arXiv subjects

Di Wang

Publications and source records attributed to Di Wang.

At least 595 records · Page 33Linked to original sources

arXiv2019

Single-Path NAS: Device-Aware Efficient ConvNet Design

Can we automatically design a Convolutional Network (ConvNet) with the highest image classification accuracy under the latency constraint of a mobile device? Neural Architecture Search (NAS) for ConvNet design is a challenging problem due to the combinatorially large design space and search time (at least 200 GPU-hours). To alleviate this complexity, we propose Single-Path NAS, a novel differentiable NAS method for designing device-efficient ConvNets in less than 4 hours. 1. Novel NAS formulation: our method introduces a single-path, over-parameterized ConvNet to encode all architectural decisions with shared convolutional kernel parameters. 2. NAS efficiency: Our method decreases the NAS search cost down to 8 epochs (30 TPU-hours), i.e., up to 5,000x faster compared to prior work. 3. On-device image classification: Single-Path NAS achieves 74.96% top-1 accuracy on ImageNet with 79ms inference latency on a Pixel 1 phone, which is state-of-the-art accuracy compared to NAS methods with similar latency (<80ms).

cs.LG↗

arXiv2019

Density Matching for Bilingual Word Embedding

Recent approaches to cross-lingual word embedding have generally been based on linear transformations between the sets of embedding vectors in the two languages. In this paper, we propose an approach that instead expresses the two monolingual embedding spaces as probability densities defined by a Gaussian mixture model, and matches the two densities using a method called normalizing flow. The method requires no explicit supervision, and can be learned with only a seed dictionary of words that have identical strings. We argue that this formulation has several intuitively attractive properties, particularly with the respect to improving robustness and generalization to mappings between difficult language pairs or word pairs. On a benchmark data set of bilingual lexicon induction and cross-lingual word similarity, our approach can achieve competitive or superior performance compared to state-of-the-art published results, with particularly strong results being found on etymologically distant and/or morphologically rich languages.

cs.CL↗

arXiv2019

Epitaxial strain adaption in chemically disordered FeRh thin films

Strain and strain adaption mechanisms in modern functional materials are of crucial importance for their performance. Understanding these mechanisms will advance innovative approaches for material properties engineering. Here we study the strain adaption mechanism in a thin film model system as function of epitaxial strain. Chemically disordered FeRh thin films are deposited on W-V buffer layers, which allow for large variation of the preset lattice constants, e.g. epitaxial boundary condition. It is shown by means of high resolution X-ray reciprocal space maps and transmission electron microscopy that the system reacts with a tilting mechanism of the structural units in order to adapt to the lattice constants of the buffer layer. This response explained by density functional theory calculations, which evidence an energetic minimum for structures with a distortion of c/a =0.87. The experimentally observed tilting mechanism is induced by this energy gain and allows the system to remain in the most favorable structure. In general, it is shown that the use of epitaxial model heterostructures consisting of alloy buffer layers of fully miscible elements and the functional material of interest allows to study strain adaption behaviors in great detail. This approach makes even small secondary effects observable, such as the directional tilting of the structural domains identified in the present case study.

cond-mat.mtrl-sci↗

arXiv2019

Resource-Guided Program Synthesis

This article presents resource-guided synthesis, a technique for synthesizing recursive programs that satisfy both a functional specification and a symbolic resource bound. The technique is type-directed and rests upon a novel type system that combines polymorphic refinement types with potential annotations of automatic amortized resource analysis. The type system enables efficient constraint-based type checking and can express precise refinement-based resource bounds. The proof of type soundness shows that synthesized programs are correct by construction. By tightly integrating program exploration and type checking, the synthesizer can leverage the user-provided resource bound to guide the search, eagerly rejecting incomplete programs that consume too many resources. An implementation in the resource-guided synthesizer ReSyn is used to evaluate the technique on a range of recursive data structure manipulations. The experiments show that ReSyn synthesizes programs that are asymptotically more efficient than those generated by a resource-agnostic synthesizer. Moreover, synthesis with ReSyn is faster than a naive combination of synthesis and resource analysis. ReSyn is also able to generate implementations that have a constant resource consumption for fixed input sizes, which can be used to mitigate side-channel attacks.

cs.PL↗

arXiv2019

Differentially Private High Dimensional Sparse Covariance Matrix Estimation

In this paper, we study the problem of estimating the covariance matrix under differential privacy, where the underlying covariance matrix is assumed to be sparse and of high dimensions. We propose a new method, called DP-Thresholding, to achieve a non-trivial $\ell_2$-norm based error bound, which is significantly better than the existing ones from adding noise directly to the empirical covariance matrix. We also extend the $\ell_2$-norm based error bound to a general $\ell_w$-norm based one for any $1\leq w\leq \infty$, and show that they share the same upper bound asymptotically. Our approach can be easily extended to local differential privacy. Experiments on the synthetic datasets show consistent results with our theoretical claims.

cs.LG↗

arXiv2019

Single-Path NAS: Designing Hardware-Efficient ConvNets in less than 4 Hours

Can we automatically design a Convolutional Network (ConvNet) with the highest image classification accuracy under the runtime constraint of a mobile device? Neural architecture search (NAS) has revolutionized the design of hardware-efficient ConvNets by automating this process. However, the NAS problem remains challenging due to the combinatorially large design space, causing a significant searching time (at least 200 GPU-hours). To alleviate this complexity, we propose Single-Path NAS, a novel differentiable NAS method for designing hardware-efficient ConvNets in less than 4 hours. Our contributions are as follows: 1. Single-path search space: Compared to previous differentiable NAS methods, Single-Path NAS uses one single-path over-parameterized ConvNet to encode all architectural decisions with shared convolutional kernel parameters, hence drastically decreasing the number of trainable parameters and the search cost down to few epochs. 2. Hardware-efficient ImageNet classification: Single-Path NAS achieves 74.96% top-1 accuracy on ImageNet with 79ms latency on a Pixel 1 phone, which is state-of-the-art accuracy compared to NAS methods with similar constraints (<80ms). 3. NAS efficiency: Single-Path NAS search cost is only 8 epochs (30 TPU-hours), which is up to 5,000x faster compared to prior work. 4. Reproducibility: Unlike all recent mobile-efficient NAS methods which only release pretrained models, we open-source our entire codebase at: https://github.com/dstamoulis/single-path-nas.

cs.LG↗

arXiv2018

An Affect-Rich Neural Conversational Model with Biased Attention and Weighted Cross-Entropy Loss

Affect conveys important implicit information in human communication. Having the capability to correctly express affect during human-machine conversations is one of the major milestones in artificial intelligence. In recent years, extensive research on open-domain neural conversational models has been conducted. However, embedding affect into such models is still under explored. In this paper, we propose an end-to-end affect-rich open-domain neural conversational model that produces responses not only appropriate in syntax and semantics, but also with rich affect. Our model extends the Seq2Seq model and adopts VAD (Valence, Arousal and Dominance) affective notations to embed each word with affects. In addition, our model considers the effect of negators and intensifiers via a novel affective attention mechanism, which biases attention towards affect-rich words in input sentences. Lastly, we train our model with an affect-incorporated objective function to encourage the generation of affect-rich words in the output responses. Evaluations based on both perplexity and human evaluations show that our model outperforms the state-of-the-art baseline model of comparable size in producing natural and affect-rich responses.

cs.CL↗

arXiv2018

Liquid Epitaxial Growth of Two-dimensional Non-layer structured hybrid Perovskite

The hybrid perovskite CH3NH3PbX3 (X= Cl, Br, I) is a promising material for developing novel optoelectronic devices. Owing to the intrinsic non-layer structure, it remains challenging to synthesize molecularly thin CH3NH3PbX3 with large size. Here, we report a low-cost and highly efficient fabrication route to obtain large-scale single-crystalline 2D CH3NH3PbX3 perovskites on a mica substrate via liquid epitaxy. The 2D perovskite is characterized as 8 nm in thickness and hundreds of micrometers in lateral size. First-principles calculations suggest the strong potassium-halogen interactions at the perovskite/mica interface lower the interface energy of perovskites, driving their fast in-plane growth. Spectroscopic investigations reveal 2D CH3NH3PbBr3 possess small exciton binding energy of 30 meV, allowing a superior visible-light photodetector with a photoresponsivity of 126 A/W and a bandwidth exceeded 80 kHz. These features demonstrate that liquid epitaxy is a bottom-up approach to fabricate the non-layer structured 2D perovskites, which offer a new material platform for the device applications and fundamental investigations.

cond-mat.mtrl-sci↗

arXiv2018

Calculated magnetic exchange interactions in Dirac magnon material Cu3TeO6

Recently topological aspects of magnon band structure have attracted much interest, and especially, the Dirac magnons in Cu3TeO6 have been observed experimentally. In this work, we calculate the magnetic exchange interactions J's using the first-principles linear-response approach and find that these J's are short-range and negligible for the Cu-Cu atomic pair apart by longer than 7 Angstrom. Moreover there are only 5 sizable magnetic exchange interactions, and according to their signs and strengths, modest magnetic frustration is expected. Based on the obtained magnetic exchange couplings, we successfully reproduce the experimental spin-wave dispersions. The calculated neutron scattering cross section also agrees very well with the experiments. We also calculate Dzyaloshinskii-Moriya interactions (DMIs) and estimate the canting angle (about 1.3°) of the magnetic non-collinearity based on the competition between DMIs and J's, which is consistent with the experiment. The small canting angle agrees with that the current experiments cannot distinguish the DMI induced nodal line from a Dirac point in the spin-wave spectrum. Finally we analytically prove that the "sum rule" conjectured in [Nat. Phys. 14, 1011 (2018)] holds but only up to the 11th nearest neighbour.

cond-mat.mtrl-sci↗

arXiv2018

Discovery of coexisting Dirac and triply degenerate magnons in a three-dimensional antiferromagnet

Topological magnons are emergent quantum spin excitations featured by magnon bands crossing linearly at the points dubbed nodes, analogous to fermions in topological electronic systems. Experimental realization of topological magnons in three dimensions has not been reported so far. Here, by measuring spin excitations (magnons) of a three-dimensional antiferromagnet Cu$_{3}$TeO$_{6}$ with inelastic neutron scattering, we provide direct spectroscopic evidence for the coexistence of symmetry-protected Dirac and triply degenerate nodes, the latter involving three-component magnons beyond the Dirac-Weyl framework. Our theoretical calculations show that the observed topological magnon band structure can be well described by the linear-spin-wave theory based on a Hamiltonian dominated by the nearest-neighbour exchange interaction $J_1$. As such, we showcase Cu$_{3}$TeO$_{6}$ as an example system where Dirac and triply degenerate magnonic nodal excitations coexist, demonstrate an exotic topological state of matter, and provide a fresh ground to explore the topological properties in quantum materials.

cond-mat.str-el↗

arXiv2018

Neural Machine Translation with Key-Value Memory-Augmented Attention

Although attention-based Neural Machine Translation (NMT) has achieved remarkable progress in recent years, it still suffers from issues of repeating and dropping translations. To alleviate these issues, we propose a novel key-value memory-augmented attention model for NMT, called KVMEMATT. Specifically, we maintain a timely updated keymemory to keep track of attention history and a fixed value-memory to store the representation of source sentence throughout the whole translation process. Via nontrivial transformations and iterative interactions between the two memories, the decoder focuses on more appropriate source word(s) for predicting the next target word at each decoding step, therefore can improve the adequacy of translations. Experimental results on Chinese=>English and WMT17 German<=>English translation tasks demonstrate the superiority of the proposed model.

cs.CL↗

arXiv2018

Techniques for Efficiently Handling Power Surges in Fuel Cell Powered Data Centers: Modeling, Analysis, Results

Fuel cells are a promising power source for future data centers, offering high energy efficiency, low greenhouse gas emissions, and high reliability. However, due to mechanical limitations related to fuel delivery, fuel cells are slow to adjust to sudden increases in data center power demands, which can result in temporary power shortfalls. To mitigate the impact of power shortfalls, prior work has proposed to either perform power capping by throttling the servers, or to leverage energy storage devices (ESDs) that can temporarily provide enough power to make up for the shortfall while the fuel cells ramp up power generation. Both approaches have disadvantages: power capping conservatively limits server performance and can lead to service level agreement (SLA) violations, while ESD-only solutions must significantly overprovision the energy storage device capacity to tolerate the shortfalls caused by the worst-case (i.e., largest) power surges, which greatly increases the total cost of ownership (TCO). We propose SizeCap, the first ESD sizing framework for fuel cell powered data centers, which coordinates ESD sizing with power capping to enable a cost-effective solution to power shortfalls in data centers. SizeCap sizes the ESD just large enough to cover the majority of power surges, but not the worst-case surges that occur infrequently, to greatly reduce TCO. It then uses the smaller capacity ESD in conjunction with power capping to cover the power shortfalls caused by the worst-case power surges. As part of our new flexible framework, we propose multiple power capping policies with different degrees of awareness of fuel cell and workload behavior, and evaluate their impact on workload performance and ESD size. Using traces from Microsoft's production data center systems, we demonstrate that SizeCap significantly reduces the ESD size without violating any SLAs.

cs.DC↗

arXiv2018

Capacity Releasing Diffusion for Speed and Locality

Diffusions and related random walk procedures are of central importance in many areas of machine learning, data analysis, and applied mathematics. Because they spread mass agnostically at each step in an iterative manner, they can sometimes spread mass "too aggressively," thereby failing to find the "right" clusters. We introduce a novel Capacity Releasing Diffusion (CRD) Process, which is both faster and stays more local than the classical spectral diffusion process. As an application, we use our CRD Process to develop an improved local algorithm for graph clustering. Our local graph clustering method can find local clusters in a model of clustering where one begins the CRD Process in a cluster whose vertices are connected better internally than externally by an $O(\log^2 n)$ factor, where $n$ is the number of nodes in the cluster. Thus, our CRD Process is the first local graph clustering algorithm that is not subject to the well-known quadratic Cheeger barrier. Our result requires a certain smoothness condition, which we expect to be an artifact of our analysis. Our empirical evaluation demonstrates improved results, in particular for realistic social graphs where there are moderately good---but not very good---clusters.

cs.DS↗

arXiv2018

Enlarged exceptional symmetries of first-quantized F-theory

The exceptional symmetries of supergravity have been reproduced from the Hamiltonian formulation of the classical mechanics of F-theory. We now find the Lagrangian formalism has even larger exceptional symmetries, simplifying its derivation: We discuss D = 5 as an example.

hep-th↗

arXiv2018

Empirical Risk Minimization in Non-interactive Local Differential Privacy: Efficiency and High Dimensional Case

In this paper, we study the Empirical Risk Minimization problem in the non-interactive local model of differential privacy. In the case of constant or low dimensionality ($p\ll n$), we first show that if the ERM loss function is $(\infty, T)$-smooth, then we can avoid a dependence of the sample complexity, to achieve error $α$, on the exponential of the dimensionality $p$ with base $1/α$ (i.e., $α^{-p}$), which answers a question in [smith 2017 interaction]. Our approach is based on polynomial approximation. Then, we propose player-efficient algorithms with $1$-bit communication complexity and $O(1)$ computation cost for each player. The error bound is asymptotically the same as the original one. Also with additional assumptions we show a server efficient algorithm. Next we consider the high dimensional case ($n\ll p$), we show that if the loss function is Generalized Linear function and convex, then we could get an error bound which is dependent on the Gaussian width of the underlying constrained set instead of $p$, which is lower than that in [smith 2017 interaction].

cs.LG↗

arXiv2018

Accelerating Light with Metasurfaces

It has been recently shown that especially engineered light beams have the remarkable ability to propagate along curved trajectories in vacuum. Current methods for generating accelerating beams use phase modulators and lenses leading to length scales on the order of tens of centimeters or larger. This poses constraints and severely limits their applicability inside materials. Here, we accelerate light inside glass using a metasurface consisting of plasmonic nanoantennas. Highly-bending beams with radii of curvature on the order of a hundred microns were generated, and the imaged intensities agree well with theory. Our approach for generating accelerating beams allows for their integration into on-chip photonic systems.

physics.optics↗

arXiv2018

On the origin of incoherent magnetic exchange coupling in MnBi/Fe$_x$Co$_{1-x}$ bilayer system

In this study we investigate the exchange coupling between the hard magnetic compound MnBi and the soft magnetic alloy FeCo including the interface structure between the two phases. Exchange spring MnBi-Fe$_x$Co$_{1-x}$ (x = 0.65 and 0.35) bilayers with various thicknesses of the soft magnetic layer were deposited onto quartz glass substrates in a DC magnetron sputtering unit from alloy targets. Magnetic measurements and density functional theory (DFT) calculations reveal that a Co-rich FeCo layer leads to more coherent exchange coupling. The optimum soft layer thickness is about 1 nm. In order to take into account the effect of incoherent interfaces with finite roughness, we have combined a cross-sectional High Resolution Transmission Electron Microscopy (HR-TEM) analysis with DFT calculations and micromagnetic simulations. The experimental results can be consistently described by modeling assuming a polycrystalline FeCo layer consisting of crystalline (110) and amorphous grains as confirmed by HR-TEM. The micromagnetic simulations show in general how the thickness of the FeCo layer and the interface roughness between the hard and soft magnetic phases both control the effectiveness of exchange coupling in an exchange spring system.

cond-mat.mtrl-sci↗

arXiv2018

$K_S^0-K_L^0$ Asymmetries and $CP$ Violation in Charmed Baryon Decays into Neutral Kaons

We study the $K^0_S-K^0_L$ asymmetries and $CP$ violations in charm-baryon decays with neutral kaons in the final state. The $K^0_S-K^0_L$ asymmetry can be used to search for two-body doubly Cabibbo-suppressed amplitudes of charm-baryon decays, with the one in $Λ^+_c\to pK^0_{S,L}$ as a promising observable. Besides, it is studied for a new $CP$-violation effect in these processes, induced by the interference between the Cabibbo-favored and doubly Cabibbo-suppressed amplitudes with the neutral kaon mixing. Once the new CP-violation effect is determined by experiments, the direct $CP$ asymmetry in neutral kaon modes can then be extracted and used to search for new physics. The numerical results based on $SU(3)$ symmetry will be tested by the experiments in the future.

hep-ph↗