Search

Based on the first-order Taylor expansion, an efficient Rigorous Coupled-Wave Analysis (RCWA) for one-dimensional ultrathin periodic structures is proposed in this paper. The derivation of the ultrathin form RCWA method is completed by using the first-order Taylor expansion to rearrange the matrix in the equations of boundary conditions. Then, the reliability of the proposed algorithm is verified by two examples. Finally, it is concluded that the proposed algorithm can reduce the CPU time of TE polarization and TM polarization by more than 50%. Meanwhile, compared with the conventional algorithm, the proposed algorithm also needs less memory.

Under the light of increasing air traffic congestion and reduced environmental impacts, the growing need for tailored approaches has arisen. To coordinate tailored approaches, accurate predictions of descent trajectories must be available. The ability of ground based automation for such endeavours however must be improved. Specifically in the ability to predict where descents begin, also referred to as the Top of descent (TOD). This paper attempts to provide a machine learning approach to TOD location prediction, considering the operational capabilities of modern-day air traffic management by working within the confines of parameters available to controllers and considerations of stakeholders. To identify the common TOD clusters and the resultant average TOD locations representing the controllers macroscopic perspective, an unsupervised learning approach is utilised. Specifically, an ensemble method of DBSCAN-K-means is used. This was compared to the more microscopic pilot controllers perspective and generic decision-making formulas to determine the proximity to expected TOD. Of which the microscopic option has proven to be more reliable in representing the individual Flight Management System (FMS) recommendations or pilots’ decisions on aircraft. In the prediction of TOD location, operationally available factors such as proximity to expected TOD, current flight level, approach heading, ground speed, fleet and aircraft type are used to train a decision tree model. A Singapore flight information region (FIR) case study is conducted to illustrate the methodology. Results of the methodology employed on the inbound A380 fleet consisting a total of 79 flights, yielded a maximum prediction range of 31NM range from actual TOD occurrence. This paper has demonstrates the possibility of using a machine learning approach to predict TOD of a flight considering both prior cruise characteristics and operational considerations of pilots. Accepted version

To meet the ever-growing demands of data throughput for forthcoming and deployed wireless networks, new wireless technologies like Long-Term Evolution License-Assisted Access (LTE-LAA) operate in shared and unlicensed bands. However, the LAA network must co-exist with incumbent IEEE 802.11 Wi-Fi systems. We consider a coexistence scenario where multiple LAA and Wi-Fi links share an unlicensed band. We aim to improve this coexistence by maximizing the key performance indicators (KPIs) of these networks simultaneously via dimension reduction and multi-criteria optimization. These KPIs are network throughputs as a function of medium access control protocols and physical layer parameters. We perform an exploratory analysis of coexistence behavior by approximating active subspaces to identify low-dimensional structure in the optimization criteria, i.e., few linear combinations of parameters for simultaneously maximizing KPIs. We leverage an aggregate low-dimensional subspace parametrized by approximated active subspaces of throughputs to facilitate multi-criteria optimization. The low-dimensional subspace approximations inform visualizations revealing convex KPIs over mixed active coordinates leading to an analytic Pareto trace of near-optimal solutions., 7 pages, 2 figures, 1 table, to appear in IEEE ICC 2021 proceedings

Recently, recommender systems that aim to suggest personalized lists of items for users to interact with online have drawn a lot of attention. In fact, many of these state-of-the-art techniques have been deep learning based. Recent studies have shown that these deep learning models (in particular for recommendation systems) are vulnerable to attacks, such as data poisoning, which generates users to promote a selected set of items. However, more recently, defense strategies have been developed to detect these generated users with fake profiles. Thus, advanced injection attacks of creating more `realistic' user profiles to promote a set of items is still a key challenge in the domain of deep learning based recommender systems. In this work, we present our framework CopyAttack, which is a reinforcement learning based black-box attack method that harnesses real users from a source domain by copying their profiles into the target domain with the goal of promoting a subset of items. CopyAttack is constructed to both efficiently and effectively learn policy gradient networks that first select, and then further refine/craft, user profiles from the source domain to ultimately copy into the target domain. CopyAttack's goal is to maximize the hit ratio of the targeted items in the Top-$k$ recommendation list of the users in the target domain. We have conducted experiments on two real-world datasets and have empirically verified the effectiveness of our proposed framework and furthermore performed a thorough model analysis., In Proceedings of 37th IEEE International Conference on Data Engineering (ICDE 2021, Full paper)

Since human observers are the ultimate receivers of an image, most of the image quality assessment (IQA) methods are based on analysis of the properties and mechanism of the human visual system. However, due to the lack of undistorted images for a reference, the accuracy of the no-reference IQA (NR-IQA) cannot compete with that of the full-reference IQA (FR-IQA). To bridge the performance gap between the FR-IQA and NR-IQA methods, we propose a NR-IQA method based on multi-task generative adversarial network, which attempts to restore dependable hallucinated images to compensate for the missing corresponding reference images. Two tasks, hallucination images and the quality maps are outputted by the generator and are combined with the specific loss to improve the reliability of hallucination images. Besides, two discriminator networks are used to respectively distinguish the undistorted images and hallucination images pairs, quality maps and structural similarity index measurement maps pairs. Finally, the hallucination images and distorted images are input into the IQA network, and quality scores are evaluated based on the differences between them. The superiority of our proposed method is verified by several different experiments on the LIVE datasets, TID2008 datasets, and TID2013 datasets.

The performance of dorsal hand vein image segmentation is limited due to low contrast and intensity inhomogeneity. In this paper, a novel method is proposed by integrating matched filter and local binary fitting level set model with the aim of overcoming the fault or incomplete segmentation in dorsal hand vein image. Following is the main work and contributions of this paper. First, 12-direction matched filters are adopted to enhance the vein patterns. Then, the local binary fitting level set model is introduced to segment the image enhanced by the first step. Third, a spurious vascular removal solution is presented to reduce the interference of metacarpal bones. 380 dorsal hand vein images collected from 69 subjects are used to evaluate the performance of the proposed algorithm. Compared with 5 existing vein segmentation methods, the proposed method achieves superior accuracy and shows great potential in image segmentation.

Graph Neural Networks (GNNs), which extend deep neural networks to graph-structured data, have attracted increasing attention. They have been proven to be powerful for numerous graph related tasks such as graph classification, link prediction, and node classification. To adapt GNNs to graph classification, recent works aim to learn graph-level representation through a hierarchical pooling procedure. One major direction is to select important nodes to hierarchically coarsen the input graph and gradually reduce the information into the graph representation. However, most of the existing methods only select important nodes, which can be redundant and cannot represent the original graph well. Meanwhile, the information of non-selected nodes is often overlooked when generating a new coarser graph, which may lead to the tremendous loss of important structural and node feature information. In this paper, we propose a novel pooling operator RepPool to learn hierarchical graph representations. Specifically, we introduce the concept of representativeness that is combined with the importance for node selection and we provide a learnable way to integrate non-selected nodes. By combining the RepPool operator with conventional GCN convolutional layers, a hierarchical graph classification architecture is developed. Extensive experiments on various public benchmarks have demonstrated the effectiveness of the proposed method. The implementation of the proposed framework is available11https://github.com/Juanhui28/RepPool/tree/master/RepPool.

In this paper, the generalized complex modified hybrid function projective synchronization (GCMHFPS) of time-delay complex chaotic systems is studied. In previous general GCMHFPS schemes, the parameters of system were selected as known. There are also some reports about GCMHFPS with unknown parameter. However, these chaotic systems with synchronous schemes are mostly of the same dimension and do not contain time-delay. However, time-delay is very common in life. By constructing suitable complex Lyapunov function and adaptive complex update laws, the sufficient conditions for GCMHFPS implementation are given, and the corresponding calculation formula is given, and new estimate of unknown complex parameters are given. The feasibility of the scheme is verified by a simulation example.

The growing vehicle connectivity and autonomy in the ground transportation system are not only able to improve traffic safety but also fuel efficiency. This paper proposes a computational friendly receding-horizon optimization-based nonlinear model predictive control (NMPC) algorithm to achieve fuel-saving speed planning for connected vehicles. The NMPC method solves for the fuel-optimal speed profile of connected vehicles considering a short speed preview of the preceding vehicle. By utilizing such previewing information through vehicle connectivity, the fuel consumption of the connected vehicles is reduced by avoiding unnecessary braking and acceleration, particularly in transient operating conditions. In order to analyze the effectiveness of NMPC design, dynamic programming (DP) method is adopted as a benchmark algorithm where the full speed preview of the preceding vehicle is known. The performances of NMPC and DP designs in driving behavior and fuel economy are quantitatively explored and compared under several standard driving cycles. Results show a promising improvement of the performance by adopting the proposed design and reveal the potential fuel benefits brought by vehicle connectivity and autonomy.

Machine learning (ML) approaches have been extensively exploited to model and to improve wireless communication networks in the past few years. Nonetheless, the estimation of key performance indicators (KPIs) and their uncertainties in Long Term Evolution License Assisted Access (LTE-LAA) based coexistence systems is not adequately addressed. For example, it is not clear if an ML method can accurately predict achievable KPIs (e.g. throughput) and the probability of coexistence (PoC) of LTE-LAA coexistence systems based on partial or no information of MAC and physical layer protocols and parameters. In this paper, we develop a novel ML method by combining a neural network with a logistic regression algorithm to track and estimate KPIs and PoC of coexisting LTE-LAA and wireless local area network (WLAN) links. This ML method can be applied when KPI samples at the base stations (BSs) and access points (APs) are available, without using knowledge of MAC and physical layer parameters. Comparison between the ML and simulation results indicate that the proposed ML method can track the system KPIs and predict the system PoC with good accuracy.

Dynamic spectrum access (DSA) to achieve spectrum sharing in unlicensed bands is a promising approach for meeting the growing demands of forthcoming and deployed wireless networks, such as long-term evolution license-assisted access (LTE-LAA) and IEEE 802.11 Wi-Fi systems. In this paper, we consider a coexistence scenario where multiple LAA and Wi-Fi links compete for spectrum sharing subchannel access. We introduce a reinforcement-learning-based subchannel selection technique which allows access points (APs) and eNBs to select best subchannel distributively considering their medium access control (MAC) channel access protocols along with the physical layer parameters. The performance of this scheme is investigated through simulations, including the convergence property and sum throughput. Numerical results show that the proposed reinforcement-learning scheme converges fast and the sum throughput of the LAA and Wi-Fi systems is reasonably close to the result based on exhaustive search.

Future unlicensed spectrum sharing scenarios involve multi-cell, multi-tier access of incumbent and emerging wireless systems, such as wireless local area network (WLAN), New-Radio unlicensed (NR-U), and Long-Term Evolution (LTE) with Licensed-Assisted Access (LAA). Listen-before-talk (LBT) medium access control (MAC) is a common technique used for channel sensing and access control. Despite intense research efforts, the majority of available results have not accurately analyzed multi-cell LBT with imperfect spectrum sensing. Furthermore, while past studies involved two major spectrum sharing strategies – shared cell access (SCA) and exclusive cell access (ECA), they missed a systematic comparison between the two. In this paper, we develop a unified analytical approach which maps the effects of imperfect spectrum sensing and multi-cell, multi-tier LBT to the key performance indicators (KPIs) of LAA and WLAN cells. We provide an analytical comparison between the SCA and ECA schemes, and show that the SCA can provide a significantly higher system throughput than the ECA as a function of sensing thresholds. We program the SCA and ECA algorithms with a new simulation method and implement Monte Carlo simulations, which verify our analytical results. Numerical results provide insightful observations on effects of various parameters. These results provide powerful analytical and simulation tools to evaluate the performance of multi-tier LBT coexistence systems with imperfect sensing, and effectively support coexistence system optimization.

To meet the ever growing wireless network demands, in terms of subscribers and data throughput, operating long term evolution (LTE) in unlicensed bands, also known as license assisted access (LAA), is introduced as a promising solution. However, the LAA network must co-exist with incumbent IEEE $802.11$ systems. In this paper, we introduce MIMO operation to boost the LAA-Wi- Fi coexistence. We take the impact of imperfect sensing into account and analyze the key performance indicators (KPIs), such as throughput, and probabilities of successful transmissions and collisions. Moreover, the optimal LAA contention window size is characterized to maximize the LAA sum rate while assuring Wi-Fi throughput above a predetermined threshold. Numerical results show the efficiency of the introduced algorithm and that if the parameters are appropriately selected, the throughput of both systems increases.

Carrier sense multiple access with collision avoidance (CSMA/CA) procedures have been specified for medium access control (MAC) in several incumbent and emerging wireless systems, such as wireless local area network (WLAN) and long-term evolution (LTE) with license assisted access (LAA). Clear channel assessment (CCA) errors in carrier sensing can cause significantly degraded network performance. Analyzing the impact of CCA errors in the MAC backoff and transmission process is a challenging task, and very few works have explicitly addressed this. Existing analytical work is only valid for special cases such as independent CCA errors, and the result lacks generality for extension to coexistence systems. In this paper, we try to fill this technical gap by modelling generalized CCA sensing errors which can be either fully correlated or independent due to the fading channel. We develop a new Markov model using matrix-vector representation which captures generalized CCA error events, and analyze the impact of CCA errors on the key performance indicators (KPIs), such as the throughput of both LTE-LAA and WLAN systems. To mitigate the effects of mis-detection and collisions which can cause the network throughput to drop to nearly zero, we propose a soft-collision method to reduce the performance loss. Finally, we program the LTE-LAA and WLAN CCA algorithms and implement extensive computer simulations. Comparisons between analytical and simulation results show consistent matching, and illustrate loss caused by sensing errors and improvement brought by the soft-collision method. This result provides a powerful analytical tool on CSMA/CA MAC-layer performance evaluation with imperfect sensing, applicable to both single and coexistence systems, and has practical value for countermeasure designs against sensing errors.

In system reliability problems, the lifetimes and the repair times are usually indeterministic. When insufficient samples are available to estimate distribution functions of the lifetimes and the repair times, it is rational to invite some experts to evaluate the belief degrees. Uncertainty theory is a branch of mathematics for modelling belief degrees. In this paper, the lifetimes and the repair times of typical repairable systems are assumed to be uncertain variables. Based on the above-mentioned assumptions, the uncertain reliability mathematical models of simple repairable series systems, simple repairable parallel systems, simple repairable series-parallel systems, and simple repairable parallel–series systems are established, respectively. The reliability indices, such as steady state availability, steady state failure frequency, mean up time, and mean down time, are proposed. Finally, some numerical examples are given for illustration.

This paper presents a predictive control method for automotive Selective Catalytic Reduction (SCR) systems to minimize vehicle tailpipe Nitrogen Oxides (NO x ) and ammonia (NH3) emissions. SCR systems have been indispensable in Diesel-powered vehicles to reduce the toxic emissions. To balance the tradeoff between NO x and NH3, the ammonia storage level in an SCR needs to be critically controlled. The proposed control method consists of an ammonia coverage ratio tracking controller and a predictive reference ammonia coverage ratio generator. The reference generator will utilize the predictive information, enabled by growing vehicle connectivity and intelligence, to determine an optimal level of ammonia coverage ratio within the preview horizon. The tracking controller will then drive ammonia coverage ratio to a desired level. The effectiveness of the proposed design approach is demonstrated through simulation studies with experimentally acquired data.

We present AdaFrame, a framework that adaptively selects relevant frames on a per-input basis for fast video recognition. AdaFrame contains a Long Short-Term Memory network augmented with a global memory that provides context information for searching which frames to use over time. Trained with policy gradient methods, AdaFrame generates a prediction, determines which frame to observe next, and computes the utility, i.e., expected future rewards, of seeing more frames at each time step. At testing time, AdaFrame exploits predicted utilities to achieve adaptive lookahead inference such that the overall computational costs are reduced without incurring a decrease in accuracy. Extensive experiments are conducted on two large-scale video benchmarks, FCVID and ActivityNet. AdaFrame matches the performance of using all frames with only 8.21 and 8.65 frames on FCVID and ActivityNet, respectively. We further qualitatively demonstrate learned frame usage can indicate the difficulty of making classification decisions; easier samples need fewer frames while harder ones require more, both at instance-level within the same class and at class-level among different categories., CVPR 2019

We present a one-dimensional model for a piezoelectric semiconductor (PS) rod with varying cross section subject to an axial loading. Using the linearized method, we obtain exact solutions for an n-type PS rod under static extension. Based on the obtained solutions, the effect of the initial concentration of electrons, the axial force, and the radius ratio of two cross sections at the two ends on the redistribution of electrons in the ZnO rod is investigated. Numerical results show that there are a larger variation of electrons near the two ends of the rod, and the piezotronic effect in the PS rod can be controlled through applying an axial force as well as changing the initial concentration of carriers and geometric shape of PS rods.

Due to the fact much of today's data can be represented as graphs, there has been a demand for generalizing neural network models for graph data. One recent direction that has shown fruitful results, and therefore growing interest, is the usage of graph convolutional neural networks (GCNs). They have been shown to provide a significant improvement on a wide range of tasks in network analysis, one of which being node representation learning. The task of learning low-dimensional node representations has shown to increase performance on a plethora of other tasks from link prediction and node classification, to community detection and visualization. Simultaneously, signed networks (or graphs having both positive and negative links) have become ubiquitous with the growing popularity of social media. However, since previous GCN models have primarily focused on unsigned networks (or graphs consisting of only positive links), it is unclear how they could be applied to signed networks due to the challenges presented by negative links. The primary challenges are based on negative links having not only a different semantic meaning as compared to positive links, but their principles are inherently different and they form complex relations with positive links. Therefore we propose a dedicated and principled effort that utilizes balance theory to correctly aggregate and propagate the information across layers of a signed GCN model. We perform empirical experiments comparing our proposed signed GCN against state-of-the-art baselines for learning node representations in signed networks. More specifically, our experiments are performed on four real-world datasets for the classical link sign prediction problem that is commonly used as the benchmark for signed network embeddings algorithms.

Successful spectrum management requires reliable methods for determining whether communication signals are present in a spectrum portion of interest. The widely-used energy detection (ED), or radiometry method is only useful in determining whether a portion of radio-frequency (RF) spectrum contains energy, but not whether this energy carries structured signals such as communications. In this paper we introduce the Lempel-Ziv Markov-chain Sum Algorithm (LZMSA), a spectrum sensing algorithm (SSA) that can detect the presence of a structured signal by leveraging the Liv-Zempel-Markov chain algorithm (LZMA). LZMA is a lossless, general-purpose data compression algorithm that is widely available on many computing platforms. The new algorithm is shown to have good performance at distinguishing between samples containing communication signals, and samples of noise, collected with a software-defined radio (SDR). This algorithm does not require any information about the signal beforehand. The algorithm is tested with computer-generated as well as SDR-captured samples of LTE signals.

This paper presents a qualitatively study of driver's behavior impacts on the fuel consumption and travel time of vehicle platooning system under urban driving scenarios. Vehicle platooning is proven advantageous in improving traffic flow, safety, energy efficiency and emission performance. To design effective platooning strategy, a better understanding of how each driver's behavior influences the overall platoon fuel and time cost will be helpful. An integrated microscopic traffic model with fuel consumption estimation is developed to simulate urban traffic. Each driver's behavior is categorically depicted by personalized parameters in the model. Through extensive Monte Carlo simulations, a strong correlation result is found between maximum acceleration of leading car and travel time, fuel consumption, monetized cost of the platoon, respectively. The relationship is further described by a first-order rational model with reasonably well goodness of fit.

This paper presents a cooperative path planning algorithm using improved Rapidly-exploring Random Trees (RRTs) to generate paths for multiple unmanned air vehicles (UAVs) in real time, from given starting locations to goal locations in the presence of unknown pop-up obstacles. Generating no conflicting paths in obstacle environments for a group of DAVs within a short time is a challenging task. Firstly, we propose an improved RRT by taking the maneuvering constraints of the DAVs into account and a simple and efficient path pruning method is designed to delete redundant nodes on the path. Secondly, a cooperative path planning method is developed that generate a low cost path to avoid collision when DAVs detect a dynamic teammate or pop-up obstacles. Simulation studies are carried out to show the performance of the proposed algorithm.

Networks such as social networks, airplane networks, and citation networks are ubiquitous. To apply advanced machine learning algorithms to network data, low-dimensional and continuous representations are desired. To achieve this goal, many network embedding methods have been proposed recently. The majority of existing methods facilitate the local information i.e. local connections between nodes, to learn the representations, while neglecting global information (or node status), which has been proven to boost numerous network mining tasks such as link prediction and social recommendation. In this paper, we study the problem of preserving local and global information for network embedding. In particular, we introduce an approach to capture global information and propose a network embedding framework LOG, which can coherently model LOcal and Global information. Experiments demonstrate the effectiveness of the proposed framework.

To support efficient spectrum sharing and related standardization efforts in unlicensed spectrum, it is important to develop analytical tools to accurately quantify coexistence performance between long-term evolution license assisted access (LTE-LAA) and incumbent systems, such as wireless local area network (WLAN). Though joint throughput of spectrum sharing LTE-LAA and WLAN systems has been extensively studied, there lacks a systematic study on a high level metric - the probability of coexistence (PoC), which indicates whether coexistence is successful or not probabilistically. Another problem is that the majority of available results either ignored delay constraints, or studied only the mean (or variance) of delay, but have not considered the delay distribution and its impact on throughput. To address these problems, we define and analyze the original PoC metrics between LTE-LAA and WLAN systems based on two practical delay constraints. The first PoC is derived from the joint distribution probability of delays for successful transmissions; and the second PoC is defined upon the joint probability of delay-constrained throughput (DCT) of LAA and WLAN systems. To address the technical difficulties involved, we design a novel analytical framework to evaluate the moment generating function and cumulative distribution function (CDF) of the delay, and a new method to evaluate the DCT and its CDF. Consequently, the PoCs can be evaluated accurately with low complexity. The analytical results are verified by our Monte Carlo simulations, which demonstrate impacts of delay and throughput requirements on the PoCs, and illustrate design tradeoffs and insightful findings. These results provide theoretical and practical value for designing improved LTE-LAA and WLAN systems, and may be extended to other emerging spectrum sharing communication systems.

To provide differentiated quality of service in long-term evolution (LTE) license assisted access (LAA) procedure, the 3GPP has defined several channel access priority classes (CAPCs). They use distinct arbitration inter-frame space (AIFS), contention window (CW) size, and payload duration. While evaluating the effects of CW size and payload duration is relatively straightforward, accurately modelling and analyzing the effect of AIFS has not been satisfactorily addressed. Available methods on analyzing different AIFSs are accurate for only limited parameter setups, or involve systematic approximations. Different from existing results, we develop a non-homogeneous per-slot Markov chain model to represent the state of each priority class during and after the AIFS, and analyze the channel access probability (CAP), successful transmission probability (STP), and average throughput of each class. Some novel features of our method include: 1) we model and solve the per-slot class-dependent link statistics (such as CAP and STP), which vary based on the slot location; and 2) we provide an in-depth analysis on the average throughput, and design a multi-class combinatorial procedure to evaluate average time spent per successful transmission on each delay cell. We program the LAA CAPC algorithms and implement extensive Monte Carlo simulations, which validate the accuracy of our analytical results even in very low throughput region for lower priority classes, and demonstrate the effects of AIFS and other parameters in an LAA system. These results provide solid progress for evaluating priority classes in the LTE-LAA system and other spectrum sharing systems, and can be extended to support system and parameter optimization.

This paper presents a sliding-mode control method for automotive Selective Catalytic Reduction (SCR) systems to regulate ammonia coverage ratios to desired levels. SCR systems have been widely adopted on Diesel powered vehicles to reduce tailpipe NO x emissions. To maintain a high NO x removal efficiency as well as acceptable ammonia slip, the SCR ammonia coverage ratio has to be critically controlled to balance such a tradeoff. A sliding-mode control method is analytically developed to track the target ammonia coverage ratio. Its tracking performance under bounded measurement noise is evaluated. The control input saturation is explicitly considered and its effect on SCR transient behavior is assessed in the design procedure. Simulation results indicate the proposed control can effectively track the target ammonia coverage ratio and comparative study against a traditional PID control demonstrates the improvement of robustness in the presence of bounded disturbances.

Human actions often involve complex interactions across several inter-related objects in the scene. However, existing approaches to fine-grained video understanding or visual relationship detection often rely on single object representation or pairwise object relationships. Furthermore, learning interactions across multiple objects in hundreds of frames for video is computationally infeasible and performance may suffer since a large combinatorial space has to be modeled. In this paper, we propose to efficiently learn higher-order interactions between arbitrary subgroups of objects for fine-grained video understanding. We demonstrate that modeling object interactions significantly improves accuracy for both action recognition and video captioning, while saving more than 3-times the computation over traditional pairwise relationships. The proposed method is validated on two large-scale datasets: Kinetics and ActivityNet Captions. Our SINet and SINet-Caption achieve state-of-the-art performances on both datasets even though the videos are sampled at a maximum of 1 FPS. To the best of our knowledge, this is the first work modeling object interactions on open domain large-scale video datasets, and we additionally model higher-order object interactions which improves the performance with low computational costs., CVPR 2018

This paper introduces a model-based control method for automotive selective catalytic reduction (SCR) systems with preview information of road grade. SCR systems have been widely adopted in Diesel powered ground vehicles to reduce tailpipe NO x emissions. The major control problem is to properly design the ammonia dosing strategy that can efficiently remove NO x without generating excessive NH3 slip at the tailpipe. While most existing methods only utilize sensor feedback information to design controllers, the SCR controller performance can be improved by incorporating preview of road information such that the controller can act in a proactive fashion and reduce emissions over the whole trip. Road grade impact on vehicle SCR system is investigated in this paper. A corresponding controller with explicit consideration of preview road grade is developed and verified in a simulation environment. Comparison results under the US06 test cycle are presented to demonstrate the efficiency improvement of the proposed controller.

To enhance spectrum efficiency in next generation heterogeneous wireless systems, it is important to improve shared spectrum usage between unlicensed long-term evolution (LTE) systems, such as the license-assisted access (LAA), and legacy systems, such as wireless local area networks (WLANs). LTE-LAA uses listen-before-talk (LBT) schemes to enhance coexistence performance. However, available LAA-LBT schemes may incur several problems, such as a slot-jamming effect and a significant slot boundary tracking error, leading to degraded performance. In this paper, we develop an LBT scheme with improved design and software-defined radio (SDR)-based experimental procedure for a performance validation. We program the improved LBT algorithm in the SDR field-programmable gate array (FPGA), and compare the results with the original LBT algorithm. This work also presents an automated testing technique and new SDR functions that modify the LAA parameters in realtime (such as backoff idle slot durations). The experiment results confirm the predicted slot-jamming effect, and show that our improved design enhances LAA throughput. These results provide not only a more robust LAA-LBT design, but also a new method of SDR programming and testing, and insight into the coexistence performance of heterogeneous systems.

Information networks such as social networks, publication networks, and the World Wide Web are ubiquitous in the real world. Traditionally, adjacency matrices are used to represent the networks. However, adjacency matrices are too sparse and too high dimensional when the scale of the networks is large. Network embedding, which aims to learn low-dimensional continuous representations for nodes, has attracted increasing attention recent years. Many network embedding methods such as DeepWalk, LINE, and node2vec have been proposed recently. Many traditional tasks such as node classification and link prediction have been proven to be benefited from the learned representations. Most of these existing network embedding methods focus on preserving the structure of the networks but totally ignore the centrality information. Centrality information which measures the importance of each individual node has been proven to be helpful in many applications. Various centrality measures such as degree centrality, closeness centrality, betweenness centrality, eigenvector centrality and Page-rank, have been proposed. Different centrality measures should be chosen according to different applications. In this paper, we aim to learn continuous nodes representations which can preserve not only the network structure but also the centrality information. We propose a general model to incorporate the centrality information in the sense of ranking. Different centrality measures can be used in the model.