arXiv subjects

Feng Liu

Publications and source records attributed to Feng Liu.

At least 577 records · Page 32Linked to original sources

arXiv2020

Domain Contrast for Domain Adaptive Object Detection

We present Domain Contrast (DC), a simple yet effective approach inspired by contrastive learning for training domain adaptive detectors. DC is deduced from the error bound minimization perspective of a transferred model, and is implemented with cross-domain contrast loss which is plug-and-play. By minimizing cross-domain contrast loss, DC guarantees the transferability of detectors while naturally alleviating the class imbalance issue in the target domain. DC can be applied at either image level or region level, consistently improving detectors' transferability and discriminability. Extensive experiments on commonly used benchmarks show that DC improves the baseline and state-of-the-art by significant margins, while demonstrating great potential for large domain divergence.

cs.CV↗

arXiv2020

Bridging the Theoretical Bound and Deep Algorithms for Open Set Domain Adaptation

In the unsupervised open set domain adaptation (UOSDA), the target domain contains unknown classes that are not observed in the source domain. Researchers in this area aim to train a classifier to accurately: 1) recognize unknown target data (data with unknown classes) and, 2) classify other target data. To achieve this aim, a previous study has proven an upper bound of the target-domain risk, and the open set difference, as an important term in the upper bound, is used to measure the risk on unknown target data. By minimizing the upper bound, a shallow classifier can be trained to achieve the aim. However, if the classifier is very flexible (e.g., deep neural networks (DNNs)), the open set difference will converge to a negative value when minimizing the upper bound, which causes an issue where most target data are recognized as unknown data. To address this issue, we propose a new upper bound of target-domain risk for UOSDA, which includes four terms: source-domain risk, $ε$-open set difference ($Δ_ε$), a distributional discrepancy between domains, and a constant. Compared to the open set difference, $Δ_ε$ is more robust against the issue when it is being minimized, and thus we are able to use very flexible classifiers (i.e., DNNs). Then, we propose a new principle-guided deep UOSDA method that trains DNNs via minimizing the new upper bound. Specifically, source-domain risk and $Δ_ε$ are minimized by gradient descent, and the distributional discrepancy is minimized via a novel open-set conditional adversarial training strategy. Finally, compared to existing shallow and deep UOSDA methods, our method shows the state-of-the-art performance on several benchmark datasets, including digit recognition (MNIST, SVHN, USPS), object recognition (Office-31, Office-Home), and face recognition (PIE).

cs.LG↗

arXiv2020

Local Facial Makeup Transfer via Disentangled Representation

Facial makeup transfer aims to render a non-makeup face image in an arbitrary given makeup one while preserving face identity. The most advanced method separates makeup style information from face images to realize makeup transfer. However, makeup style includes several semantic clear local styles which are still entangled together. In this paper, we propose a novel unified adversarial disentangling network to further decompose face images into four independent components, i.e., personal identity, lips makeup style, eyes makeup style and face makeup style. Owing to the further disentangling of makeup style, our method can not only control the degree of global makeup style, but also flexibly regulate the degree of local makeup styles which any other approaches can't do. For makeup removal, different from other methods which regard makeup removal as the reverse process of makeup, we integrate the makeup transfer with the makeup removal into one uniform framework and obtain multiple makeup removal results. Extensive experiments have demonstrated that our approach can produce more realistic and accurate makeup transfer results compared to the state-of-the-art methods.

cs.CV↗

arXiv2020

Assessing the Impact of COVID-19 on the Objective and Analysis of Oncology Clinical Trials -- Application of the Estimand Framework

COVID-19 outbreak has rapidly evolved into a global pandemic. The impact of COVID-19 on patient journeys in oncology represents a new risk to interpretation of trial results and its broad applicability for future clinical practice. We identify key intercurrent events that may occur due to COVID-19 in oncology clinical trials with a focus on time-to-event endpoints and discuss considerations pertaining to the other estimand attributes introduced in the ICH E9 addendum. We propose strategies to handle COVID-19 related intercurrent events, depending on their relationship with malignancy and treatment and the interpretability of data after them. We argue that the clinical trial objective from a world without COVID-19 pandemic remains valid. The estimand framework provides a common language to discuss the impact of COVID-19 in a structured and transparent manner. This demonstrates that the applicability of the framework may even go beyond what it was initially intended for.

q-bio.OT↗

arXiv2020

Event Arguments Extraction via Dilate Gated Convolutional Neural Network with Enhanced Local Features

Event Extraction plays an important role in information-extraction to understand the world. Event extraction could be split into two subtasks: one is event trigger extraction, the other is event arguments extraction. However, the F-Score of event arguments extraction is much lower than that of event trigger extraction, i.e. in the most recent work, event trigger extraction achieves 80.7%, while event arguments extraction achieves only 58%. In pipelined structures, the difficulty of event arguments extraction lies in its lack of classification feature, and the much higher computation consumption. In this work, we proposed a novel Event Extraction approach based on multi-layer Dilate Gated Convolutional Neural Network (EE-DGCNN) which has fewer parameters. In addition, enhanced local information is incorporated into word features, to assign event arguments roles for triggers predicted by the first subtask. The numerical experiments demonstrated significant performance improvement beyond state-of-art event extraction approaches on real-world datasets. Further analysis of extraction procedure is presented, as well as experiments are conducted to analyze impact factors related to the performance improvement.

cs.CL↗

arXiv2020

Asynchrony-Resilient and Privacy-Preserving Charging Protocol for Plug-in Electric Vehicles

The proliferation of plug-in electric vehicles (PEVs) advocates a distributed paradigm for the coordination of PEV charging. Distinct from existing primal-dual decomposition or consensus methods, this paper proposes a cutting-plane based distributed algorithm, which enables an asynchronous coordination while well preserving individual's private information. To this end, an equivalent surrogate model is first constructed by exploiting the duality of the original optimization problem, which masks the private information of individual users by a transformation. Then, a cutting-plane based algorithm is derived to solve the surrogate problem in a distributed manner with intrinsic superiority to cope with various asynchrony. Critical implementation issues, such as the distributed initialization, cutting-plane generation and localized stopping criteria, are discussed in detail. Numerical tests on IEEE 37- and 123-node feeders with real data show that the proposed method is resilient to a variety of asynchrony and admits the plug-and-play operation mode. It is expected the proposed methodology provides an alternative path toward a more practical protocol for PEV charging.

eess.SY↗

arXiv2020

A Unified View of Topological Phase Transition in Band Theory

We develop a unified view of topological phase transitions (TPTs) in solids by revising the classical band theory with the inclusion of topology. Re-evaluating the band evolution from an "atomic crystal" [a normal insulator (NI)] to a solid crystal, such as a semiconductor, we demonstrate that there exists ubiquitously an intermediate phase of topological insulator (TI), whose critical transition point displays a linear scaling between electron hopping potential and average bond length, underlined by deformation-potential theory. The validity of the scaling relation is verified in various two-dimensional (2D) lattices regardless of lattice symmetry, periodicity, and form of electron hoppings, based on a generic tight-binding model. Significantly, this linear scaling is shown to set an upper bound for the degree of structural disorder to destroy the topological order in a crystalline solid, as exemplified by formation of vacancies and thermal disorder. Our work formulates a simple framework for understanding the physical nature of TPTs with significant implications in practical applications of topological materials.

cond-mat.mtrl-sci↗

arXiv2020

Inter-sequence Enhanced Framework for Personalized Sequential Recommendation

Modeling the sequential correlation of users' historical interactions is essential in sequential recommendation. However, the majority of the approaches mainly focus on modeling the \emph{intra-sequence} item correlation within each individual sequence but neglect the \emph{inter-sequence} item correlation across different user interaction sequences. Though several studies have been aware of this issue, their method is either simple or implicit. To make better use of such information, we propose an inter-sequence enhanced framework for the Sequential Recommendation (ISSR). In ISSR, both inter-sequence and intra-sequence item correlation are considered. Firstly, we equip graph neural networks in the inter-sequence correlation encoder to capture the high-order item correlation from the user-item bipartite graph and the item-item graph. Then, based on the inter-sequence correlation encoder, we build GRU network and attention network in the intra-sequence correlation encoder to model the item sequential correlation within each individual sequence and temporal dynamics for predicting users' preferences over candidate items. Additionally, we conduct extensive experiments on three real-world datasets. The experimental results demonstrate the superiority of ISSR over many state-of-the-art methods and the effectiveness of the inter-sequence correlation encoder.

cs.IR↗

arXiv2020

Topological Phase Transitions in a Hybridized Three-Dimensional Topological Insulator

As the thickness of a three-dimensional (3D) topological insulator (TI) becomes comparable to the penetration depth of the surface states, quantum tunneling between surfaces turns their gapless Dirac electronic structure into a gapped surface state. Analytical formulation suggests that the hybridization gap scales exponentially with decrease in number of layers while the system oscillates between topologically trivial and non-trivial insulators. This work explores the transport properties of a 3D TI in the inter-surface hybridization regime. By experimentally probing the hybridization gap as a function of BiSbTeSe2 thickness using three different methods, we map the crossover from the 3D to 2D state. In the 2D topological state, we observe a finite longitudinal conductance at ~2e2/h when the Fermi level is aligned within the surface gap, indicating a quantum spin Hall (QSH) state. Additionally, we study the response of trivial and non-trivial hybridization gapped states modulated by external out-of-plane magnetic and electric fields. Our revelations of surface gap-closing and/or reopening features are strongly indicative of topological phase transitions (TPTs) in the hybridization gap regime, realizing magnetic/electric field switching between band insulating and QSH states with immense potential for practical applications.

cond-mat.mes-hall↗

arXiv2020

Distributed Generalized Nash Equilibrium Seeking for Energy Sharing Games

With the proliferation of distributed generators and energy storage systems, traditional passive consumers in power systems have been gradually evolving into the so-called "prosumers", i.e., proactive consumers, which can both produce and consume power. To encourage energy exchange among prosumers, energy sharing is increasingly adopted, which is usually formulated as a generalized Nash game (GNG). In this paper, a distributed approach is proposed to seek the Generalized Nash equilibrium (GNE) of the energy sharing game. To this end, we convert the GNG into an equivalent optimization problem. A Krasnosel'ski{ǐ}-Mann iteration type algorithm is thereby devised to solve the problem and consequently find the GNE in a distributed manner. The convergence of the proposed algorithm is proved rigorously based on the nonexpansive operator theory. The performance of the algorithm is validated by experiments with three prosumers, and the scalability is tested by simulations using 123 prosumers.

eess.SY↗

arXiv2020

Deep Homography Estimation for Dynamic Scenes

Homography estimation is an important step in many computer vision problems. Recently, deep neural network methods have shown to be favorable for this problem when compared to traditional methods. However, these new methods do not consider dynamic content in input images. They train neural networks with only image pairs that can be perfectly aligned using homographies. This paper investigates and discusses how to design and train a deep neural network that handles dynamic scenes. We first collect a large video dataset with dynamic content. We then develop a multi-scale neural network and show that when properly trained using our new dataset, this neural network can already handle dynamic scenes to some extent. To estimate a homography of a dynamic scene in a more principled way, we need to identify the dynamic content. Since dynamic content detection and homography estimation are two tightly coupled tasks, we follow the multi-task learning principles and augment our multi-scale network such that it jointly estimates the dynamics masks and homographies. Our experiments show that our method can robustly estimate homography for challenging scenarios with dynamic scenes, blur artifacts, or lack of textures.

cs.CV↗

arXiv2020

Global mild solutions to three-dimensional magnetohydrodynamic system in Morrey spaces

In this article, the Cauchy problem of three-dimensional (3-D) incompressible magnetohydrodynamic system was investigated. If the initial $\mathcal{M}^{1,1}$ norms of the vorticity $ω$ and the current density $j$ are both sufficiently small, then some uniform estimates with respect to time for the coupling terms between the fluid and the magnetic field can be established, which lead to a global-in-time well-posedness of mild solutions in Morrey spaces via some effective arguments.

math.AP↗

arXiv2020

Exponential Stability of Partial Primal-Dual Gradient Dynamics with Nonsmooth Objective Functions

In this paper, we investigate the continuous time partial primal-dual gradient dynamics (P-PDGD) for solving convex optimization problems with the form $ \min\limits_{x\in X,y\inΩ}\ f({x})+h(y),\ \textit{s.t.}\ A{x}+By=C $, where $ f({x}) $ is strongly convex and smooth, but $ h(y) $ is strongly convex and non-smooth. Affine equality and set constraints are included. We prove the exponential stability of P-PDGD, and bounds on decaying rates are provided. Moreover, it is also shown that the decaying rates can be regulated by setting the stepsize.

math.OC↗

arXiv2020

Softmax Splatting for Video Frame Interpolation

Differentiable image sampling in the form of backward warping has seen broad adoption in tasks like depth estimation and optical flow prediction. In contrast, how to perform forward warping has seen less attention, partly due to additional challenges such as resolving the conflict of mapping multiple pixels to the same target location in a differentiable way. We propose softmax splatting to address this paradigm shift and show its effectiveness on the application of frame interpolation. Specifically, given two input frames, we forward-warp the frames and their feature pyramid representations based on an optical flow estimate using softmax splatting. In doing so, the softmax splatting seamlessly handles cases where multiple source pixels map to the same target location. We then use a synthesis network to predict the interpolation result from the warped representations. Our softmax splatting allows us to not only interpolate frames at an arbitrary time but also to fine tune the feature pyramid and the optical flow. We show that our synthesis approach, empowered by softmax splatting, achieves new state-of-the-art results for video frame interpolation.

cs.CV↗

arXiv2020

Terahertz composite plasmonic slabs based on double-layer metallic gratings

A composite plasmonic slab based on double-layer metallic gratings and a dielectric film is experimentally and numerically demonstrated in terahertz (THz) regions, which can support two resonance modes and then form a broad bandgap (40%). As compared to the double-layer metal grating, the dielectric film in composite THz slabs significantly enhances the transmission of the transverse magnetic (TM) mode. Electric field vector proved that the low-frequency resonance mode originates from the symmetric plasmonic mode and the high-frequency resonance mode is induced by the hybrid mode of plasmonic and dielectric modes. The inherently near field coupling between metal gratings and dielectric film has been analyzed by changing the structural parameters. We further demonstrate that by tuning the metallic grating width, the plasmonic bandgap can be manipulated. These results suggest that this composite plasmonic slab has great potential for use as a filter, polarizer, and sensor in THz regions.

physics.optics↗

arXiv2020

Higher-order Quantum Spin Hall Effect of Light

Band topology and related spin (or pseudo-spin) physics of photons provide us with a new dimension for manipulating light, which is potentially useful for information communication and data storage. Especially, the quantum spin Hall effect of light, where electromagnetic waves propagate along surfaces of samples with strong spin-momentum locking, paves the way for achieving topologically protected photonic spin transport. Recently, the conventional bulk-edge correspondence of the band topology has been extended to higher-order cases that enables the explorations of topological states with codimensions larger than 1 such as hinge and corner states. Here, for the first time, we demonstrate a higherorder quantum spin Hall effect of light by utilizing an all-dielectric C6v-symmetric photonic crystal. We observe corner states with opposite pseudospin polarizations at different corners owing to nontrivial higher-order topology and finite spin-spin coupling. By applying the spin-polarized excitation sources, we can selectively excite the corner states at different spatial positions through spin-momentum-locked decaying edge states, resembling the quantum spin Hall effect in a higher-order manner. Our work which breaks the barriers between the spin photonics and higher-order topology opens the frontiers for studying lower-dimensional spinful classical surface waves and supports explorations in robust communications.

cond-mat.mtrl-sci↗

arXiv2020

Terahertz metallic photonic crystals integrated with dielectric waveguides

Compact and low-loss photonic crystal waveguides are critical in integrated terahertz (THz) applications. Compared with pure metal or dielectric photonic crystal waveguides, hybrid (metal-dielectric) integrated waveguides provide a simple way to further improve the field confinement and the propagation loss. In this work, we investigate a hybrid waveguide consisting of metallic photonic crystals and dielectric films in 0.1-1.0 THz. Photonic crystal waveguides based on metal pillar arrays (MPAs) support two resonance modes including the fundamental and high-order transverse magnetic (TM) modes and then form one apparent bandgap in 0.45-0.55 THz. The high-order TM-mode shows higher confinement than the fundamental mode and are thus sensitive to the dielectric film on the MPAs. The propagation loss and field confinement can be optimized by changing the dielectric film thickness and refractive index. The investigation shows that the lowest loss occurs at 0.68 THz because the high-order TM-mode THz waves are tightly confined inside the hybrid waveguide. This work proves that such hybrid waveguides based on metallic photonic crystals are promising to develop as a compact integrated terahertz device.

physics.optics↗

arXiv2020

Segmentation-based Method combined with Dynamic Programming for Brain Midline Delineation

The midline related pathological image features are crucial for evaluating the severity of brain compression caused by stroke or traumatic brain injury (TBI). The automated midline delineation not only improves the assessment and clinical decision making for patients with stroke symptoms or head trauma but also reduces the time of diagnosis. Nevertheless, most of the previous methods model the midline by localizing the anatomical points, which are hard to detect or even missing in severe cases. In this paper, we formulate the brain midline delineation as a segmentation task and propose a three-stage framework. The proposed framework firstly aligns an input CT image into the standard space. Then, the aligned image is processed by a midline detection network (MD-Net) integrated with the CoordConv Layer and Cascade AtrousCconv Module to obtain the probability map. Finally, we formulate the optimal midline selection as a pathfinding problem to solve the problem of the discontinuity of midline delineation. Experimental results show that our proposed framework can achieve superior performance on one in-house dataset and one public dataset.

eess.IV↗