizations of topological superfluid states in the experimentally relevant systems.

cond-mat.quant-gas↗
arXiv2019

Multicore fiber-based quantum access network

We propose a quantum access network based on multicore fiber (MCF) to scale up the number of users in quantum key distribution (QKD) networks. The MCF is used as feeder fiber and single core single mode fibers (SSMFs) are used as drop fibers. Quantum signals (QSs) are integrated with classical signals (CSs) in both MCF and SSMFs to save deployment cost since access networks are cost sensitive. Due to the integration, spontaneous Raman scattering (SRS) and intercore crosstalk (ICXT) are the main impairment sources to QKD. To alleviate the noise, we propose a core and wavelength assignment scheme (CWAS) that a dedicated core of MCF is used to transmit QSs and the wavelengths of QSs are set lower than those of upstream signals. Also, we demonstrate that wavelength time division multiplexing (WTDM) is suitable for QSs which are required to support large number of quantum users, since WTDM can realize higher secure key rate (SKR) than time division multiplexing (TDM) and require lower cost than wavelength division multiplexing (WDM). Finally, the proposed quantum access network is verified experimentally. The experiment results are consistent with our analysis. The properties of SRS in the proposed architecture are shown in the experiments through the quantum bit error rates (QBERs) in different experimental case, which verifies the superiority of the proposed CWAS. Also, the characteristics of the SKRs prove that the number of receivers has a great impact on the performance of QSs using WTDM.

quant-ph↗
arXiv2019

Support Relation Analysis for Objects in Multiple View RGB-D Images

Understanding physical relations between objects, especially their support relations, is crucial for robotic manipulation. There has been work on reasoning about support relations and structural stability of simple configurations in RGB-D images. In this paper, we propose a method for extracting more detailed physical knowledge from a set of RGB-D images taken from the same scene but from different views using qualitative reasoning and intuitive physical models. Rather than providing a simple contact relation graph and approximating stability over convex shapes, our method is able to provide a detailed supporting relation analysis based on a volumetric representation. Specifically, true supporting relations between objects (e.g., if an object supports another object by touching it on the side or if the object above contributes to the stability of the object below) are identified. We apply our method to real-world structures captured in warehouse scenarios and show our method works as desired.

cs.CV↗
arXiv2019

Novel two-dimensional Ca-Cl crystals with metallicity, piezoelectric effect and room-temperature ferromagnetism

Recently we have reported the direct observation of two-dimensional (2D) Ca-Cl crystals on reduced graphene oxide (rGO) membranes, in which the calcium ions are only about monovalent (i.e. ~+1) and metallic rather than insulating properties are displayed by those CaCl crystals. Here, we report the experimental observation and demonstration of the formation of graphene-Ca-Cl heterojunction owing to the metallicity of 2D Ca-Cl crystals, unexpected piezoelectric effect, room-temperature ferromagnetism, as well as the distinct hydrogen storage and release capability of the Ca-Cl crystals in rGO membranes. Theoretical studies show that the formation of those abnormal crystals is attributed to the strong cation-pi interactions of the Ca2+ with the aromatic rings in the graphitic surfaces. Since strong cation-pi interactions also exist between other metal ions (such as Mg2+, Fe2+, Co2+, Cu2+, Cd2+, Cr2+ and Pb2+) and graphitic surfaces, similar 2D crystals with abnormal valence state of the metal cations and corresponding abnormal properties as well as novel applications are highly expected. Those findings further show the realistically potential applications of such abnormal CaCl material with unusual electronic properties in designing novel transistors and magnetic devices, hydrogen storage, catalyzer, high-performance conducting electrodes and sensors, with a size down to atomic scale.

physics.chem-ph↗
arXiv2019

Periodic Anderson model with Holstein phonons for the description of the Cerium volume collapse

Recent experiments have suggested that the electron-phonon coupling may play an important role in the $γ\rightarrow α$ volume collapse transition in Cerium. A minimal model for the description of such transition is the periodic Anderson model. In order to better understand the effect of the electron-phonon interaction on the volume collapse transition, we study the periodic Anderson model with coupling between Holstein phonons and electrons in the conduction band. We find that the electron-phonon coupling enhances the volume collapse, which is consistent with experiments in Cerium. While we start with the Kondo Volume Collapse scenario in mind, our results capture some interesting features of the Mott scenario, such as a gap in the conduction electron spectra which grows with the effective electron-phonon coupling.

cond-mat.str-el↗
arXiv2019

FeatherNets: Convolutional Neural Networks as Light as Feather for Face Anti-spoofing

Face Anti-spoofing gains increased attentions recently in both academic and industrial fields. With the emergence of various CNN based solutions, the multi-modal(RGB, depth and IR) methods based CNN showed better performance than single modal classifiers. However, there is a need for improving the performance and reducing the complexity. Therefore, an extreme light network architecture(FeatherNet A/B) is proposed with a streaming module which fixes the weakness of Global Average Pooling and uses less parameters. Our single FeatherNet trained by depth image only, provides a higher baseline with 0.00168 ACER, 0.35M parameters and 83M FLOPS. Furthermore, a novel fusion procedure with ``ensemble + cascade'' structure is presented to satisfy the performance preferred use cases. Meanwhile, the MMFD dataset is collected to provide more attacks and diversity to gain better generalization. We use the fusion method in the Face Anti-spoofing Attack Detection Challenge@CVPR2019 and got the result of 0.0013(ACER), 0.999(TPR@FPR=10e-2), 0.998(TPR@FPR=10e-3) and 0.9814(TPR@FPR=10e-4).

cs.CV↗
arXiv2019

Enhancing the long-term performance of recommender system

Recommender system is a critically important tool in online commercial system and provide users with personalized recommendation on items. So far, numerous recommendation algorithms have been made to further improve the recommendation performance in a single-step recommendation, while the long-term recommendation performance is neglected. In this paper, we proposed an approach called Adjustment of Recommendation List (ARL) to enhance the long-term recommendation accuracy. In order to observe the long-term accuracy, we developed an evolution model of network to simulate the interaction between the recommender system and user's behaviour. The result shows that not only long-term recommendation accuracy can be enhanced significantly but the diversity of item in online system maintains healthy. Notably, an optimal parameter n* of ARL existed in long-term recommendation, indicating that there is a trade-off between keeping diversity of item and user's preference to maximize the long-term recommendation accuracy. Finally, we confirmed that the optimal parameter n* is stable during evolving network, which reveals the robustness of ARL method.

physics.soc-ph↗
arXiv2019

Predictability of diffusion-based recommender systems

The recommendation methods based on network diffusion have been shown to perform well in both recommendation accuracy and diversity. Nowdays, numerous extensions have been made to further improve the performance of such methods. However, to what extent can items be predicted by diffusion-based algorithms still lack of understanding. Here, we mainly propose a method to quantify the predictability of diffusion-based algorithms. Accordingly, we conduct experiments on Movielens and Netflix data sets. The results show that the higher recommendation accuracy based on diffusion algorithms can still be achieved by optimizing the way of resource allocation on a density network. On a sparse network, the possibility of improving accuracy is relatively low due to the fact that the current accuracy of diffusion-based methods is very close its predictability. In this case, we find that the predictability can be improved significantly by multi-steps diffusion, especially for users with less historical information. In contrast to common belief, there are plausible circumstances where the higher predictability of diffusion-based methods do not correspond to those users with more historical recording. Thus, we proposed the diffusion coverage and item average degree to explain this phenomenon. In addition, we demonstrate the recommendation accuracy in real online system is overestimated by random partition used in the literature, suggesting the recommendation in real online system may be a harder task.

physics.soc-ph↗