Browsing Preprints by Title
Now showing items 177195 of 195

Tagging like Humans: Diverse and Distinct Image Annotation(arXiv, 20180331)In this work we propose a new automatic image annotation model, dubbed {\bf diverse and distinct image annotation} (D2IA). The generative model D2IA is inspired by the ensemble of human annotations, which create semantically relevant, yet distinct and diverse tags. In D2IA, we generate a relevant and distinct tag subset, in which the tags are relevant to the image contents and semantically distinct to each other, using sequential sampling from a determinantal point process (DPP) model. Multiple such tag subsets that cover diverse semantic aspects or diverse semantic levels of the image contents are generated by randomly perturbing the DPP sampling process. We leverage a generative adversarial network (GAN) model to train D2IA. Extensive experiments including quantitative and qualitative comparisons, as well as human subject studies, on two benchmark datasets demonstrate that the proposed model can produce more diverse and distinct tags than the stateofthearts.

Teaching UAVs to Race With Observational Imitation Learning(arXiv, 20180303)Recent work has tackled the problem of autonomous navigation by imitating a teacher and learning an endtoend policy, which directly predicts controls from raw images. However, these approaches tend to be sensitive to mistakes by the teacher and do not scale well to other environments or vehicles. To this end, we propose a modular network architecture that decouples perception from control, and is trained using Observational Imitation Learning (OIL), a novel imitation learning variant that supports online training and automatic selection of optimal behavior from observing multiple teachers. We apply our proposed methodology to the challenging problem of unmanned aerial vehicle (UAV) racing. We develop a simulator that enables the generation of large amounts of synthetic training data (both UAV captured images and its controls) and also allows for online learning and evaluation. We train a perception network to predict waypoints from raw image data and a control network to predict UAV controls from these waypoints using OIL. Our modular network is able to autonomously fly a UAV through challenging race tracks at high speeds. Extensive experiments demonstrate that our trained network outperforms its teachers, endtoend baselines, and even human pilots in simulation. The supplementary video can be viewed at https://youtu.be/PeTXSoriflc

Teaching UAVs to Race: EndtoEnd Regression of Agile Controls in Simulation(arXiv, 20181122)Automating the navigation of unmanned aerial vehicles (UAVs) in diverse scenarios has gained much attention in recent years. However, teaching UAVs to fly in challenging environments remains an unsolved problem, mainly due to the lack of training data. In this paper, we train a deep neural network to predict UAV controls from raw image data for the task of autonomous UAV racing in a photorealistic simulation. Training is done through imitation learning with data augmentation to allow for the correction of navigation mistakes. Extensive experiments demonstrate that our trained network (when sufficient data augmentation is used) outperforms stateoftheart methods and flies more consistently than many human pilots. Additionally, we show that our optimized network architecture can run in realtime on embedded hardware, allowing for efficient onboard processing critical for realworld deployment.

TGFb2, catalase activity, H2O2 output and metastatic potential of diverse types of tumour(Cold Spring Harbor Laboratory, 20181114)Theileria annulata is a protozoan parasite that infects and transforms bovine macrophages causing a myeloidleukaemialike disease called tropical theileriosis. TGFb2 is highly expressed in many cancer cells and is significantly increased in Theileriatransformed macrophages, as are levels of Reactive Oxygen Species (ROS), notably H2O2. Here, we describe the interplay between TGFb2 and ROS in cellular transformation. We show that TGFb2 drives expression of catalase to reduce the amount of H2O2 produced by T. annulatatransformed bovine macrophages, as well as by human lung (A549) and colon cancer (HT29) cell lines. Theileriatransformed macrophages attenuated for dissemination express less catalase and produce more H2O2, but regain both virulent migratory and matrigel traversal phenotypes when stimulated with TGFb2, or catalase that reduce H2O2 output. Increased H2O2 output therefore, underpins the aggressive dissemination phenotype of diverse tumour cell types, but in contrast, too much H2O2 can dampen dissemination.

Theory of Topological Spin Hall Effect in Antiferromagnetic Skyrmion: Impact on Currentinduced Motion(arXiv, 20170909)We demonstrate that the nontrivial magnetic texture of antiferromagnetic skyrmions (AFMSks) promotes a nonvanishing topological spin Hall effect (TSHE) on the flowing electrons. This results in a substantial enhancement of the nonadiabatic torque and hence improves the skyrmion mobility. This nonadiabatic torque increases when decreasing the skyrmion size, and therefore scaling down results in a much higher torque efficiency. In clean AFMSks, we find a significant boost of the TSHE close to van Hove singularity. Interestingly, this effect is enhanced away from the band gap in the presence of nonmagnetic interstitial defects. Furthermore, unlike their ferromagnetic counterpart, TSHE in AFMSks increases with increase in disorder strength thus opening promising avenues for materials engineering of this effect.

Thermodynamically consistent modeling and simulation of multicomponent twophase flow model with partial miscibility(arXiv, 20161125)A general diffuse interface model with a realistic equation of state (e.g. PengRobinson equation of state) is proposed to describe the multicomponent twophase fluid flow based on the principles of the NVTbased framework which is a latest alternative over the NPTbased framework to model the realistic fluids. The proposed model uses the Helmholtz free energy rather than Gibbs free energy in the NPTbased framework. Different from the classical routines, we combine the first law of thermodynamics and related thermodynamical relations to derive the entropy balance equation, and then we derive a transport equation of the Helmholtz free energy density. Furthermore, by using the second law of thermodynamics, we derive a set of unified equations for both interfaces and bulk phases that can describe the partial miscibility of two fluids. A relation between the pressure gradient and chemical potential gradients is established, and this relation leads to a new formulation of the momentum balance equation, which demonstrates that chemical potential gradients become the primary driving force of fluid motion. Moreover, we prove that the proposed model satisfies the total (free) energy dissipation with time. For numerical simulation of the proposed model, the key difficulties result from the strong nonlinearity of Helmholtz free energy density and tight coupling relations between molar densities and velocity. To resolve these problems, we propose a novel convexconcave splitting of Helmholtz free energy density and deal well with the coupling relations between molar densities and velocity through very careful physical observations with a mathematical rigor. We prove that the proposed numerical scheme can preserve the discrete (free) energy dissipation. Numerical tests are carried out to verify the effectiveness of the proposed method.

Thermodynamically consistent simulation of nonisothermal diffuseinterface twophase flow with PengRobinson equation of state(arXiv, 20171206)In this paper, we consider a diffuseinterface gasliquid twophase flow model with inhomogeneous temperatures, in which we employ the PengRobinson equation of state and the temperaturedependent influence parameter instead of the van der Waals equation of state and the constant influence parameter used in the existing models. As a result, our model can characterize accurately the physical behaviors of numerous realistic gasliquid fluids, especially hydrocarbons. Furthermore, we prove a relation associating the pressure gradient with the gradients of temperature and chemical potential, and thereby derive a new formulation of the momentum balance equation, which shows that gradients of the chemical potential and temperature become the primary driving force of the fluid motion. It is rigorously proved that the new formulations of the model obey the first and second laws of thermodynamics. To design efficient numerical methods, we prove that Helmholtz free energy density is a concave function with respect to the temperature under certain physical conditions. Based on the proposed modeling formulations and the convexconcave splitting of Helmholtz free energy density, we propose a novel thermodynamically stable numerical scheme. We rigorously prove that the proposed method satisfies the first and second laws of thermodynamics. Finally, numerical tests are carried out to verify the effectiveness of the proposed simulation method.

Total Variation Depth for Functional Data(arXiv, 20161115)There has been extensive work on data depthbased methods for robust multivariate data analysis. Recent developments have moved to infinitedimensional objects such as functional data. In this work, we propose a new notion of depth, the total variation depth, for functional data. As a measure of depth, its properties are studied theoretically, and the associated outlier detection performance is investigated through simulations. Compared to magnitude outliers, shape outliers are often masked among the rest of samples and harder to identify. We show that the proposed total variation depth has many desirable features and is well suited for outlier detection. In particular, we propose to decompose the total variation depth into two components that are associated with shape and magnitude outlyingness, respectively. This decomposition allows us to develop an effective procedure for outlier detection and useful visualization tools, while naturally accounting for the correlation in functional data. Finally, the proposed methodology is demonstrated using real datasets of curves, images, and video frames.

UE4Sim: A PhotoRealistic Simulator for Computer Vision Applications(arXiv, 20170819)We present a photorealistic training and evaluation simulator (UE4Sim) with extensive applications across various fields of computer vision. Built on top of the Unreal Engine, the simulator integrates full featured physics based cars, unmanned aerial vehicles (UAVs), and animated human actors in diverse urban and suburban 3D environments. We demonstrate the versatility of the simulator with two case studies: autonomous UAVbased tracking of moving objects and autonomous driving using supervised learning. The simulator fully integrates both several stateoftheart tracking algorithms with a benchmark evaluation tool and a deep neural network (DNN) architecture for training vehicles to drive autonomously. It generates synthetic photorealistic datasets with automatic ground truth annotations to easily extend existing realworld datasets and provides extensive synthetic data variety through its ability to reconfigure synthetic worlds on the fly using an automatic world generation tool.

The ultrasensitive Nodewalk technique identifies stochastic from virtual, populationbased enhancer hubs regulating MYC in 3D: Implications for the fitness of cancer cells(Cold Spring Harbor Laboratory, 20180327)The relationship between stochastic transcriptional bursts and dynamic 3D chromatin states is not well understood due to poor sensitivity and/or resolution of current chromatin structurebased assays. Consequently, it is not well established if enhancers operate individually and/or in clusters to coordinate gene transcription. In the current study, we introduce Nodewalk, which uniquely combines high sensitivity with high resolution to enable the analysis of chromatin networks in minute input material. The >10,000fold increase in sensitivity over other manytoall competing methods uncovered that active chromatin hubs identified in large input material, corresponding to 10 000 cells, flanking the MYC locus are primarily virtual. Thus, the close agreement between chromatin interactomes generated from aliquots corresponding to less than 10 cells with randomly resampled interactomes, we find that numerous distal enhancers positioned within flanking topologically associating domains (TADs) converge on MYC in largely mutually exclusive manners. Moreover, when comparing with several enhancer baits, the assignment of the MYC locus as the node with the highest dynamic importance index, indicates that it is MYC targeting its enhancers, rather than vice versa. Dynamic changes in the configuration of the boundary between TADs flanking MYC underlie numerous stochastic encounters with a diverse set of enhancers to depict the plasticity of its transcriptional regulation. Such an arrangement might increase the fitness of the cancer cell by increasing the probability of MYC transcription in response to a wide range of environmental cues encountered by the cell during the neoplastic process.

Underwater Optical Wireless Communications, Networking, and Localization: A Survey(arXiv, 20180228)Underwater wireless communications can be carried out through acoustic, radio frequency (RF), and optical waves. Compared to its bandwidth limited acoustic and RF counterparts, underwater optical wireless communications (UOWCs) can support higher data rates at low latency levels. However, severe aquatic channel conditions (e.g., absorption, scattering, turbulence, etc.) pose great challenges for UOWCs and significantly reduce the attainable communication ranges, which necessitates efficient networking and localization solutions. Therefore, we provide a comprehensive survey on the challenges, advances, and prospects of underwater optical wireless networks (UOWNs) from a layer by layer perspective which includes: 1) Potential network architectures; 2) Physical layer issues including propagation characteristics, channel modeling, and modulation techniques 3) Data link layer problems covering link configurations, link budgets, performance metrics, and multiple access schemes; 4) Network layer topics containing relaying techniques and potential routing algorithms; 5) Transport layer subjects such as connectivity, reliability, flow and congestion control; 6) Application layer goals and stateoftheart UOWN applications, and 7) Localization and its impacts on UOWN layers. Finally, we outline the open research challenges and point out the future directions for underwater optical wireless communications, networking, and localization research.

Unemployment estimation: Spatial point referenced methods and models(arXiv, 20170626)Portuguese Labor force survey, from 4th quarter of 2014 onwards, started georeferencing the sampling units, namely the dwellings in which the surveys are carried. This opens new possibilities in analysing and estimating unemployment and its spatial distribution across any region. The labor force survey choose, according to an preestablished sampling criteria, a certain number of dwellings across the nation and survey the number of unemployed in these dwellings. Based on this survey, the National Statistical Institute of Portugal presently uses direct estimation methods to estimate the national unemployment figures. Recently, there has been increased interest in estimating these figures in smaller areas. Direct estimation methods, due to reduced sampling sizes in small areas, tend to produce fairly large sampling variations therefore model based methods, which tend to

Unidirectional MagnonDriven Domain Wall Motion due to Interfacial DzyaloshinskiiMoriya Interaction(arXiv, 20180328)We theoretically study magnondriven motion of a tranverse domain wall in the presence of interfacial DzyaloshinskiiMoriya interaction (DMI). Contrary to previous studies, the domain wall moves along the same direction regardless of the magnonflow direction. Our symmetry analysis reveals that the odd order DMI contributions to the domain wall velocity are independent of the magnonflow direction. Corresponding DMIinduced asymmetric transitions from a spinwave state to another give rise to a large momentum transfer to the domain wall without nonreciprocity and much reflection. This counterintuitive unidirectional motion occurs not only for a spin wave with a single wavevector but also for thermal magnons with distributed wavevectors.

Using MultiSpectral UAV Imagery to Extract Tree Crop Structural Properties and Assess Pruning Effects(MDPI AG, 20180418)Unmanned aerial vehicles (UAV) provide an unprecedented capacity to monitor the development and dynamics of tree growth and structure through time. It is generally thought that the pruning of tree crops encourages new growth, has a positive effect on fruiting, makes fruitpicking easier, and may increase yield, as it increases light interception and tree crown surface area. To establish the response of pruning in an orchard of lychee trees, an assessment of changes in tree structure, i.e. tree crown perimeter, width, height, area and Plant Projective Cover (PPC), was undertaken using multispectral UAV imagery collected before and after a pruning event. While tree crown perimeter, width and area could be derived directly from the delineated tree crowns, height was estimated from a produced canopy height model and PPC was most accurately predicted based on the NIR band. Pre and postpruning results showed significant differences in all measured tree structural parameters, including an average decrease in tree crown perimeter of 1.94 m, tree crown width of 0.57 m, tree crown height of 0.62 m, tree crown area of 3.5 m2, and PPC of 14.8%. In order to provide guidance on data collection protocols for orchard management, the impact of flying height variations was also examined, offering some insight into the influence of scale and the scalability of this UAV based approach for larger orchards. The different flying heights (i.e. 30, 50 and 70 m) produced similar measurements of tree crown width and PPC, while tree crown perimeter, area and height measurements decreased with increasing flying height. Overall, these results illustrate that routine collection of multispectral UAV imagery can provide a means of assessing pruning effects on changes in tree structure in commercial orchards, and highlight the importance of collecting imagery with consistent flight configurations, as varying flying heights may cause changes to tree structural measurements.

Vec2SPARQL: integrating SPARQL queries and knowledge graph embeddings(Cold Spring Harbor Laboratory, 20181108)Recent developments in machine learning have lead to a rise of large number of methods for extracting features from structured data. The features are represented as a vectors and may encode for some semantic aspects of data. They can be used in a machine learning models for different tasks or to compute similarities between the entities of the data. SPARQL is a query language for structured data originally developed for querying Resource Description Framework (RDF) data. It has been in use for over a decade as a standardized NoSQL query language. Many different tools have been developed to enable data sharing with SPARQL. For example, SPARQL endpoints make your data interoperable and available to the world. SPARQL queries can be executed across multiple endpoints. We have developed a Vec2SPARQL, which is a general framework for integrating structured data and their vector space representations. Vec2SPARQL allows jointly querying vector functions such as computing similarities (cosine, correlations) or classifications with machine learning models within a single SPARQL query. We demonstrate applications of our approach for biomedical and clinical use cases. Our source code is freely available at https://github.com/bioontologyresearchgroup/vec2sparql and we make a Vec2SPARQL endpoint available at http://sparql.bio2vec.net/.

VQABQ: Visual Question Answering by Basic Questions(arXiv, 20170319)Taking an image and question as the input of our method, it can output the textbased answer of the query question about the given image, so called Visual Question Answering (VQA). There are two main modules in our algorithm. Given a natural language question about an image, the first module takes the question as input and then outputs the basic questions of the main given question. The second module takes the main question, image and these basic questions as input and then outputs the textbased answer of the main question. We formulate the basic questions generation problem as a LASSO optimization problem, and also propose a criterion about how to exploit these basic questions to help answer main question. Our method is evaluated on the challenging VQA dataset and yields stateoftheart accuracy, 60.34% in openended task.

Weakly intrusive lowrank approximation method for nonlinear parameterdependent equations(arXiv, 20170630)This paper presents a weakly intrusive strategy for computing a lowrank approximation of the solution of a system of nonlinear parameterdependent equations. The proposed strategy relies on a Newtonlike iterative solver which only requires evaluations of the residual of the parameterdependent equation and of a preconditioner (such as the differential of the residual) for instances of the parameters independently. The algorithm provides an approximation of the set of solutions associated with a possibly large number of instances of the parameters, with a computational complexity which can be orders of magnitude lower than when using the same Newtonlike solver for all instances of the parameters. The reduction of complexity requires efficient strategies for obtaining lowrank approximations of the residual, of the preconditioner, and of the increment at each iteration of the algorithm. For the approximation of the residual and the preconditioner, weakly intrusive variants of the empirical interpolation method are introduced, which require evaluations of entries of the residual and the preconditioner. Then, an approximation of the increment is obtained by using a greedy algorithm for lowrank approximation, and a lowrank approximation of the iterate is finally obtained by using a truncated singular value decomposition. When the preconditioner is the differential of the residual, the proposed algorithm is interpreted as an inexact Newton solver for which a detailed convergence analysis is provided. Numerical examples illustrate the efficiency of the method.

Weighted LowRank Approximation of Matrices and Background Modeling(arXiv, 20180415)We primarily study a special a weighted lowrank approximation of matrices and then apply it to solve the background modeling problem. We propose two algorithms for this purpose: one operates in the batch mode on the entire data and the other one operates in the batchincremental mode on the data and naturally captures more background variations and computationally more effective. Moreover, we propose a robust technique that learns the background frame indices from the data and does not require any training frames. We demonstrate through extensive experiments that by inserting a simple weight in the Frobenius norm, it can be made robust to the outliers similar to the $\ell_1$ norm. Our methods match or outperform several stateoftheart online and batch background modeling methods in virtually all quantitative and qualitative measures.

Wetting of Water on Graphene(arXiv, 20161128)The wetting properties of graphene have proven controversial and difficult to assess. The presence of a graphene layer on top of a substrate does not significantly change the wetting properties of the solid substrate, suggesting that a single graphene layer does not affect the adhesion between the wetting phase and the substrate. However, wetting experiments of water on graphene show contact angles that imply a large amount of adhesion. Here, we investigate the wetting of graphene by measuring the mass of water vapor adsorbing to graphene flakes of different thickness at different relative humidities. Our experiments unambiguously show that the thinnest of graphene flakes do not adsorb water, from which it follows that the contact angle of water on these flakes is ~180o. Thicker flakes of graphene nanopowder, on the other hand, do adsorb water. A calculation of the van der Waals (vdW) interactions that dominate the adsorption in this system confirms that the adhesive interactions between a single atomic layer of graphene and water are so weak that graphene is superhydrophobic. The observations are confirmed in an independent experiment on graphenecoated water droplets that shows that it is impossible to make liquid 'marbles' with molecularly thin graphene.