Deep neural network (NN) has been widely accepted for enabling various AI applications, however, the limitation of computational and memory resources is a major problem on mobile devices. Quantized NN with a reduced bit precision is an effective solution, which relaxes the resource requirements, but the accuracy degradation due to its numerical approximation is another problem. We propose a novel quantized NN model employing the “dithering” technique to improve the accuracy with the minimal additional hardware requirement at the view point of the hardware-algorithm co-designing. Dithering distributes the quantization error occurring at each pixel (neuron) spatially so that the total information loss of the plane would be minimized. The experiment we conducted using the software-based accuracy evaluation and FPGA-based hardware resource estimation proved the effectiveness and efficiency of the concept of an NN model with dithering.

Stance detection in twitter aims at mining user stances expressed in a tweet towards a single or multiple target entities. Detecting and analyzing user stances from massive opinion-oriented twitter posts provide enormous opportunities to journalists, governments, companies, and other organizations. Most of the prior studies have explored the traditional deep learning models, e.g., long short-term memory (LSTM) and gated recurrent unit (GRU) for detecting stance in tweets. However, compared to these traditional approaches, recently proposed densely connected bidirectional LSTM and nested LSTMs architectures effectively address the vanishing-gradient and overfitting problems as well as dealing with long-term dependencies. In this paper, we propose a neural network model that adopts the strengths of these two LSTM variants to learn better long-term dependencies, where each module coupled with an attention mechanism that amplifies the contribution of important elements in the final representation. We also employ a multi-kernel convolution on top of them to extract the higher-level tweet representations. Results of extensive experiments on single and multi-target benchmark stance detection datasets show that our proposed method achieves substantial improvement over the current state-of-the-art deep learning based methods.

Most of latency-sensitive mobile applications depend on computational resources provided by a cloud computing service. The problem of relying on cloud computing is that, sometimes, the physical locations of cloud servers are distant from mobile users and the communication latency is long. As a result, the concept of distributed cloud service, called mobile edge computing (MEC), is being introduced in the 5G network. However, MEC can reduce only the communication latency. The computing latency in MEC must also be considered to satisfy the required total latency of services. In this research, we study the impact of both latencies in MEC architecture with regard to latency-sensitive services. We also consider a centralized model, in which we use a controller to manage flows between users and mobile edge resources to analyze MEC in a practical architecture. Simulations show that the interval and controller latency trigger some blocking and error in the system. However, the permissive system which relaxes latency constraints and chooses an edge server by the lowest total latency can improve the system performance impressively.

Personalized trip planning is a challenging problem given that places of interest should be selected according to user preferences and sequentially arranged while satisfying various constraints. In this study, we aimed to model various uncertain aspects that should be considered during trip planning and efficiently generate personalized plans that maximize user satisfaction based on preferences and constraints. Specifically, we propose a probabilistic itinerary evaluation model based on a hybrid temporal Bayesian network that determines suitable itineraries considering preferences, constraints, and uncertain environmental variables. The model retrieves the sum of time-weighted user satisfaction, and ant colony optimization generates the trip plan that maximizes the objective function. First, the optimization algorithm generates candidate itineraries and evaluates them using the proposed model. Then, we improve candidate itineraries based on the evaluation results of previous itineraries. To validate the proposed trip planning approach, we conducted an extensive user study by asking participants to choose their preferred trip plans from options created by a human planner and our approach. The results show that our approach provides human-like trip plans, as participants selected our generated plans in 57% of the pairs. We also evaluated the efficiency of the employed ant colony optimization algorithm for trip planning by performance comparisons with other optimization methods.

A pupil function of aperture in image capturing systems is theoretically derived such that one can perfectly reconstruct all-in-focus image through linear filtering of the focal stack. The perfect reconstruction filters are also designed based on the derived pupil function. The designed filters are space-invariant; hence the presented method does not require region segmentation. Simulation results using synthetic scenes shows effectiveness of the derived pupil function and the filters.

Parameter estimation theorems for LFM signals have been developed due to the advantages of fractional Fourier transform (FrFT). The traditional estimation methods in the fractional Fourier domain (FrFD) are almost based on two-dimensional search which have the contradiction between estimation performance and complexity. In order to solve this problem, we introduce the orthogonal matching pursuit (OMP) into the FrFD, propose a modified optimization method to estimate initial frequency and final frequency of fractional bandlimited LFM signals. In this algorithm, the differentiation fractional spectrum which is used to form observation matrix in OMP is derived from the spectrum analytical formulations of the LFM signal, and then, based on that the LFM signal has approximate rectangular spectrum in the FrFD and the correlation between the LFM signal and observation matrix yields a maximal value at the edge of the spectrum (see Sect.3.3 for details), the edge spectrum information can be extracted by OMP. Finally, the estimations of initial frequency and final frequency are obtained through multiplying the edge information by the sampling frequency resolution. The proposed method avoids reconstruction and the traditional peak-searching procedure, and the iterations are needed only twice. Thus, the computational complexity is much lower than that of the existing methods. Meanwhile, Since the vectors at the initial frequency and final frequency points both have larger modulus, so that the estimations are closer to the actual values, better normalized root mean squared error (NRMSE) performance can be achieved. Both theoretical analysis and simulation results demonstrate that the proposed algorithm bears a relatively low complexity and its estimation precision is higher than search-based and reconstruction-based algorithms.

Liquid crystal director distributions between strong and weak polar anchoring surfaces in hybrid aligned cells are numerically analyzed. When the anchoring is a critical one, homogeneously or homeotropicly liquid crystal alignment can be obtained. Such cells have no threshold voltage and a driving voltage can be reduced less than 0.5 volt.

Person Re-Identification has received extensive study in the past few years and achieves impressive progress. Recent outstanding methods extract discriminative features by slicing feature maps of deep neural network into several stripes. Still there have improvement on feature fusion and metric learning strategy which can help promote slice-based methods. In this paper, we propose a novel framework that is end-to-end trainable, called Multi-level Slice-based Network (MSN), to capture features both in different levels and body parts. Our model consists of a dual-branch network architecture, one branch for global feature extraction and the other branch for local ones. Both branches process multi-level features using pyramid feature alike module. By concatenating the global and local features, distinctive features are exploited and properly compared. Also, our proposed method creatively introduces a triplet-center loss to elaborate combined loss function, which helps train the joint-learning network. By demonstrating the comprehensive experiments on the mainstream evaluation datasets including Market-1501, DukeMTMC, CUHK03, it indicates that our proposed model robustly achieves excellent performance and outperforms many of existing approaches. For example, on DukeMTMC dataset in single-query mode, we obtain a great result of Rank-1/mAP =85.9%(+1.0%)/74.2%(+4.7%).

When the track density of two-dimensional magnetic recording (TDMR) systems is increased, intertrack interference (ITI) inevitably grows, resulting in the extreme degradation of an overall system performance. In this work, we present coding, writing, and reading techniques which allow TDMR systems with multi-readers to overcome severe ITI. A rate-5/6 two-dimensional (2D) modulation code is adopted to protect middle-track data from ITI based on cross-track data dependence. Since the rate-5/6 2D modulation code greatly improves the reliability of the middle-track, there is a bit-error rate gap between middle-track and sidetracks. Therefore, we propose the different track width writing technique to optimize the reliability of all three data tracks. In addition, we also evaluate the TDMR system performance using an user areal density capability (UADC) as a main key parameter. Here, an areal density capability (ADC) can be measured by finding the bit-error rate of the system with sweeping track and linear densities. The UADC is then obtained by removing redundancy from the ADC. Simulation results show that a system with our proposed techniques gains the UADC of about 4.66% over the conventional TDMR systems.

Super resolution (SR) image reconstruction has attracted increasing attention these years and many SR image reconstruction algorithms have been proposed for restoring a high-resolution image from one or multiple low-resolution images. However, how to objectively evaluate the quality of SR reconstructed images remains an open problem. Although a great number of image quality metrics have been proposed, they are quite limited to evaluate the quality of SR reconstructed images. Inspired by this, this paper presents a blind quality index for SR reconstructed images using first- and second-order structural degradation. First, the SR reconstructed image is decomposed into multi-order derivative magnitude maps, which are effective for first- and second-order structural representation. Then, log-energy based features are extracted on these multi-order derivative magnitude maps in the frequency domain. Finally, support vector regression is used to learn the quality model for SR reconstructed images. The results of extensive experiments that were conducted on one public database demonstrate the superior performance of the proposed method over the existing quality metrics. Moreover, the proposed method is less dependent on the number of training images and has low computational cost.

Low-density parity-check (LDPC) codes can be used to improve the storage reliability of multi-level cell (MLC) flash memories because of their strong error-correcting capability. In order to improve the weighted bit-flipping (WBF) decoding of LDPC codes in MLC flash memories with cell-to-cell interference (CCI), we propose two strategies of normalizing weights and adjusting log-likelihood ratio (LLR) values. Simulation results show that the WBF decoding under the proposed strategies is much advantageous in both error and convergence performances over existing WBF decoding algorithms. Based on complexity analysis, the strategies provide the WBF decoding with a good tradeoff between performance and complexity.

Modulation spectrogram is effective for analyzing underwater signals which consist of tonal and modulated components. This method can analyze the acoustic and modulation frequency at the same time, but has the trade-off issue of time-frequency localization. This letter introduces a reassignment method for overcoming the localization issue in conventional spectrograms, and then presents an alignment scheme for implementing modulation spectrogram. Relevant experiments show improvement in acoustic frequency estimation perspective and an increment in analyzable modulation frequency range.

The least mean squares (LMS) algorithm has been widely used for adaptive filtering because of easily implementing at a computational complexity of O(2N) where N is the number of taps. The drawback of the LMS algorithm is that its performance is sensitive to the scaling of the input. The normalized LMS (NLMS) algorithm solves this problem on the LMS algorithm by normalizing with the sliding-window power of the input; however, this normalization increases the computational cost to O(3N) per iteration. In this work, we derive a new formula to strictly perform the NLMS algorithm at a computational complexity of O(2N), that is referred to as the C-NLMS algorithm. The derivation of the C-NLMS algorithm uses the H∞ framework presented previously by one of the authors for creating a unified view of adaptive filtering algorithms. The validity of the C-NLMS algorithm is verified using simulations.

Pseudo-random sequences with good statistical property, such as low autocorrelation, high linear complexity and 2-adic complexity, have been widely applied to designing reliable stream ciphers. In this paper, we explicitly determine the 2-adic complexities of two classes of generalized cyclotomic binary sequences with order 4. Our results show that the 2-adic complexities of both of the sequences attain the maximum. Thus, they are large enough to resist the attack of the rational approximation algorithm for feedback with carry shift registers. We also present some examples to illustrate the validity of the results by Magma programs.

In this letter, a construction of sparse measurement matrices is presented. Based on finite fields, a base matrix is obtained. Then a Hadamard matrix or a discrete Fourier transform (DFT) matrix is nested in the base matrix, which eventually formes a new deterministic measurement matrix. The coherence of the proposed matrices is low, which meets the Welch bound asymptotically. Thus these matrices could satisfy the restricted isometry property (RIP). Simulation results demonstrate that the proposed matrices give better performance than Gaussian counterparts.

The estimation of the matrix rank of harmonic components of a music spectrogram provides some useful information, e.g., the determination of the number of basis vectors of the matrix-factorization-based algorithms, which is required for the automatic music transcription or in post-processing. In this work, we develop an algorithm based on Stein's unbiased risk estimator (SURE) algorithm with the matrix factorization model. The noise variance required for the SURE algorithm is estimated by suppressing the harmonic component via median filtering. An evaluation performed using the MIDI-aligned piano sounds (MAPS) database revealed an average estimation error of -0.26 (standard deviation: 4.4) for the proposed algorithm.

We propose a method to realize a subjective super-resolution on a high-speed LED display, which dynamically shows a set of four neighboring pixels on every LED pixel. We have experimentally confirmed the subjective super-resolution effect. This paper proposes a subjective super-resolution hypothesis in human visual system and reports simulation results with pseudo fixation eye movements.

Distributed compressive video sensing (DCVS) has received considerable attention due to its potential in source-limited communication, e.g., wireless video sensor networks (WVSNs). Multi-hypothesis (MH) prediction, which treats the target block as a linear combination of hypotheses, is a state-of-the-art technique in DCVS. The common approach is under the supposition that blocks that are dissimilar from the target block are given lower weights than blocks that are more similar. This assumption can yield acceptable reconstruction quality, but it is not suitable for scenarios with more details. In this paper, based on the joint sparsity model (JSM), the authors present a Tikhonov-regularized MH prediction scheme in which the most similar block provides the similar common portion and the others blocks provide respective unique portions, differing from the common supposition. Specifically, a new scheme for generating hypotheses and a Euclidean distance-based metric for the regularized term are proposed. Compared with several state-of-the-art algorithms, the authors show the effectiveness of the proposed scheme when there are a limited number of hypotheses.

A multi-camera setup for a surveillance system enables a larger coverage area, especially when a single camera has limited monitoring capability due to certain obstacles. Therefore, for large-scale coverage, multiple cameras are the best option. In this paper, we present a method for detecting multiple objects using several cameras with large overlapping views as this allows synchronization of object identification from a number of views. The proposed method uses a graph structure that is robust enough to represent any detected moving objects by defining their vertices and edges to determine their relationships. By evaluating these object features, represented as a set of attributes in a graph, we can perform lightweight multiple object detection using several cameras, as well as performing object tracking within each camera's field of view and between two cameras. By evaluating each vertex hierarchically as a subgraph, we can further observe the features of the detected object and perform automatic separation of occluding objects. Experimental results show that the proposed method would improve the accuracy of object tracking by reducing the occurrences of incorrect identification compared to individual camera-based tracking.

In this letter, we investigate a design of efficient antenna allocation at the full duplex receiver (FDR) in a multi-input multi-output multi-eavesdropper (MIMOME) wiretap channel for physical layer security improvement. Specifically, we propose the allocation which are feasible for the practical scenario with self-interference (SI) taken into account, because the jamming signals from FDR not only confuse the eavesdropper but also inevitably cause SI at the FDR. Due to the nolinear and coupling of the antenna allocation optimization problem, we transform the original problem into an integer programming problem. Then, we derive the optimal solution and the corresponding beamforming matrices in closed-form by means of combining spatial alignment and null-space projection method. Furthermore, we present the feasibility condition and full-protection condition, which offer insight into principles that enable more efficient and effective use of FDR in the wiretap channel for security improvement. From the simulation results, we validate the theoretical analysis and demonstrate the outstanding performance of the proposed antennas allocation at FDR.