arXiv subjects

Ye Liu

Publications and source records attributed to Ye Liu.

At least 181 records · Page 10Linked to original sources

arXiv2020

Interpretable Multi-Step Reasoning with Knowledge Extraction on Complex Healthcare Question Answering

Healthcare question answering assistance aims to provide customer healthcare information, which widely appears in both Web and mobile Internet. The questions usually require the assistance to have proficient healthcare background knowledge as well as the reasoning ability on the knowledge. Recently a challenge involving complex healthcare reasoning, HeadQA dataset, has been proposed, which contains multiple-choice questions authorized for the public healthcare specialization exam. Unlike most other QA tasks that focus on linguistic understanding, HeadQA requires deeper reasoning involving not only knowledge extraction, but also complex reasoning with healthcare knowledge. These questions are the most challenging for current QA systems, and the current performance of the state-of-the-art method is slightly better than a random guess. In order to solve this challenging task, we present a Multi-step reasoning with Knowledge extraction framework (MurKe). The proposed framework first extracts the healthcare knowledge as supporting documents from the large corpus. In order to find the reasoning chain and choose the correct answer, MurKe iterates between selecting the supporting documents, reformulating the query representation using the supporting documents and getting entailment score for each choice using the entailment model. The reformulation module leverages selected documents for missing evidence, which maintains interpretability. Moreover, we are striving to make full use of off-the-shelf pre-trained models. With less trainable weight, the pre-trained model can easily adapt to healthcare tasks with limited training samples. From the experimental results and ablation study, our system is able to outperform several strong baselines on the HeadQA dataset.

cs.AI↗

arXiv2020

Revisiting Convolutional Neural Networks for Citywide Crowd Flow Analytics

Citywide crowd flow analytics is of great importance to smart city efforts. It aims to model the crowd flow (e.g., inflow and outflow) of each region in a city based on historical observations. Nowadays, Convolutional Neural Networks (CNNs) have been widely adopted in raster-based crowd flow analytics by virtue of their capability in capturing spatial dependencies. After revisiting CNN-based methods for different analytics tasks, we expose two common critical drawbacks in the existing uses: 1) inefficiency in learning global spatial dependencies, and 2) overlooking latent region functions. To tackle these challenges, in this paper we present a novel framework entitled DeepLGR that can be easily generalized to address various citywide crowd flow analytics problems. This framework consists of three parts: 1) a local feature extraction module to learn representations for each region; 2) a global context module to extract global contextual priors and upsample them to generate the global features; and 3) a region-specific predictor based on tensor decomposition to provide customized predictions for each region, which is very parameter-efficient compared to previous methods. Extensive experiments on two typical crowd flow analytics tasks demonstrate the effectiveness, stability, and generality of our framework.

cs.CV↗

arXiv2020

The Impact of the Physical Layer on the Performance of Concurrent Transmissions

The popularity of concurrent transmissions (CT) has soared after recent studies have shown their feasibility on the four physical layers specified by BLE 5, hence providing an alternative to the use of IEEE 802.15.4 for the design of reliable and efficient low-power wireless protocols. However, to date, the extent to which physical layer properties affect the performance of CT has not yet been investigated in detail. This paper fills this gap and provides the first extensive study on the impact of the physical layer on CT-based solutions using IEEE 802.15.4 and BLE 5. We first highlight through simulation how the impact of errors induced by de-synchronization and beating on the performance of CT highly depends on the choice of the underlying physical layer. We then confirm these observations experimentally on real hardware through an analysis of the bit error distribution across received packets, unveiling possible techniques to effectively handle these errors. We further study the performance of CT-based flooding protocols in the presence of radio interference on a large-scale, and derive important insights on how the used physical layer affects their dependability.

cs.NI↗

arXiv2020

Optimal Emergency Frequency Control Based on Coordinated Droop in Multi-Infeed Hybrid AC-DC System

In multi-infeed hybrid AC-DC (MIDC) systems, the asynchronous interconnection between regional grids, the complicated system dynamics and possible cascading failures have an enormous effect on the frequency stability. In order to deal with the frequency instability problems in emergency situations, this paper proposes a decentralized emergency frequency control strategy based on coordinated droop for the MIDC system. First, a P-f droop control for LCC-HVDC systems is introduced and the coordinated droop mechanism among LCC-HVDC systems and generators is designed. Then, to reasonably allocate the power imbalance among LCC-HVDC systems and generators, an optimal emergency frequency control (OEFC) problem is formulated, and the optimal droop coefficients are selected in a decentralized approach, which can deal with various control objectives. A Lyapunov stability analysis shows that the closed-loop equilibrium is locally asymptotically stable considering the LCC-HVDC dynamics. The effectiveness of the proposed emergency control strategy is verified through simulations.

eess.SY↗

arXiv2020

Commonsense Evidence Generation and Injection in Reading Comprehension

Human tackle reading comprehension not only based on the given context itself but often rely on the commonsense beyond. To empower the machine with commonsense reasoning, in this paper, we propose a Commonsense Evidence Generation and Injection framework in reading comprehension, named CEGI. The framework injects two kinds of auxiliary commonsense evidence into comprehensive reading to equip the machine with the ability of rational thinking. Specifically, we build two evidence generators: the first generator aims to generate textual evidence via a language model; the other generator aims to extract factual evidence (automatically aligned text-triples) from a commonsense knowledge graph after graph completion. Those evidences incorporate contextual commonsense and serve as the additional inputs to the model. Thereafter, we propose a deep contextual encoder to extract semantic relationships among the paragraph, question, option, and evidence. Finally, we employ a capsule network to extract different linguistic units (word and phrase) from the relations, and dynamically predict the optimal option based on the extracted units. Experiments on the CosmosQA dataset demonstrate that the proposed CEGI model outperforms the current state-of-the-art approaches and achieves the accuracy (83.6%) on the leaderboard.

cs.AI↗

arXiv2020

Fine-Grained Urban Flow Inference

The ubiquitous deployment of monitoring devices in urban flow monitoring systems induces a significant cost for maintenance and operation. A technique is required to reduce the number of deployed devices, while preventing the degeneration of data accuracy and granularity. In this paper, we present an approach for inferring the real-time and fine-grained crowd flows throughout a city based on coarse-grained observations. This task exhibits two challenges: the spatial correlations between coarse- and fine-grained urban flows, and the complexities of external impacts. To tackle these issues, we develop a model entitled UrbanFM which consists of two major parts: 1) an inference network to generate fine-grained flow distributions from coarse-grained inputs that uses a feature extraction module and a novel distributional upsampling module; 2) a general fusion subnet to further boost the performance by considering the influence of different external factors. This structure provides outstanding effectiveness and efficiency for small scale upsampling. However, the single-pass upsampling used by UrbanFM is insufficient at higher upscaling rates. Therefore, we further present UrbanPy, a cascading model for progressive inference of fine-grained urban flows by decomposing the original tasks into multiple subtasks. Compared to UrbanFM, such an enhanced structure demonstrates favorable performance for larger-scale inference tasks.

cs.CV↗

arXiv2020

A General Initialization Scheme for Electromagnetic Transient Simulation: Towards Large-Scale Hybrid AC-DC Grids

With the large-scale hybrid AC-DC grids coming into being, electromagnetic transient (EMT) simulation is required to accurately describe the dynamics of systems. However, the EMT steady-state initialization for hybrid AC-DC system is difficult and time-consuming when the system scale is huge. In order to provide a stable snapshot for EMT simulation with nonlinear components and black-box components, this paper proposes a general initialization scheme for EMT simulation (EMT-GIS) which can be implemented in the electromagnetic transient program (EMTP)-type simulators. First, an integrated power flow (IPF) algorithm is introduced to provide the steady-state results. Then, an initialized snapshot calculation-and-splicing mechanism is designed for EMT-GIS. The proposed EMT-GIS is tested using a hybrid AC-DC system in China on the CloudPSS simulation platform. Test results verify the effectiveness of the proposed EMT-GIS.

eess.SY↗

arXiv2019

Bas-relief Generation from Point Clouds Based on Normal Space Compression with Real-time Adjustment on CPU

Bas-relief generation based on 3d models is a hot topic in computer graphics. State-of-the-art algorithms take a mesh surface as input, but real-time interaction via CPU cannot be realized. In this paper, a bas-relief generation algorithm that takes a scattered point cloud as input is proposed. The algorithm takes normal vectors as the operation object and the variation of the local surface as the compression criterion. By constructing and solving linear equations of bas-relief vertices, the closed-form solution can be obtained. Since there is no need to compute discrete gradients on a point cloud lacking topology information, it is easier to implement and more intuitive than gradient domain methods. The algorithm provides parameters to adjust the bas-relief height, saturation and detail richness. At the same time, through the solution strategy based on the subspace, it realizes the real-time adjustment of the bas-relief effect based on the computing power of a consumer CPU. In addition, an iterative solution to generate a bas-relief model of a specified height is presented to meet specific application requirements. Experiments show that our algorithm provides a unified solution for various types of bas-relief creation and can generate bas-reliefs with good saturation and rich details.

cs.GR↗

arXiv2019

Orthogonal Nonnegative Tucker Decomposition

In this paper, we study the nonnegative tensor data and propose an orthogonal nonnegative Tucker decomposition (ONTD). We discuss some properties of ONTD and develop a convex relaxation algorithm of the augmented Lagrangian function to solve the optimization problem. The convergence of the algorithm is given. We employ ONTD on the image data sets from the real world applications including face recognition, image representation, hyperspectral unmixing. Numerical results are shown to illustrate the effectiveness of the proposed algorithm.

stat.ML↗

arXiv2019

Point cloud ridge-valley feature enhancement based on position and normal guidance

Ridge-valley features are important elements of point clouds, as they contain rich surface information. To recognize these features from point clouds, this paper introduces an extreme point distance (EPD) criterion with scale independence. Compared with traditional methods, the EPD greatly reduces the number of potential feature points and improves the robustness of multiscale feature point recognition. On this basis, a feature enhancement algorithm based on user priori guidance is proposed that adjusts the coordinates of the feature area by solving an objective equation containing the expected position and normal constraints. Since the expected normal can be expressed as a function of neighborhood point coordinates, the above objective equation can be converted into linear sparse equations with enhanced feature positions as variables, and thus, the closed solution can be obtained. In addition, a parameterization method for scattered point clouds based on feature line guidance is proposed, which reduces the number of unknowns by 2/3 and eliminates lateral sliding in the direction perpendicular to feature lines. Finally, the application of the algorithm in multiscale ridge-valley feature recognition, freeform surface feature enhancement and computer-aided design (CAD) workpiece sharp feature restoration verifies its effectiveness.

cs.GR↗

arXiv2019

Generative Question Refinement with Deep Reinforcement Learning in Retrieval-based QA System

In real-world question-answering (QA) systems, ill-formed questions, such as wrong words, ill word order, and noisy expressions, are common and may prevent the QA systems from understanding and answering them accurately. In order to eliminate the effect of ill-formed questions, we approach the question refinement task and propose a unified model, QREFINE, to refine the ill-formed questions to well-formed question. The basic idea is to learn a Seq2Seq model to generate a new question from the original one. To improve the quality and retrieval performance of the generated questions, we make two major improvements: 1) To better encode the semantics of ill-formed questions, we enrich the representation of questions with character embedding and the recent proposed contextual word embedding such as BERT, besides the traditional context-free word embeddings; 2) To make it capable to generate desired questions, we train the model with deep reinforcement learning techniques that considers an appropriate wording of the generation as an immediate reward and the correlation between generated question and answer as time-delayed long-term rewards. Experimental results on real-world datasets show that the proposed QREFINE method can generate refined questions with more readability but fewer mistakes than the original questions provided by users. Moreover, the refined questions also significantly improve the accuracy of answer retrieval.

cs.IR↗

arXiv2019

Topology of the icosidodecahedral arrangement

The icosidodecahedral arrangement is introduced by M. Yoshinaga (arXiv:1902.06256) as the first known example that is a hyperplane arrangement whose Milnor fiber has torsions in first integral homology. In this note, we prove that the icosidodecahedral arrangement is $K(π,1)$, hence so is its Milnor fiber.

math.GT↗

arXiv2019

Emergency DC Power Support Strategy Based on Coordinated Droop Control in Multi-Infeed HVDC System

With the complex hybrid AC-DC power system in China coming into being, the HVDC faults, such as DC block faults, have an enormous effect on the frequency stability of the AC side. In multi-infeed HVDC (MIDC) system, to improve the frequency stability of the receiving-end system, this paper proposes an emergency DC power support (EDCPS) strategy, which is based on a designed droop characteristic between the active power of LCC-HVDC lines and the receiving-end system frequency. Then, a coordinated optimization method is proposed to determine the droop coefficients of the MIDC lines. To test the proposed EDCPS method, the electromagnetic transient (EMT) model of a MIDC system is built on the CloudPSS platform, and the test results verify the effectiveness of the proposed EDCPS strategy.

eess.SY↗

arXiv2019

Tucker Decomposition Network: Expressive Power and Comparison

Deep neural networks have achieved a great success in solving many machine learning and computer vision problems. The main contribution of this paper is to develop a deep network based on Tucker tensor decomposition, and analyze its expressive power. It is shown that the expressiveness of Tucker network is more powerful than that of shallow network. In general, it is required to use an exponential number of nodes in a shallow network in order to represent a Tucker network. Experimental results are also given to compare the performance of the proposed Tucker network with hierarchical tensor network and shallow network, and demonstrate the usefulness of Tucker network in image classification problems.

cs.LG↗

arXiv2019

Urban greenery and mental wellbeing in adults: Cross-sectional mediation analyses on multiple pathways across different greenery measures

Multiple mechanisms have been proposed to explain how greenery enhances their mental wellbeing. Mediation studies, however, focus on a limited number of mechanisms and rely on remotely sensed greenery measures, which do not accurately capture how neighborhood greenery is perceived on the ground. To examine: 1) how streetscape and remote sensing-based greenery affect people's mental wellbeing in Guangzhou, China; 2) whether and, if so, to what extent the associations are mediated by physical activity, stress, air quality and noise, and social cohesion; and 3) whether differences in the mediation across the streetscape greenery and NDVI exposure metrics occurred. Mental wellbeing was quantified by the WHO-5 wellbeing index. Greenery measures were extracted at the neighborhood level: 1) streetscape greenery from street view data via a convolutional neural network, and 2) the NDVI remote sensing images. Single and multiple mediation analyses with multilevel regressions were conducted. Streetscape and NDVI greenery were weakly and positively, but not significantly, correlated. Our regression results revealed that streetscape greenery and NDVI were, individually and jointly, positively associated with mental wellbeing. Significant partial mediators for the streetscape greenery were physical activity, stress, air quality and noise, and social cohesion; together, they explained 62% of the association. For NDVI, only physical activity and social cohesion were significant partial mediators, accounting for 22% of the association. Mental health and wellbeing and both streetscape and satellite-derived greenery seem to be both directly correlated and indirectly mediated. Our findings signify that both greenery measures capture different aspects of natural environments and may contribute to people's wellbeing by means of different mechanisms.

stat.AP↗

arXiv2019

MMKG: Multi-Modal Knowledge Graphs

We present MMKG, a collection of three knowledge graphs that contain both numerical features and (links to) images for all entities as well as entity alignments between pairs of KGs. Therefore, multi-relational link prediction and entity matching communities can benefit from this resource. We believe this data set has the potential to facilitate the development of novel multi-modal learning approaches for knowledge graphs.We validate the utility ofMMKG in the sameAs link prediction task with an extensive set of experiments. These experiments show that the task at hand benefits from learning of multiple feature types.

cs.AI↗

arXiv2019

UrbanFM: Inferring Fine-Grained Urban Flows

Urban flow monitoring systems play important roles in smart city efforts around the world. However, the ubiquitous deployment of monitoring devices, such as CCTVs, induces a long-lasting and enormous cost for maintenance and operation. This suggests the need for a technology that can reduce the number of deployed devices, while preventing the degeneration of data accuracy and granularity. In this paper, we aim to infer the real-time and fine-grained crowd flows throughout a city based on coarse-grained observations. This task is challenging due to two reasons: the spatial correlations between coarse- and fine-grained urban flows, and the complexities of external impacts. To tackle these issues, we develop a method entitled UrbanFM based on deep neural networks. Our model consists of two major parts: 1) an inference network to generate fine-grained flow distributions from coarse-grained inputs by using a feature extraction module and a novel distributional upsampling module; 2) a general fusion subnet to further boost the performance by considering the influences of different external factors. Extensive experiments on two real-world datasets, namely TaxiBJ and HappyValley, validate the effectiveness and efficiency of our method compared to seven baselines, demonstrating the state-of-the-art performance of our approach on the fine-grained urban flow inference problem.

cs.CV↗

arXiv2019

Automatic trajectory measurement of large numbers of crowded objects

Complex motion patterns of natural systems, such as fish schools, bird flocks, and cell groups, have attracted great attention from scientists for years. Trajectory measurement of individuals is vital for quantitative and high-throughput study of their collective behaviors. However, such data are rare mainly due to the challenges of detection and tracking of large numbers of objects with similar visual features and frequent occlusions. We present an automatic and effective framework to measure trajectories of large numbers of crowded oval-shaped objects, such as fish and cells. We first use a novel dual ellipse locator to detect the coarse position of each individual and then propose a variance minimization active contour method to obtain the optimal segmentation results. For tracking, cost matrix of assignment between consecutive frames is trainable via a random forest classifier with many spatial, texture, and shape features. The optimal trajectories are found for the whole image sequence by solving two linear assignment problems. We evaluate the proposed method on many challenging data sets.

cs.CV↗