arXiv subjects

Bowen Li

Publications and source records attributed to Bowen Li.

At least 235 records · Page 13Linked to original sources

arXiv2020

Heads-up! Unsupervised Constituency Parsing via Self-Attention Heads

Transformer-based pre-trained language models (PLMs) have dramatically improved the state of the art in NLP across many tasks. This has led to substantial interest in analyzing the syntactic knowledge PLMs learn. Previous approaches to this question have been limited, mostly using test suites or probes. Here, we propose a novel fully unsupervised parsing approach that extracts constituency trees from PLM attention heads. We rank transformer attention heads based on their inherent properties, and create an ensemble of high-ranking heads to produce the final tree. Our method is adaptable to low-resource languages, as it does not rely on development sets, which can be expensive to annotate. Our experiments show that the proposed method often outperform existing approaches if there is no development set present. Our unsupervised parser can also be used as a tool to analyze the grammars PLMs learn implicitly. For this, we use the parse trees induced by our method to train a neural PCFG and compare it to a grammar derived from a human-annotated treebank.

cs.CL↗

arXiv2020

Reliable Liver Fibrosis Assessment from Ultrasound using Global Hetero-Image Fusion and View-Specific Parameterization

Ultrasound (US) is a critical modality for diagnosing liver fibrosis. Unfortunately, assessment is very subjective, motivating automated approaches. We introduce a principled deep convolutional neural network (CNN) workflow that incorporates several innovations. First, to avoid overfitting on non-relevant image features, we force the network to focus on a clinical region of interest (ROI), encompassing the liver parenchyma and upper border. Second, we introduce global heteroimage fusion (GHIF), which allows the CNN to fuse features from any arbitrary number of images in a study, increasing its versatility and flexibility. Finally, we use 'style'-based view-specific parameterization (VSP) to tailor the CNN processing for different viewpoints of the liver, while keeping the majority of parameters the same across views. Experiments on a dataset of 610 patient studies (6979 images) demonstrate that our pipeline can contribute roughly 7% and 22% improvements in partial area under the curve and recall at 90% precision, respectively, over conventional classifiers, validating our approach to this crucial problem.

eess.IV↗

arXiv2020

Knowledge Graph Extraction from Videos

Nearly all existing techniques for automated video annotation (or captioning) describe videos using natural language sentences. However, this has several shortcomings: (i) it is very hard to then further use the generated natural language annotations in automated data processing, (ii) generating natural language annotations requires to solve the hard subtask of generating semantically precise and syntactically correct natural language sentences, which is actually unrelated to the task of video annotation, (iii) it is difficult to quantitatively measure performance, as standard metrics (e.g., accuracy and F1-score) are inapplicable, and (iv) annotations are language-specific. In this paper, we propose the new task of knowledge graph extraction from videos, i.e., producing a description in the form of a knowledge graph of the contents of a given video. Since no datasets exist for this task, we also include a method to automatically generate them, starting from datasets where videos are annotated with natural language. We then describe an initial deep-learning model for knowledge graph extraction from videos, and report results on MSVD* and MSR-VTT*, two datasets obtained from MSVD and MSR-VTT using our method.

cs.CL↗

arXiv2020

Atomic origin for hydrogenation promoted bulk oxygen vacancies removal in vanadium dioxide

Oxygen vacancies (VO), a common type of point defects in metal oxides materials, play important roles on the physical and chemical properties. To obtain stoichiometric oxide crystal, the pre-existing VO is always removed via careful post-annealing treatment at high temperature in air or oxygen atmosphere. However, the annealing conditions is difficult to control and the removal of VO in bulk phase is restrained due to high energy barrier of VO migration. Here, we selected VO2 crystal film as the model system and developed an alternative annealing treatment aided by controllable hydrogen doping, which can realizes effective removal of VO defects in VO2-δ crystal at lower temperature. This finding is attributed to the hydrogenation accelerated oxygen vacancies recovery in VO2-δ crystal. Theoretical calculations revealed that the H-doping induced electrons are prone to accumulate around the oxygen defects in VO2-δ film, which facilitates the diffusion of VO and thus makes it easier to be removed. The methodology is expected to be applied to other metal oxides for oxygen-related point defects control.

cond-mat.mtrl-sci↗

arXiv2020

An improved transformation between Fibonacci FSRs and Galois FSRs

Feedback shift registers (FSRs), which have two configurations: Fibonacci and Galois, are a primitive building block in stream ciphers. In this paper, an improved transformation is proposed between Fibonacci FSRs and Galois FSRs. In the previous results, the number of stages is identical when constructing the equivalent FSRs. In this paper, there is no requirement to keep the number of stages equal for two equivalent FSRs here. More precisely, it is verified that an equivalent Galois FSR with fewer stages cannot be found for a Fibonacci FSR, but the converse is not true. Furthermore, the total number of equivalent Galois FSRs for a given Fibonacci FSR with n stages is calculated. In order to reduce the propagation time and memory, an effective algorithm is developed to find equivalent Galois FSR and is proved to own minimal operators and stages. Finally, the feasibility of our proposed strategies, to mutually transform Fibonacci FSRs and Galois FSRs, is demonstrated by numerical examples.

cs.IT↗

arXiv2020

Electron Correlation driven Metal-Insulator transition in Strained and Disordered VO2 films

The Metal-Insulator transition (MIT) in VO2 is characterized by the complex interplay among lattice, electronic and orbital degrees of freedom. In this contribution we investigated the strain-modulation of the orbital hierarchy and the influence over macroscopic properties of the metallic phase of VO2 such as Fermi Level (FL) population and metallicity, i.e., the material ability to screen an electric field, by means of temperature-dependent X-ray Absorption Near Edge Structure (XANES) and Resonant Photoemission spectroscopy (ResPES). We demonstrate that the MIT in strained VO2 is of the Filling Control type, hence it is generated by electron correlation effects. In addition, we show that the MIT in Nanostructured (NS) disordered VO2, where the structural phase transition is quenched, is driven by electron correlation. Therefore a fine tuning of the correlation could lead to a precise control and tuning of the transition features.

cond-mat.mtrl-sci↗

arXiv2020

Spectral self-adaptive absorber/emitter for harvesting energy from the sun and outer space

The sun (~6000 K) and outer space (~3 K) are the original heat source and sink for human beings on Earth. The energy applications of absorbing solar irradiation and harvesting the coldness of outer space for energy utilization have attracted considerable interest from researchers. However, combining these two functions in a static device for continuous energy harvesting is unachievable due to the intrinsic infrared spectral conflict. In this study, we developed spectral self-adaptive absorber/emitter (SSA/E) for daytime photothermal and nighttime radiative sky cooling modes depending on the phase transition of the vanadium dioxide coated layer. A 24-hour day-night test showed that the fabricated SSA/E has continuous energy harvesting ability and improved overall energy utilization performance, thus showing remarkable potential in future energy applications.

cond-mat.mtrl-sci↗

arXiv2020

ManiGAN: Text-Guided Image Manipulation

The goal of our paper is to semantically edit parts of an image matching a given text that describes desired attributes (e.g., texture, colour, and background), while preserving other contents that are irrelevant to the text. To achieve this, we propose a novel generative adversarial network (ManiGAN), which contains two key components: text-image affine combination module (ACM) and detail correction module (DCM). The ACM selects image regions relevant to the given text and then correlates the regions with corresponding semantic words for effective manipulation. Meanwhile, it encodes original image features to help reconstruct text-irrelevant contents. The DCM rectifies mismatched attributes and completes missing contents of the synthetic image. Finally, we suggest a new metric for evaluating image manipulation results, in terms of both the generation of new attributes and the reconstruction of text-irrelevant contents. Extensive experiments on the CUB and COCO datasets demonstrate the superior performance of the proposed method. Code is available at https://github.com/mrlibw/ManiGAN.

cs.CV↗

arXiv2020

Singularities of plane gravitational waves and their memory effects

Similar to the Schwarzschild coordinates for spherical black holes, the Baldwin, Jeffery and Rosen (BJR) coordinates for plane gravitational waves are often singular, and extensions beyond such singularities are necessary, before studying asymptotic properties of such spacetimes at the null infinity of the plane, on which the gravitational waves propagate. The latter is closely related to the studies of memory effects and soft graviton theorems. In this paper, we point out that in the BJR coordinates all the spacetimes are singular physically at the focused point $u = u_s$, except for the two cases: (1) $α=1/2, \; \forall \; χ_n$; and (2) $α=1, \; χ_i = 0\; (i = 1, 2, 3)$, where $χ_n$ are the coefficients in the expansion $χ\equiv \left[{\mbox{det}}\left(g_{ab}\right) \right]^{1/4} = \left(u - u_s\right)^α\sum_{n = 0}^{\infty}χ_n \left(u - u_s\right)^n$ with $χ_0 \not= 0$, the constant $α\in (0, 1]$ characterizes the strength of the singularities, and $g_{ab}$ denotes the reduced metric on the two-dimensional plane orthogonal to the propagation direction of the wave. Therefore, the hypersurfaces $u= u_s$ already represent the boundaries of such spacetimes, and the null infinity does not belong to them. As a result, they cannot be used to study properties of plane gravitational waves at null infinities, including memory effects and soft graviton theorems.

gr-qc↗

arXiv2020

Image-to-Image Translation with Text Guidance

The goal of this paper is to embed controllable factors, i.e., natural language descriptions, into image-to-image translation with generative adversarial networks, which allows text descriptions to determine the visual attributes of synthetic images. We propose four key components: (1) the implementation of part-of-speech tagging to filter out non-semantic words in the given description, (2) the adoption of an affine combination module to effectively fuse different modality text and image features, (3) a novel refined multi-stage architecture to strengthen the differential ability of discriminators and the rectification ability of generators, and (4) a new structure loss to further improve discriminators to better distinguish real and synthetic images. Extensive experiments on the COCO dataset demonstrate that our method has a superior performance on both visual realism and semantic consistency with given descriptions.

cs.CV↗

arXiv2020

Super-resolution in recovering embedded electromagnetic sources in high contrast media

The purpose of this work is to provide a rigorous mathematical analysis of the expected super-resolution phenomenon in the time-reversal imaging of electromagnetic (EM) radiating sources embedded in a high contrast medium. It is known that the resolution limit is essentially determined by the sharpness of the imaginary part of the EM Green's tensor for the associated background. We first establish the close connection between the resolution and the material parameters and the resolvent of the electric integral operator, via the Lippmann-Schwinger representation formula. We then present an insightful characterization of the spectral structure of the integral operator for a general bounded domain and derive the pole-pencil decomposition of its resolvent in the high contrast regime. For the special case of a spherical domain, we provide some quantitative asymptotic behavior of the eigenvalues and eigenfunctions. These mathematical findings shall enable us to provide a concise and rigorous illustration of the super-resolution in the EM source reconstruction in high contrast media. Some numerical examples are also presented to verify our main theoretical results.

math.AP↗

arXiv2019

Controllable Text-to-Image Generation

In this paper, we propose a novel controllable text-to-image generative adversarial network (ControlGAN), which can effectively synthesise high-quality images and also control parts of the image generation according to natural language descriptions. To achieve this, we introduce a word-level spatial and channel-wise attention-driven generator that can disentangle different visual attributes, and allow the model to focus on generating and manipulating subregions corresponding to the most relevant words. Also, a word-level discriminator is proposed to provide fine-grained supervisory feedback by correlating words with image regions, facilitating training an effective generator which is able to manipulate specific visual attributes without affecting the generation of other content. Furthermore, perceptual loss is adopted to reduce the randomness involved in the image generation, and to encourage the generator to manipulate specific attributes required in the modified text. Extensive experiments on benchmark datasets demonstrate that our method outperforms existing state of the art, and is able to effectively manipulate synthetic images using natural language descriptions. Code is available at https://github.com/mrlibw/ControlGAN.

cs.CV↗

arXiv2019

Balancibility: Existence and Uniqueness of Power Flow Solutions under Voltage Balance Requirements

In distribution systems, power injection variability due to growing penetrations of distributed energy resources (DERs) and dispatchable loads can lead to power quality issues such as severe voltage unbalance. To ensure the safe operation of phase-balance-sensitive components such as transformers and induction motor loads, the amount of voltage unbalance must be maintained within specified limits for a range of uncertain loading conditions. This paper builds on existing "solvability conditions" that characterize operating regions for which the power flow equations are guaranteed to have a unique high-voltage solution. We extend these existing solvability conditions to be applicable to distribution systems and augment them with a "balancibility" condition which quantifies an operating region within which a unique, adequately balanced power flow solution exists. To build this condition, we consider different unbalance definitions and derive closed-form representations through reformulations or safe approximations. Using case studies, we evaluate these closed-form representations and compare the balancibility conditions associated with different unbalance definitions.

math.OC↗

arXiv2019

Real-Time FPGA-based Digital Signal Processing and Correction for a Small Animal PET

Small animal Positron Emission Tomography (PET) is dedicated to small animal imaging, which requires high position and energy precision, as well as good flexibility and efficiency of the electronics. This paper presents the design of a digital signal processing logic for a marmoset brain PET system based on LYSO crystal arrays, SiPMs, and the resistive network readout method. We implement 32-channel signal processing in a single Xilinx Artix-7 Field-Programmable Gate Array (FPGA). The logic is designed to support four online modes which are regular data processing mode, flood map construction mode, energy spectrum construction mode, and raw data mode. Several functions are integrated, including two-dimensional (2D) raw position calculation, crystal locating, events filtering, and synchronization detection. Furthermore, a series of online corrections is also integrated, such as photon peak correction to 511 keV and time measurement result correction with crystal granularity. A Gigabit Ethernet interface is utilized for data transfer, Look-Up Tables (LUTs) configuration, and command issuing. The pipeline logic works at 125 MHz with a signal processing capability beyond the required data rate of 1,000,000 events/s/channel. A series of initial tests are conducted. The results indicate that the logic design meets the application requirement.

physics.ins-det↗

arXiv2019

Prototype design of singles processing unit for the small animal PET

Position Emission Tomography (PET) is an advanced clinical diagnostic imaging technique for nuclear medicine. Small animal PET is increasingly uesd for studying the animal model of disease, new drugs and new therapies. A prototype of Singles Processing Unit (SPU) for a small animal PET system was designed to obtain the time, energy, and position information. The energy and position is actually calculated through high precison charge measurement, which is based on amplification, shaping, A/D conversion and area calculation in digital signal processing domian. Analysis and simulations were also conducted to optimize the key parameters in system design. Initial tests indicate that the charge and time precision is better than 0.3% FWHM and 350 ps FWHM respectively, while the position resolution is better than 0.35% FWHM. Commination tests of the SPU prototype with the PET detector indicate that the system time precision is better than 2.5 ns, while the flood map and energy spectra concored well with the expected.

physics.ins-det↗

arXiv2019

The Value of Including Unimodality Information in Distributionally Robust Optimal Power Flow

To manage renewable generation and load consumption uncertainty, chance-constrained optimal power flow (OPF) formulations and various solution methodologies have been proposed. However, conventional solution approaches often rely on accurate estimates of uncertainty distributions, which may not exist. When the distributions are not known but can be limited to a set of plausible distributions, termed an ambiguity set, distributionally robust (DR) optimization can be used to ensure that chance constraints hold for all distributions in that set. However, DR OPF yields conservative solutions if the ambiguity set is too large. In this paper, we assess the value of using both moment and unimodality information, which shrinks the ambiguity set and reduces conservatism, in DR OPF problems. Most practical uncertainty distributions in power systems are unimodal. Exact reformulations, approximations, and efficient solving techniques were developed in a previous paper. This paper develops an optimal parameter selection approach that searches for an optimal approximation, significantly improving the computational efficiency and solution quality. We evaluate the performance of the approach against existing chance-constrained OPF approaches using modified IEEE 118-bus and 300-bus systems with high penetrations of renewable generation. Results show that including unimodality information reduces solution conservatism and cost without significantly degrading reliability.

math.OC↗

arXiv2019

Behavioral similarity of dissipative solitons in an ultrafast fiber laser

Dissipative solitons are non-dispersive localized waves that exhibit intriguing dynamics. In optical systems, the discovery of new soliton dynamics and how to accurately manipulate dissipative solitons are still largely unexplored and challenging in the ultrafast fiber laser community. Herein, we unveil a new type of dissipative soliton behavior in a net-normal-dispersion bidirectional ultrafast fiber laser. That is, the bidirectional dissipative solitons will always reveal similar spectral and temporal characteristics through common gain and loss modulation, even if the transient instability is involved. The behavioral similarity enables us to accurately design the soliton patterns by introducing seed pulses through loss modulation. As a proof-of-concept application, the precise and flexible manipulation of multi-soliton patterns is demonstrated. These findings will shed new insights into the complex dissipative soliton dynamics, and also beneficial for designing ultrafast lasers with desirable soliton patterns for practical applications.

physics.optics↗

arXiv2019

Mathematical analysis of electromagnetic plasmonic metasurfaces

We study the anomalous electromagnetic scattering in the homogenization regime, by a subwavelength thin layer of periodically distributed plasmonic nanoparticles on a perfect conducting plane. By using layer potential techniques, we derive the asymptotic expansion of the electromagnetic field away from the thin layer and quantitatively analyze the field enhancement due to the mixed collective plasmonic resonances, which can be characterized by the spectra of periodic Neumann-Poincaré type operators. Based on the asymptotic behavior of the scattered field in the macroscopic scale, we further demonstrate that the optical effect of this thin layer can be effectively approximated by a Leontovich boundary condition, which is uniformly valid no matter whether the incident frequency is near the resonant range but varies with the magnetic property of the plasmonic nanoparticles. The quantitative approximation clearly shows the blow-up of the field energy and the conversion of polarization when resonance occurs, resulting in a significant change of the reflection property of the conducting plane. These results confirm essential physical changes of electromagnetic metasurface at resonances mathematically, whose occurrence was verified earlier for the acoustic case \cite{ammari2017bubble} and the transverse magnetic case \cite{ammari2016mathematical}.

math.AP↗