arXiv subjects

Qiong Liu

Publications and source records attributed to Qiong Liu.

At least 73 records · Page 4Linked to original sources

arXiv2022

Reducing Learning Difficulties: One-Step Two-Critic Deep Reinforcement Learning for Inverter-based Volt-Var Control

A one-step two-critic deep reinforcement learning (OSTC-DRL) approach for inverter-based volt-var control (IB-VVC) in active distribution networks is proposed in this paper. Firstly, considering IB-VVC can be formulated as a single-period optimization problem, we formulate the IB-VVC as a one-step Markov decision process rather than the standard Markov decision process, which simplifies the DRL learning task. Then we design the one-step actor-critic DRL scheme which is a simplified version of recent DRL algorithms, and it avoids the issue of Q value overestimation successfully. Furthermore, considering two objectives of VVC: minimizing power loss and eliminating voltage violation, we utilize two critics to approximate the rewards of two objectives separately. It simplifies the approximation tasks of each critic, and avoids the interaction effect between two objectives in the learning process of critic. The OSTC-DRL approach integrates the one-step actor-critic DRL scheme and the two-critic technology. Based on the OSTC-DRL, we design two centralized DRL algorithms. Further, we extend the OSTC-DRL to multi-agent OSTC-DRL for decentralized IB-VVC and design two multi-agent DRL algorithms. Simulations demonstrate that the proposed OSTC-DRL has a faster convergence rate and a better control performance, and the multi-agent OSTC-DRL works well for decentralized IB-VVC problems.

cs.AI↗

arXiv2022

Resolvent Analysis of an Under-expanded Planar Supersonic Impinging Jet

This investigation aims to assess the effect of different types of actuator forcing on the feedback loop of an under-expanded Mach 1.27 planar impinging jet using a resolvent framework. To this end, we employ a Large Eddy Simulation database as a truth model. The time and spanwise-averaged mean flow is taken as an input to global stability and resolvent analyses with the purpose of examining both the intrinsic instability and input-output characteristics. The results show that the inherent instability and primary energy amplification are attributed to the Kelvin-Helmholtz (K-H) instability. Moreover, the K-H response modes obtained from the resolvent analysis are in reasonable agreement with Spectral Proper Orthogonal Decomposition (SPOD) modes from the unsteady LES data. Insights into noise control are obtained by localizing the actuator forcing to the nozzle lip and the ground plate by imposing component-wise forcing to mimic different notional actuators. It is observed that energy amplification obtained for the localized component-wise forcing is different from the global resolvent analysis and dependent on the type of actuator. This provides insights into the type, wavenumber, and frequency of actuators for active flow control.

physics.flu-dyn↗

arXiv2021

ELSA: Enhanced Local Self-Attention for Vision Transformer

Self-attention is powerful in modeling long-range dependencies, but it is weak in local finer-level feature learning. The performance of local self-attention (LSA) is just on par with convolution and inferior to dynamic filters, which puzzles researchers on whether to use LSA or its counterparts, which one is better, and what makes LSA mediocre. To clarify these, we comprehensively investigate LSA and its counterparts from two sides: \emph{channel setting} and \emph{spatial processing}. We find that the devil lies in the generation and application of spatial attention, where relative position embeddings and the neighboring filter application are key factors. Based on these findings, we propose the enhanced local self-attention (ELSA) with Hadamard attention and the ghost head. Hadamard attention introduces the Hadamard product to efficiently generate attention in the neighboring case, while maintaining the high-order mapping. The ghost head combines attention maps with static matrices to increase channel capacity. Experiments demonstrate the effectiveness of ELSA. Without architecture / hyperparameter modification, drop-in replacing LSA with ELSA boosts Swin Transformer \cite{swin} by up to +1.4 on top-1 accuracy. ELSA also consistently benefits VOLO \cite{volo} from D1 to D5, where ELSA-VOLO-D5 achieves 87.2 on the ImageNet-1K without extra training images. In addition, we evaluate ELSA in downstream tasks. ELSA significantly improves the baseline by up to +1.9 box Ap / +1.3 mask Ap on the COCO, and by up to +1.9 mIoU on the ADE20K. Code is available at \url{https://github.com/damo-cv/ELSA}.

cs.CV↗

arXiv2021

Varifocal Multiview Images: Capturing and Visual Tasks

Multiview images have flexible field of view (FoV) but inflexible depth of field (DoF). To overcome the limitation of multiview images on visual tasks, in this paper, we present varifocal multiview (VFMV) images with flexible DoF. VFMV images are captured by focusing a scene on distinct depths by varying focal planes, and each view only focused on one single plane.Therefore, VFMV images contain more information in focal dimension than multiview images, and can provide a rich representation for 3D scene by considering both FoV and DoF. The characteristics of VFMV images are useful for visual tasks to achieve high quality scene representation. Two experiments are conducted to validate the advantages of VFMV images in 4D light field feature detection and 3D reconstruction. Experiment results show that VFMV images can detect more light field features and achieve higher reconstruction quality due to informative focus cues. This work demonstrates that VFMV images have definite advantages over multiview images in visual tasks.

eess.IV↗

arXiv2021

Global stability analysis of flow behind an upswept aftbody

Wakes of aircraft and automobiles with relatively flat slanted aftbodies are often characterized by a streamwise-oriented vortex pair, whose strength affects drag and other crucial performance parameters. We examine the stability characteristics of the vortex pair emerging over an abstraction comprised of a streamwise-aligned cylinder terminated with an upswept plane. The Reynolds number is fixed at 5000 and the upsweep angle is increased from 20deg to 32deg. At 20deg, the LES yields a steady streamwise-oriented vortex pair, and the global modes are also stable. At 32deg, the LES displays unsteady flow behavior. Linear analysis of the mean flow reveals different unstable modes. The lowest oscillation frequency is an antisymmetric mode, which is attached to the entire slanted base. At the highest frequency, the mode is symmetric and has the same rotational orientation as the mean vortex pair. Its support is prominent in the rear part of the slanted base and spreads relatively rapidly downstream with prominent helical structures. A receptivity analysis of low- and high-frequency modes suggests the latter holds promise to affect the vortical flow, providing a potential starting point for a control strategy to modify the vortex pair.

physics.flu-dyn↗

arXiv2021

Decoupled Dynamic Filter Networks

Convolution is one of the basic building blocks of CNN architectures. Despite its common use, standard convolution has two main shortcomings: Content-agnostic and Computation-heavy. Dynamic filters are content-adaptive, while further increasing the computational overhead. Depth-wise convolution is a lightweight variant, but it usually leads to a drop in CNN performance or requires a larger number of channels. In this work, we propose the Decoupled Dynamic Filter (DDF) that can simultaneously tackle both of these shortcomings. Inspired by recent advances in attention, DDF decouples a depth-wise dynamic filter into spatial and channel dynamic filters. This decomposition considerably reduces the number of parameters and limits computational costs to the same level as depth-wise convolution. Meanwhile, we observe a significant boost in performance when replacing standard convolution with DDF in classification networks. ResNet50 / 101 get improved by 1.9% and 1.3% on the top-1 accuracy, while their computational costs are reduced by nearly half. Experiments on the detection and joint upsampling networks also demonstrate the superior performance of the DDF upsampling variant (DDF-Up) in comparison with standard convolution and specialized content-adaptive layers.

cs.CV↗

arXiv2021

Adaptive Feedforward Neural Network Control with an Optimized Hidden Node Distribution

Composite adaptive radial basis function neural network (RBFNN) control with a lattice distribution of hidden nodes has three inherent demerits: 1) the approximation domain of adaptive RBFNNs is difficult to be determined a priori; 2) only a partial persistence of excitation (PE) condition can be guaranteed; and 3) in general, the required number of hidden nodes of RBFNNs is enormous. This paper proposes an adaptive feedforward RBFNN controller with an optimized distribution of hidden nodes to suitably address the above demerits. The distribution of the hidden nodes calculated by a K-means algorithm is optimally distributed along the desired state trajectory. The adaptive RBFNN satisfies the PE condition for the periodic reference trajectory. The weights of all hidden nodes will converge to the optimal values. This proposed method considerably reduces the number of hidden nodes, while achieving a better approximation ability. The proposed control scheme shares a similar rationality to that of the classical PID control in two special cases, which can thus be seen as an enhanced PID scheme with a better approximation ability. For the controller implemented by digital devices,the proposed method, for a manipulator with unknown dynamics, potentially achieves better control performance than model-based schemes with accurate dynamics.Simulation results demonstrate the effectiveness of the proposed scheme. This result provides a deeper insight into the coordination of the adaptive neural network control and the deterministic learning theory.

eess.SY↗

arXiv2021

Acoustic response of turbulent cavity flow using resolvent analysis

The decomposition of hydrodynamic and acoustic components of cavity flows can aid the understanding of the flow-acoustic interaction, which produces a variety of adverse effects in applications. We apply the approach of combining the resolvent analysis and Doak's momentum potential theory to examine the input-output flow-acoustic characteristics of compressible flow over an open cavity at Re=10,000. The methodology can decompose hydrodynamic and acoustic components from an input-output framework. We further localize the forcing at the leading-edge and front wall of the cavity and filter with a single component of velocity, density, and temperature forcing. For the subsonic flow at M=0.6, The strong acoustic component appears at the trailing edge of the cavity at a lower frequency. while as the frequency increases, the intense acoustic structure moves close to upstream and overlap with the hydrodynamic component. This tendency is also observed in the supersonic flow condition M=1.4. Compared to different types of forcing, the streamwise velocity forcing achieves the largest energy amplification. However, it substantially reduces by moving the forcing location from the leading-edge to the front wall of the cavity. This work provides an alternative approach to examine input-output flow-acoustic characteristics where resonance is dominant and provides guidance for the development of flow control strategies.

physics.flu-dyn↗

arXiv2020

FREA-Unet: Frequency-aware U-net for Modality Transfer

While Positron emission tomography (PET) imaging has been widely used in diagnosis of number of diseases, it has costly acquisition process which involves radiation exposure to patients. However, magnetic resonance imaging (MRI) is a safer imaging modality that does not involve patient's exposure to radiation. Therefore, a need exists for an efficient and automated PET image generation from MRI data. In this paper, we propose a new frequency-aware attention U-net for generating synthetic PET images. Specifically, we incorporate attention mechanism into different U-net layers responsible for estimating low/high frequency scales of the image. Our frequency-aware attention Unet computes the attention scores for feature maps in low/high frequency layers and use it to help the model focus more on the most important regions, leading to more realistic output images. Experimental results on 30 subjects from Alzheimers Disease Neuroimaging Initiative (ADNI) dataset demonstrate good performance of the proposed model in PET image synthesis that achieved superior performance, both qualitative and quantitative, over current state-of-the-arts.

eess.IV↗

arXiv2020

Using Sensory Time-cue to enable Unsupervised Multimodal Meta-learning

As data from IoT (Internet of Things) sensors become ubiquitous, state-of-the-art machine learning algorithms face many challenges on directly using sensor data. To overcome these challenges, methods must be designed to learn directly from sensors without manual annotations. This paper introduces Sensory Time-cue for Unsupervised Meta-learning (STUM). Different from traditional learning approaches that either heavily depend on labels or on time-independent feature extraction assumptions, such as Gaussian distribution features, the STUM system uses time relation of inputs to guide the feature space formation within and across modalities. The fact that STUM learns from a variety of small tasks may put this method in the camp of Meta-Learning. Different from existing Meta-Learning approaches, STUM learning tasks are composed within and across multiple modalities based on time-cue co-exist with the IoT streaming data. In an audiovisual learning example, because consecutive visual frames usually comprise the same object, this approach provides a unique way to organize features from the same object together. The same method can also organize visual object features with the object's spoken-name features together if the spoken name is presented with the object at about the same time. This cross-modality feature organization may further help the organization of visual features that belong to similar objects but acquired at different location and time. Promising results are achieved through evaluations.

cs.CV↗

arXiv2020

A WISE View of IRAS Debris Disks: Revising the Dust Properties

Debris disks around stars are considered as components of planetary systems. Constrain the dust properties of these disks can give crucial information to formation and evolution of planetary systems. As an all-sky survey, \textit{InfRared Astronomical Satellite} (\iras) gave great contribution to the debris disk searching which discovered the first debris disk host star (Vega). The \iras-detected debris disk sample published by Rhee \citep{rhe07} contains 146 stars with detailed information of dust properties. While the dust properties of 45 of them still can not be determined due to the limitations with the \iras\ database (have \iras\ detection at 60 $μ$m only). Therefore, using more sensitivity data of \textit{Wide-field Infrared Survey Explorer} (\wise), we can better characterize the sample stars: For the stars with \iras\ detection at 60 $μ$m only, we refit the excessive flux densities and obtain the dust temperatures and fractional luminosities; While for the remaining stars with multi-bands \iras\ detections, the dust properties are revised which show that the dust temperatures were over estimated in high temperatures band before. Moreover, we identify 17 stars with excesses at the \wise\ 22 $μ$m which have smaller distribution of distance from Earth and higher fractional luminosities than the other stars without mid-infrared excess emission. Among them, 15 stars can be found in previous works.

astro-ph.SR↗

arXiv2020

Bright Debris Disk Candidates Observed with AKARI/Far-Infrared Surveyor (FIS)

We cross-correlate \hip\ main-sequence star catalog with \fis\ catalog, and identify 136 stars (at $>90$% reliability) with far-infrared detections at least in one band. After rejecting 51 stars classified as young stellar objects, Be stars, other type stars with known dust disks or with potential contaminations and 2 stars without infrared excess emission, we obtain a sample of 83 candidate stars with debris disks. Stars in our sample cover spectral types from B to K-types with most being early types. This represents an unique sample of luminous debris disks that derived uniformly from all sky survey with a spatial resolution a factor of two better than the previous such survey by \iras. Moreover, by collecting the infrared photometric data from other public archives, 85% of them have infrared excesses in more than one bands, allowing the estimate of the dust temperatures. We fit the blackbody model to the broad band spectral energy distribution of these stars to derive the statistical distribution of the disk parameters. 7 stars require an additional warm component of temperature around 200 K. While a substantial fraction of our sample(58 stars) have weak 12 \micron\ excess indicating that a warm dust component maybe common among these bright debris disk systems.

astro-ph.EP↗

arXiv2020

Unsteady control of supersonic turbulent cavity flow based on resolvent analysis

We use resolvent analysis to determine an unsteady active control setup to attenuate pressure fluctuations in turbulent supersonic flow over a rectangular cavity with a length-to-depth ratio of 6 at a Mach number of 1.4 and a Reynolds number based on cavity depth of 10,000. Large-eddy simulations (LES) and dynamic modal decomposition (DMD) of the supersonic cavity flow reveal the dominance of two-dimensional Rossiter modes II and IV. These predominantly two-dimensional vortical structures generate high-amplitude unsteadiness over the cavity through trailing-edge impingement and create oblique shock waves by obstructing the freestream. To disrupt the undesired formation of vortical structures, we introduce three-dimensional unsteady forcing along the cavity leading edge to generate streamwise vortical structures. Resolvent analysis with respect to the time-averaged base flow is leveraged to determine the optimal combination of forcing frequency and spanwise wavenumber. Instead of selecting the most amplified resolvent forcing modes, we seek the combination of control parameters that yields sustained amplification of the primary resolvent-based kinetic energy distribution over the entire length of the cavity. The sustained amplification is critical to ensure that the selected forcing input remains effective to prevent the formation of the large spanwise vortices. This resolvent-analysis-based flow control guideline is validated with a number of companion LES of the controlled cavity flows. The optimal control setup is verified to be the most effective in reducing the pressure root-mean-square level up to 52% along the aft and bottom cavity walls compared to the baseline cavity flow. The present flow control guideline derived from resolvent analysis should be applicable to flows that require the actuation input to remain effective over an extended region of interest.

physics.flu-dyn↗

arXiv2020

Active flow control of a pump-induced wall-normal vortex with steady blowing

The emergence of a submerged vortex upstream of a pump can reduce pump intake efficiency and cause structural damage. In this study, we consider the use of active flow control with steady blowing to increase the pressure distribution within a single-phase pump-induced wall-normal vortex model, which is based on the Burgers vortex with a no-slip boundary condition prescribed along its symmetry plane. The goal of our control is to modify the vortex core velocity profile. These changes are sought to increase the core pressure such that detrimental effects on the pump are alleviated. Three-dimensional direct numerical simulations (DNS) are performed to examine the dynamics of the vortex with the application of axial momentum injection at and around the root of the vortex. We find that the active flow control approach can effectively modify the wall-normal vortical structure and significantly increase the low-core pressure by up to 81% compared to that of the uncontrolled case. The result shows that the control setup is also effective when it is introduced in an off-centered manner. Compared to the unsteady blowing and suction based actuation from our previous work (Liu et al. 2018), the current steady control technique offers an effective and simple flow control setup that can support robust operations of pumps.

physics.flu-dyn↗

arXiv2019

Resolvent analysis of compressible laminar and turbulent cavity flows

The present work demonstrates the use of resolvent analysis to obtain physical insights for open-cavity flows. Resolvent analysis identifies the flow response to harmonic forcing, given a steady base state, in terms of the response and forcing modes and the amplification gain. The response and forcing modes reveal the spatial structures associated with this amplification process. In this study, we perform resolvent analysis on both laminar and turbulent flows over a rectangular cavity with length-to-depth ratio of $L/D=6$ at a free stream Mach number of $M_\infty=0.6$ in a spanwise periodic setting. Based on the dominant instability of the base state, a discount parameter is introduced to resolvent analysis to examine the harmonic characteristics over a finite-time window. We first uncover the underlying flow physics and interpret findings from laminar flow at $Re_D = 502$. These findings from laminar flow are extended to a more practical cavity flow example at a much higher Reynolds number of $Re_D = 10^4$. The features of response and forcing modes from the laminar and turbulent cavity flows are similar to the spatial structures from the laminar analysis. We further find that the large amplification of energy in flow response is associated with high frequency for turbulent flow, while the flow is more responsive to low frequency excitation in the laminar case. These findings from resolvent analysis provide valuable insights for flow control studies with regard to parameter selection and placement of actuators and sensors.

physics.flu-dyn↗

arXiv2019

Atomic-Scale Investigation on the Ultra-large Bending Behaviours of Layered Sodium Titanate Nanowires

Study on mechanical properties of one-dimensional layered titanate nanomaterials is crucial since they demonstrate important applications in various fields. Here, we conducted ex situ and in situ atomic-scale investigation on bending properties of a kind of ceramic layered titanate (Na2Ti2O4(OH)2) nanowires in a transmission electron microscopy. The nanowires showed flexibility along <100> direction and could obtain a maximum bending strain of nearly 37%. By analysing the defect behaviours, the unique bending properties of this ceramic material was found to correlate with a novel arrangement of dislocations, an accessible nucleation and movement along the axial direction resulting from the weak electrostatic interaction between the TiO6 layers and the low b/a ratio. These results provide pioneering and key understanding on bending behaviours of layered titanate nanowire families and potentially other one-dimensional nanomaterials with layered crystalline structures.

cond-mat.mtrl-sci↗

arXiv2019

A physical mechanism of heterogeneity in stem cell, cancer and cancer stem cell

Heterogeneity is ubiquitous in stem cells (SC), cancer cells (CS), and cancer stem cells (CSC). SC and CSC heterogeneity is manifested as diverse sub-populations with self-renewing and unique regeneration capacity. Moreover, the CSC progeny possesses multiple plasticity and cancerous characteristics. Many studies have demonstrated that cancer heterogeneity is one of the greatest obstacle for therapy. This leads to the incomplete anti-cancer therapies and transitory efficacy. Furthermore, numerous micro-metastasis leads to the wide spread of the tumor cells across the body which is the beginning of metastasis. The epigenetic processes (DNA methylation or histone remodification etc.) can provide a source for certain heterogeneity. In this study, we develop a mathematical model to quantify the heterogeneity of SC, CSC and cancer taking both genetic and epigenetic effects into consideration. We uncovered the roles and physical mechanisms of heterogeneity from the three aspects (SC, CSC and cancer). In the adiabatic regime (relatively fast regulatory binding and effective coupling among genes), seven native states (SC, CSC, Cancer, Premalignant, Normal, Lesion and Hyperplasia) emerge. In non-adiabatic regime (relatively slow regulatory binding and effective weak coupling among genes), multiple meta-stable SC, CS, CSC and differentiated states emerged which can explain the origin of heterogeneity. In other words, the slow regulatory binding mimicking the epigenetics can give rise to heterogeneity. Elucidating the origin of heterogeneity and dynamical interrelationship between intra-tumoral cells has clear clinical significance in helping to understand the cellular basis of treatment response, therapeutic resistance, and tumor relapse.

q-bio.MN↗

arXiv2018

Quantification of the underlying mechanisms and relationship among cancer, metastasis and differentiation/development

Recurrence and metastasis have been regarded as two of the greatest obstacles for curing cancer. Cancer stem cell (CSC) have been found. They contribute to cancer development with the distinct feature of recurrence and resistance to the popular treatments such as drugs and chemotherapy. In addition, recent discoveries suggest that the epithelial mesenchymal transition (EMT) is an essential process in normal embryogenesis and tissue repair, which is a required step in cancer metastasis. Although there are many indications showing the connections between metastasis and stem cell, researches often studied them separately or at most bi-laterally, not in an integrated way. In this study, we aim at exploring the global mechanisms and interrelationship among cancer, development and metastasis which are currently poorly understood. To start, we constructed a core gene regulatory network motif which contain specific genes and microRNAs of CSC, EMT and cancer. We uncovered seven distinct states emerged from the underlying landscape. They are identified as Normal, Premalignant, Cancer, stem cell (SC), cancer stem cell (CSC), Lesion and Hyperlasia state. Given the biological definition of each state, we also discussed the metastasis ability of each state. We show how and which types of cells can be transformed to a cancer state and the connections among cancer, CSC and EMT. The barrier height and flux of the kinetic paths are explored to quantify how and which cells switch stochastically between the states. Our landscape model provides a quantitative way which reveals the global mechanisms of cancer, development and metastasis.

q-bio.CB↗