Search

Graph-based clustering plays an important role in the clustering area. Recent studies about graph convolution neural networks have achieved impressive success on graph type data. However, in general clustering tasks, the graph structure of data does not exist such that the strategy to construct a graph is crucial for performance. Therefore, how to extend graph convolution networks into general clustering tasks is an attractive problem. In this paper, we propose a graph auto-encoder for general data clustering, which constructs the graph adaptively according to the generative perspective of graphs. The adaptive process is designed to induce the model to exploit the high-level information behind data and utilize the non-Euclidean structure sufficiently. We further design a novel mechanism with rigorous analysis to avoid the collapse caused by the adaptive construction. Via combining the generative model for network embedding and graph-based clustering, a graph auto-encoder with a novel decoder is developed such that it performs well in weighted graph used scenarios. Extensive experiments prove the superiority of our model., Comment: Accepted by IEEE Transactions on Pattern Analysis and Machine Intelligence

As to unsupervised learning, most discriminative information is encoded in the cluster labels. To obtain the pseudo labels, unsupervised feature selection methods usually utilize spectral clustering to generate them. Nonetheless, two related disadvantages exist accordingly: 1) the performance of feature selection highly depends on the constructed Laplacian matrix and 2) the pseudo labels are obtained with mixed signs, while the real ones should be nonnegative. To address this problem, a novel approach for unsupervised feature selection is proposed by extending orthogonal least square discriminant analysis (OLSDA) to the unsupervised case, such that nonnegative pseudo labels can be achieved. Additionally, an orthogonal constraint is imposed on the class indicator to hold the manifold structure. Furthermore, l2,1 regularization is imposed to ensure that the projection matrix is row sparse for efficient feature selection and proved to be equivalent to l2,0 regularization. Finally, extensive experiments on nine benchmark data sets are conducted to demonstrate the effectiveness of the proposed approach.

Graph clustering, aiming to partition nodes of a graph into various groups via an unsupervised approach, is an attractive topic in recent years. To improve the representative ability, several graph auto-encoder (GAE) models, which are based on semisupervised graph convolution networks (GCN), have been developed and they have achieved impressive results compared with traditional clustering methods. However, all existing methods either fail to utilize the orthogonal property of the representations generated by GAE or separate the clustering and the training of neural networks. We first prove that the relaxed k-means will obtain an optimal partition in the inner-product distance used space. Driven by theoretical analysis about relaxed k-means, we design a specific GAE-based model for graph clustering to be consistent with the theory, namely Embedding GAE (EGAE). The learned representations are well explainable so that the representations can be also used for other tasks. To induce the neural network to produce deep features that are appropriate for the specific clustering model, the relaxed k-means and GAE are learned simultaneously. Meanwhile, the relaxed k-means can be equivalently regarded as a decoder that attempts to learn representations that can be linearly constructed by some centroid vectors. Accordingly, EGAE consists of one encoder and dual decoders. Extensive experiments are conducted to prove the superiority of EGAE and the corresponding theoretical analyses.

The conventional subspace clustering method obtains explicit data representation that captures the global structure of data and clusters via the associated subspace. However, due to the limitation of intrinsic linearity and fixed structure, the advantages of prior structure are limited. To address this problem, in this brief, we embed the structured graph learning with adaptive neighbors into the deep autoencoder networks such that an adaptive deep clustering approach, namely, autoencoder constrained clustering with adaptive neighbors (ACC_AN), is developed. The proposed method not only can adaptively investigate the nonlinear structure of data via a parameter-free graph built upon deep features but also can iteratively strengthen the correlations among the deep representations in the learning process. In addition, the local structure of raw data is preserved by minimizing the reconstruction error. Compared to the state-of-the-art works, ACC_AN is the first deep clustering method embedded with the adaptive structured graph learning to update the latent representation of data and structured deep graph simultaneously.

Neural network based clustering methods usually have better performance compared to the conventional approaches due to more efficient feature extraction. Most of existing deep clustering techniques either exploit graph information as prior to extract pivotal deep structure from the raw data and simply utilizes stochastic gradient descent (SGD). However, they often suffer from separating the learning steps regarding dimensionality reduction and clustering. To address these issues, a novel deep model named as deep fuzzy k-means (DFKM) with adaptive loss function and entropy regularization is proposed. DFKM performs deep feature extraction and fuzzy clustering simultaneously to generate a more appropriate nonlinear feature map. Additionally, DFKM incorporates FKM so that fuzzy information is utilized to represent a clear structure of deep clusters. To further promote the robustness of the model, a robust loss function is applied to the objective with adaptive weights. Moreover, an entropy regularization is employed for affinity to provide confidence of each assignment and the corresponding membership and centroid matrices are updated by close-form solutions rather than SGD. Extensive experiments show that DFKM has better performance compared to the state-of-the-art fuzzy clustering techniques under three clustering metrics.

Water storage change of the lakes in the Tibetan Plateau is regarded as one of the most critical regional hydrological consequences owing to climate change. In this study, we investigate the water storage changes in 22 lakes in the Tibetan Plateau based on sequential remote sensed lake area and water level, which are derived from moderate resolution imaging spectroradiometer (MODIS) surface reflectance and Laboratoire D’Etudes en Geophysique et Oceanographie Spatiales (LEGOS) altimetry data, respectively. Water storage of the lake is estimated on the basis of the relationship between lake area and water level. The method can be seen as an alternative to the conventional hydrological approaches. The results show that, during 2001–2017, most of the studied lakes in the Tibetan Plateau have shown significant increasing trends in water storage accompanied with larger lake area and higher water level. The changes in lake water storage are found in close relation to variations of climate factors such as precipitation, potential evaporation, and temperature in most lakes. The climate change impacts, however, can be amplified or attenuated by other environmental factors in some lake catchments.

A massive green tide (i.e., Ulva prolifera bloom) event in summer 2015 in the Yellow Sea was investigated using satellite imagery and a trajectory model. First, the occurrence and evolution of the macroalgal bloom were studied based on time series of cloud-free Moderate Resolution Imaging Spectroradiometer imagery during May to August using the Floating Algae Index detection method. A Lagrangian spill trajectory model, i.e., the General NOAA (National Oceanic and Atmospheric Administration) Operational Modeling Environment (GNOME) model, which was originally designed to simulate the transport of oil spills, was then implemented to produce trajectories of the macroalgae. The model was forced by the 3-hourly ocean surface current data from the Naval Oceanographic Office global-scale operational ocean prediction system, and 6-hourly blended surface wind products from NOAA/National Climatic Data Center. The simulated transport of the green tide agrees fairly well with satellite observations over the time span of about 10 days, indicating that the combination of GNOME model and satellite data can be employed for rapid algal bloom response and environmental impact assessment. The results of numerical experiments also show that the surface winds play a significant role in the movement of the macroalgae in the Yellow Sea where the surface currents in summer are primarily driven by sea winds.

To exploit the potential of a dual-comb absolute distance ranging system outside a well-controlled laboratory, a compact dual-comb structure is proposed. A beat frequency generated by the repetition rates of two frequency combs serves as an electric reference, which is easily built and maintains a long range and high resolution compared with traditional dual-comb systems. The performance of the proposed method is compared with that of a heterodyne interferometer. The residuals range within -116.6 to 117.2 nm, the standard deviations vary from 46.3 to 137.9 nm, and the non-ambiguity range extension remains reliable throughout a 10-m test. Compared with Michelson-type dual-comb interferometers, this compact dual-comb system omits the redundant optical reference arm, promising practical applications of distance ranging.

In this letter, a quasi-planar waveguide tuner is proposed, which is comprised of a main waveguide, three waveguide shorting stubs, and three ridged waveguide shorting stubs. All waveguide stubs are connected to the E-plane of the main waveguide. It can be used to replace the well-known E-H waveguide tuners with a simpler structure, smaller size, and lower fabrication cost. The matching ability is verified by using the quasi-planar tuner to match the discontinuities from an inductive iris and a capacitive iris. The matching results are shown to be the best compared with that of three other conventional tuners, an E-plane tuner, an H-plane tuner, and an E-H tuner. Tested results are also presented.

The merit factor (MF) introduced by Golay has long been accepted as the standard criterion to evaluate and design binary sequences with good anti-multipath property in sonar, radar, and communication systems for its theoretical tightness and practical simplicity. In this paper, we first show that the MF is a biased anti-multipath performance-evaluation metric in theory and, more importantly, that it is not a pertinent sequence design criterion in practice for most binary sequences of practical interest. Then, we propose the weighted MF (WMF) based on a nonuniform weighting of the out-of-phase aperiodic autocorrelation function (ACF) that provides accurate measurement of self-generated interference for the constant-amplitude complex-valued sequences and the nonconstant modulus sequences. Based on the WMF, a list of ldquobadrdquo (of low MFs) binary sequences (lengths 33-95) with better anti-multipath performance than the ldquobestrdquo (known) binary sequences has been designed to verify its greater pertinence over the MF as a sequence design criterion for sonar, radar, and communication systems. Moreover, we extend the weighted correlation model of the WMF to code-division multiple-access (CDMA) systems and propose the weighted cross-correlation factor (WCF) to evaluate the sequence set's multiple-access interference (MAI) rejection property in the context of multipath propagation. Theoretical analysis corroborated by simulations confirms that the WCF provides greater practical pertinence and analytical tractability than the current standard criterion.

The diversity-multiplexing tradeoff for multiple-input multiple-output (MIMO) point-to-point channels and multiple access channels were first proposed and studied by Zheng and Tse recently. While the optimal tradeoff curves for MIMO channels have been explicitly explored, those corresponding to some suboptimal and practical MIMO schemes are still open. One such important problem is the diversity-multiplexing tradeoff for a V-BLAST type system employing ordered successive interference cancellation (SIC) receivers with zero forcing (ZF) or minimum mean square error (MMSE) processing at each stage. In this paper, we take a novel geometrical approach and rigorously verify that under general settings, the optimal ordering rule for a V-BLAST SIC receiver will not improve its performance regarding diversity-multiplexing tradeoff in point-to- point channels. The same geometrical tool is then applied to MIMO spatial-division multiple access channels, leading to some first results in this area. Particularly, we reveal that when the rates of data streams are fixed (i.e., zero spatial multiplexing gain), the diversity order is not improved by user ordering.

Cooperative transmission by base stations (BSs) can significantly improve the spectral efficiency of multiuser, multi-cell, multiple input multiple output (MIMO) systems. We show that contrary to what is often assumed in the literature, the multiuser interference in such systems is fundamentally asynchronous. Intuitively, perfect timing-advance mechanisms can at best only ensure that the desired signal components -but not also the interference components -are perfectly aligned at their intended mobile stations. We develop an accurate mathematical model for the asynchronicity, and show that it leads to a significant performance degradation of existing designs that ignore the asynchronicity of interference. Using three previously proposed linear preceding design methods for BS cooperation, we develop corresponding algorithms that are better at mitigating the impact of the asynchronicity of the interference. Furthermore, we also address timing-advance inaccuracies (jitter), which are inevitable in a practical system. We show that using jitter-statistics-aware precoders can mitigate the impact of these inaccuracies as well. The insights of this paper are critical for the practical implementation of BS cooperation in multiuser MIMO systems, a topic that is typically oversimplified in the literature.

A combination of orthogonal frequency division multiplexing (OFDM) and Multiple-input-multiple-output (MIMO) systems appears to be a promising solution for the PHY layer of indoor multimedia transmission via wireless Local Area Networks (WLANs). Antenna selection is an excellent way of reducing the hardware costs of MIMO-OFDM systems while retaining high performance. This paper addresses two major practical concerns for the application of antenna selection: antenna selection training protocol design, and calibration to solve RF imbalance. We present novel solutions that are especially suitable for slowly time-varying environments, e.g., indoor scenarios, sports stadiums, and shopping malls. Specifically, a low Doppler spread associated with such environments enables us to train all antenna subsets by multiple training packets transmitted in burst; consequently antenna selection techniques can be accommodated in the emerging standards with minimum modifications. In order to deal with RF imbalance, we propose a novel calibration procedure that reduces the performance degradations. Both numerical and analytical approaches are used to verify the effectiveness of the proposed solutions, which make antenna selection more easily adaptable for high-throughput WLAN systems. Our solutions have been accommodated in the current draft of the IEEE 802.11n standard for high-throughput WLANs

In recent years, the remarkable ability of multiple-input-multiple-output (MIMO) wireless communication systems to provide spatial diversity or multiplexing gains has been clearly demonstrated. For MIMO diversity schemes, it is well known that antenna selection methods that optimize the postprocessing signal-to-noise ratio (SNR) can preserve the diversity order of the original full-size MIMO system. On the other hand, the diversity order achieved by antenna selection in spatial multiplexing systems, especially those exploiting practical coding and decoding schemes, has not thus far been rigorously analyzed. In this paper, a geometrical framework is proposed to theoretically analyze the diversity order achieved by transmit antenna selection for separately encoded spatial multiplexing systems with linear and decision-feedback receivers. When two antennas are selected from the transmitter, the exact achievable diversity order is rigorously derived, which previously only appears as conjectures based on numerical results in the literature. If more than two antennas are selected, we give lower and upper bounds on the achievable diversity order. Furthermore, the same geometrical approach is used to evaluate the diversity-multiplexing tradeoff in spatial multiplexing systems with transmit antenna selection

Motivated by matrix determinant properties, this letter develops a fast transmit antenna selection algorithm for MIMO systems: the G-circles method. This novel scheme is shown to achieve many advantages over other existing algorithms