arXiv subjects

Yang Wu

Publications and source records attributed to Yang Wu.

At least 199 records · Page 11Linked to original sources

arXiv2018

Dynamic Ensemble Active Learning: A Non-Stationary Bandit with Expert Advice

Active learning aims to reduce annotation cost by predicting which samples are useful for a human teacher to label. However it has become clear there is no best active learning algorithm. Inspired by various philosophies about what constitutes a good criteria, different algorithms perform well on different datasets. This has motivated research into ensembles of active learners that learn what constitutes a good criteria in a given scenario, typically via multi-armed bandit algorithms. Though algorithm ensembles can lead to better results, they overlook the fact that not only does algorithm efficacy vary across datasets, but also during a single active learning session. That is, the best criteria is non-stationary. This breaks existing algorithms' guarantees and hampers their performance in practice. In this paper, we propose dynamic ensemble active learning as a more general and promising research direction. We develop a dynamic ensemble active learner based on a non-stationary multi-armed bandit with expert advice algorithm. Our dynamic ensemble selects the right criteria at each step of active learning. It has theoretical guarantees, and shows encouraging results on $13$ popular datasets.

cs.LG↗

arXiv2018

A programmable two-qubit solid-state quantum processor under ambient conditions

Quantum computers, which take advantage of the superposition and entanglement of physical states, could outperform their classical counterparts in solving problems with technological impact, such as factoring large numbers and searching databases. A quantum processor executes algorithms by applying a programmable sequence of gates to an initialized state of qubits, which coherently evolves into a final state containing the result of the computation. Although quantum processors with a few qubits have been demonstrated on multiple quantum computing platforms, realization of solid-state programmable quantum processor under ambient conditions remains elusive. Here we report a programable quantum processor that can be programmed with fifteen parameters to realize arbitrary unitary transformations on two spin-qubits in a nitrogen-vacancy (NV) center in diamond. We implemented the Deutsch-Jozsa and Grover search algorithms with average success rates above 80\%. This programmable two-qubit processor could form a core component of a large-scale quantum processor, and the methods used here are suitable for such a device.

quant-ph↗

arXiv2018

Barkhausen effect in the first order structural phase transition in type-II Weyl semimetal MoTe2

We report the first observation of the non-magnetic Barkhausen effect in van der Waals layered crystals, specifically, between the Td and 1T' phases in type-II Weyl semimetal MoTe2. Thinning down the MoTe2 crystal from bulk material to about 25nm results in a drastic strengthening of the hysteresis in the phase transition, with the difference in critical temperature increasing from 40K to more than 300K. The Barkhausen effect appears for thin samples and the temperature range of the Barkhausen zone grows approximately linearly with reducing sample thickness, pointing to a surface origin of the phase pinning defects. The distribution of the Barkhausen jumps shows a power law behavior, with its critical exponent α = 1.27, in good agreement with existing scaling theory. Temperature-dependent Raman spectroscopy on MoTe2 crystals of various thicknesses shows results consistent with our transport measurements.

cond-mat.mtrl-sci↗

arXiv2018

Knowledge-Embedded Representation Learning for Fine-Grained Image Recognition

Humans can naturally understand an image in depth with the aid of rich knowledge accumulated from daily lives or professions. For example, to achieve fine-grained image recognition (e.g., categorizing hundreds of subordinate categories of birds) usually requires a comprehensive visual concept organization including category labels and part-level attributes. In this work, we investigate how to unify rich professional knowledge with deep neural network architectures and propose a Knowledge-Embedded Representation Learning (KERL) framework for handling the problem of fine-grained image recognition. Specifically, we organize the rich visual concepts in the form of knowledge graph and employ a Gated Graph Neural Network to propagate node message through the graph for generating the knowledge representation. By introducing a novel gated mechanism, our KERL framework incorporates this knowledge representation into the discriminative image feature learning, i.e., implicitly associating the specific attributes with the feature maps. Compared with existing methods of fine-grained image classification, our KERL framework has several appealing properties: i) The embedded high-level knowledge enhances the feature representation, thus facilitating distinguishing the subtle differences among subordinate categories. ii) Our framework can learn feature maps with a meaningful configuration that the highlighted regions finely accord with the nodes (specific attributes) of the knowledge graph. Extensive experiments on the widely used Caltech-UCSD bird dataset demonstrate the superiority of our KERL framework over existing state-of-the-art methods.

cs.CV↗

arXiv2018

Meta-Learning Transferable Active Learning Policies by Deep Reinforcement Learning

Active learning (AL) aims to enable training high performance classifiers with low annotation cost by predicting which subset of unlabelled instances would be most beneficial to label. The importance of AL has motivated extensive research, proposing a wide variety of manually designed AL algorithms with diverse theoretical and intuitive motivations. In contrast to this body of research, we propose to treat active learning algorithm design as a meta-learning problem and learn the best criterion from data. We model an active learning algorithm as a deep neural network that inputs the base learner state and the unlabelled point set and predicts the best point to annotate next. Training this active query policy network with reinforcement learning, produces the best non-myopic policy for a given dataset. The key challenge in achieving a general solution to AL then becomes that of learner generalisation, particularly across heterogeneous datasets. We propose a multi-task dataset-embedding approach that allows dataset-agnostic active learners to be trained. Our evaluation shows that AL algorithms trained in this way can directly generalise across diverse problems.

cs.LG↗

arXiv2018

Terahertz Emission from Compensated Magnetic Heterostructures

Terahertz emission spectroscopy (TES) has recently played an important role in unveiling the spin dynamics at a terahertz (THz) frequency range. So far, ferromagnetic (FM)/nonmagnetic (NM) heterostructures have been intensively studied as THz sources. Compensated magnets such as a ferrimagnet (FIM) and antiferromagnet (AFM) are other types of magnetic materials with interesting spin dynamics. In this work, we study TES from compensated magnetic heterostructures including CoGd FIM alloy or IrMn AFM layers. Systematic measurements on composition and temperature dependences of THz emission from CoGd/Pt bilayer structures are conducted. It is found that the emitted THz field is determined by the net spin polarization of the laser induced spin current rather than the net magnetization. The temperature robustness of the FIM based THz emitter is also demonstrated. On the other hand, an AFM plays a different role in THz emission. The IrMn/Pt bilayer shows negligible THz signals, whereas Co/IrMn induces sizable THz outputs, indicating that IrMn is not a good spin current generator, but a good detector. Our results not only suggest that a compensated magnet can be utilized for robust THz emission, but also provide a new approach to study the magnetization dynamics especially near the magnetization compensation point.

cond-mat.mes-hall↗

arXiv2018

Direct visualization of current induced spin accumulation in topological insulators

Charge-to-spin conversion in various material systems is the key for the fundamental understanding of spin-orbitronis as well as the development of efficient means to manipulate the magnetization. We report the direct spatial imaging of current induced spin accumulation at the channel edges of Bi2Se3 and BiSbTeSe2 topological insulators by a scanning photovoltage microscope at room temperature. The spin polarization is along the out-of-plane direction with opposite signs for the two channel edges. The accumulated spin direction reverses sign upon changing the current direction and the detected spin signal shows a linear dependence on the magnitude of currents, which indicates that our observed phenomena are the current induced effects. The spin Hall angle of Bi2Se3 and BiSbTeSe2 is determined to be 0.0085 and 0.0616, respectively. We further image the current induced spin accumulation in a Pt heavy metal. Our results open up the possibility of optically detecting the current induced spin accumulations, and thus point towards a better understanding of the interaction between spins and circularly polarized light.

cond-mat.mes-hall↗

arXiv2018

Widely tunable band gap in a multivalley semiconductor SnSe by potassium doping

SnSe, a group IV-VI monochalcogenide with layered crystal structure similar to black phosphorus, has recently attracted extensive interests due to its excellent thermoelectric properties and potential device applications. Experimental electronic structure of both the valence and conduction bands is critical for understanding the effects of hole versus electron doping on the thermoelectric properties, and to further reveal possible change of the band gap upon doping. Here, we report the multivalley valence bands with a large effective mass on semiconducting SnSe crystals and reveal single-valley conduction bands through electron doping to provide a complete picture of the thermoelectric physics. Moreover, by electron doping through potassium deposition, the band gap of SnSe can be widely tuned from 1.2 eV to 0.4 eV, providing new opportunities for tunable electronic and optoelectronic devices.

cond-mat.mtrl-sci↗

arXiv2018

Pose-Normalized Image Generation for Person Re-identification

Person Re-identification (re-id) faces two major challenges: the lack of cross-view paired training data and learning discriminative identity-sensitive and view-invariant features in the presence of large pose variations. In this work, we address both problems by proposing a novel deep person image generation model for synthesizing realistic person images conditional on the pose. The model is based on a generative adversarial network (GAN) designed specifically for pose normalization in re-id, thus termed pose-normalization GAN (PN-GAN). With the synthesized images, we can learn a new type of deep re-id feature free of the influence of pose variations. We show that this feature is strong on its own and complementary to features learned with the original images. Importantly, under the transfer learning setting, we show that our model generalizes well to any new re-id dataset without the need for collecting any training data for model fine-tuning. The model thus has the potential to make re-id model truly scalable.

cs.CV↗

arXiv2018

Depth-Based 3D Hand Pose Estimation: From Current Achievements to Future Goals

In this paper, we strive to answer two questions: What is the current state of 3D hand pose estimation from depth images? And, what are the next challenges that need to be tackled? Following the successful Hands In the Million Challenge (HIM2017), we investigate the top 10 state-of-the-art methods on three tasks: single frame 3D pose estimation, 3D hand tracking, and hand pose estimation during object interaction. We analyze the performance of different CNN structures with regard to hand shape, joint visibility, view point and articulation distributions. Our findings include: (1) isolated 3D hand pose estimation achieves low mean errors (10 mm) in the view point range of [70, 120] degrees, but it is far from being solved for extreme view points; (2) 3D volumetric representations outperform 2D CNNs, better capturing the spatial structure of the depth data; (3) Discriminative methods still generalize poorly to unseen hand shapes; (4) While joint occlusions pose a challenge for most methods, explicit modeling of structure constraints can significantly narrow the gap between errors on visible and occluded joints.

cs.CV↗

arXiv2018

Room-temperature nanoseconds spin relaxation in WTe2 and MoTe2 thin films

The Weyl semimetal WTe2 and MoTe2 show great potential in generating large spin currents since they possess topologically-protected spin-polarized states and can carry a very large current density. In addition, the intrinsic noncentrosymmetry of WTe2 and MoTe2 endows with a unique property of crystal symmetry-controlled spin-orbit torques. An important question to be answered for developing spintronic devices is how spins relax in WTe2 and MoTe2. Here, we report a room-temperature spin relaxation time of 1.2 ns (0.4 ns) in WTe2 (MoTe2) thin film using the time-resolved Kerr rotation (TRKR). Based on ab initio calculation, we identify a mechanism of long-lived spin polarization resulting from a large spin splitting around the bottom of the conduction band, low electron-hole recombination rate and suppression of backscattering required by time-reversal and lattice symmetry operation. In addition, we find the spin polarization is firmly pinned along the strong internal out-of-plane magnetic field induced by large spin splitting. Our work provides an insight into the physical origin of long-lived spin polarization in Weyl semimetals which could be useful to manipulate spins for a long time at room temperature.

cond-mat.mtrl-sci↗

arXiv2018

Metric Learning via Maximizing the Lipschitz Margin Ratio

In this paper, we propose the Lipschitz margin ratio and a new metric learning framework for classification through maximizing the ratio. This framework enables the integration of both the inter-class margin and the intra-class dispersion, as well as the enhancement of the generalization ability of a classifier. To introduce the Lipschitz margin ratio and its associated learning bound, we elaborate the relationship between metric learning and Lipschitz functions, as well as the representability and learnability of the Lipschitz functions. After proposing the new metric learning framework based on the introduced Lipschitz margin ratio, we also prove that some well known metric learning algorithms can be shown as special cases of the proposed framework. In addition, we illustrate the framework by implementing it for learning the squared Mahalanobis metric, and by demonstrating its encouraging results on eight popular datasets of machine learning.

cs.LG↗

arXiv2017

Ultrafast Electronic Dynamics of a Weyl Semimetal MoTe$_2$ Revealed by Time and Angle Resolved Photoemission Spectroscopy

A Weyl semimetal is a new type of topological quantum phase with intriguing physics near the Weyl nodes. Although the equilibrium state of Weyl semimetals has been investigated, the ultrafast dynamics near the Weyl node in the nonequilibrium state is still missing. Here by performing time and angle resolved photoemission spectroscopy on type-II Weyl semimetal MoTe$_2$, we reveal the dispersion of the unoccupied states and identify the Weyl node at 70 meV above E$_F$. Moreover, by tracking the ultrafast relaxation dynamics near the Weyl node upon photo-excitation with energy, momentum and temporal resolution, two intrinsic recovery timescales are observed, a fast one of 430 fs and a slow one of 4.1 ps, which are associated with hot electron cooling by optical phonon cascade emission and anharmonic decay of hot optical phonons respectively. The electron population shows a metallic response, and the two temperature model fitting of the transient electronic temperature gives an electron-phonon coupling constant of $λ\langleΩ^2\rangle\simeq32$ $\textrm{meV}^2$. Our work provides important dynamic information for understanding the relaxation mechanism of a Weyl semimetal and for exploiting potential applications using ultrafast optical control.

cond-mat.mtrl-sci↗

arXiv2017

Joint Hierarchical Category Structure Learning and Large-Scale Image Classification

We investigate the scalable image classification problem with a large number of categories. Hierarchical visual data structures are helpful for improving the efficiency and performance of large-scale multi-class classification. We propose a novel image classification method based on learning hierarchical inter-class structures. Specifically, we first design a fast algorithm to compute the similarity metric between categories, based on which a visual tree is constructed by hierarchical spectral clustering. Using the learned visual tree, a test sample label is efficiently predicted by searching for the best path over the entire tree. The proposed method is extensively evaluated on the ILSVRC2010 and Caltech 256 benchmark datasets. Experimental results show that our method obtains significantly better category hierarchies than other state-of-the-art visual tree-based methods and, therefore, much more accurate classification.

cs.CV↗

arXiv2017

Room temperature magnetization switching in topological insulator-ferromagnet heterostructures by spin-orbit torques

Topological insulators (TIs) with spin momentum locked topological surface states (TSS) are expected to exhibit a giant spin-orbit torque (SOT) in the TI/ferromagnet systems. To date, the TI SOT driven magnetization switching is solely reported in a Cr doped TI at 1.9 K. Here, we directly show giant SOT driven magnetization switching in a Bi2Se3/NiFe heterostructure at room temperature captured using a magneto-optic Kerr effect microscope. We identify a large charge to spin conversion efficiency of ~1-1.75 in the thin TI films, where the TSS is dominant. In addition, we find the current density required for the magnetization switching is extremely low, ~6x10^5 A cm-2, which is one to two orders of magnitude smaller than that with heavy metals. Our demonstration of room temperature magnetization switching of a conventional 3d ferromagnet using Bi2Se3 may lead to potential innovations in TI based spintronic applications.

cond-mat.mtrl-sci↗

arXiv2017

Experimental protection of the coherence of a molecular qubit exceeding a millisecond

There are several important solid-state systems, such as defects in solids, superconducting circuits and molecular qubits, for attractive candidates of quantum computations. Molecular qubits, which benefit from the power of chemistry for the tailored and inexpensive synthesis of new systems, face the challenge from decoherence effect. The decoherence effect is due to the molecular qubits' inevitable interactions to their environment. Strategies to overcome decoherence effect have been developed, such as designing a nuclear spin free environment and working at atomic clock transitions. These chemical approaches, however, have some fundamental limitations. For example, chemical approach restricts the ligand selection and design to ligands with zero nuclear magnetic dipole moment, such as carbon, oxygen, and sulfur. Herein, a physical approach, named Dynamical decoupling (DD), is utilized to combat decoherence, while the limitations of the chemical approaches can be avoided. The phase memory time $T_2$ for a transition metal complex has been prolonged to exceed one millisecond ($1.4~$ms) by employing DD. The single qubit figure of merit $Q_M $ reaches $ 1.4\times 10^5$, which is $40$ times better than that previously reported value for such system. Our results show that molecular qubits, with milliseconds $T_2$, are promising candidates for quantum information processing.

quant-ph↗

arXiv2017

High quality atomically thin PtSe2 films grown by molecular beam epitaxy

Atomically thin PtSe2 films have attracted extensive research interests for potential applications in high-speed electronics, spintronics and photodetectors. Obtaining high quality, single crystalline thin films with large size is critical. Here we report the first successful layer-by-layer growth of high quality PtSe2 films by molecular beam epitaxy. Atomically thin films from 1 ML to 22 ML have been grown and characterized by low-energy electron diffraction, Raman spectroscopy and X-ray photoemission spectroscopy. Moreover, a systematic thickness dependent study of the electronic structure is revealed by angle-resolved photoemission spectroscopy (ARPES), and helical spin texture is revealed by spin-ARPES. Our work provides new opportunities for growing large size single crystalline films for investigating the physical properties and potential applications of PtSe2.

cond-mat.mtrl-sci↗

arXiv2017

Experimental evidence of type-II Dirac fermions in PtSe2

Topological semimetals have attracted extensive research interests for realizing condensed matter physics counterparts of three-dimensional Dirac and Weyl fermions, which were originally introduced in high energy physics. Recently it has been proposed that type-II Dirac semimetal can host a new type of Dirac fermions which break Lorentz invariance and therefore does not have counterpart in high energy physics. Here we report the electronic structure of high quality PtSe$_2$ crystals to provide direct evidence for the existence of three-dimensional type-II Dirac fermions. A comparison of the crystal, vibrational and electronic structure to a sister compound PtTe$_2$ is also discussed. Our work provides an important platform for exploring the novel quantum phenomena in the PtSe$_2$ class of type-II Dirac semimetals.

cond-mat.mtrl-sci↗