Mutual Information Metric Research Articles

Preprocessing and postprocessing analysis for hot-mix asphalt dynamic modulus experimental data

Dynamic modulus (|E*|) measurements of hot-mix asphalt (HMA) mixtures are critical for understanding material behavior but present significant challenges due to the complex testing procedures and precision required. While substantial efforts have focused on developing predictive models for |E*|, the critical role of experimental data preprocessing has been largely overlooked in the existing literature. This study addresses this gap by proposing a novel, comprehensive framework for both preprocessing and post-processing of dynamic modulus data. Leveraging the well-established ASU |E*| database and the NCHRP 1–40D Witczak prediction model as benchmarks, we introduce advanced empirical techniques, including probability distribution analysis, scatter and box plots, correlation coefficients, and mutual information metrics, to refine data quality and enhance the interpretability of predictive models. Our approach reveals new insights into the interaction of various input factors and |E*| values, leading to improved model robustness and reliability. The post-processing techniques further substantiate the predictive power of the Witczak model, yielding significant enhancements in accuracy and reliability. This research pioneers a standardized data preparation methodology that sets a new precedent in asphalt material engineering, offering a robust foundation for future |E*| modeling and analysis.

Secure multi-server coded caching

This paper considers multiple cache-enabled end-users connected to multiple servers through a linear network. We also prevent an external eavesdropper from obtaining any information about the original files in cache-aided networks. The eavesdropper does not exist in the placement phase. He can analyze and capture the multi-cast messages in the content delivery phase. He receives a single linear combination derived from each multi-cast message. Hence, we only consider the security of the delivery phase. Our work generalizes the secure caching problem with one server to a multi-server setup. A secure centralized multi-server coded caching scenario is considered, and closed-form coding delay and secret shared key storage expression are provided. Regarding our security guarantee, we show that the delivery phase does not reveal any information to the eavesdropper in terms of the mutual information metric. We analyze the system’s performance in terms of coding delay and guarantee the security of our scheme using the mutual information metric. Numerical evaluations verify that security incurs a negligible cost in terms of memory usage when the number of files and users increases, i.e., the secure and insecure bounds almost coincide. Also, we numerically show that our proposed scheme outperforms the secure coded caching problem with one server.

Proof of biased behavior of Normalized Mutual Information

The Normalized Mutual Information (NMI) metric is widely utilized in the evaluation of clustering and community detection algorithms. This study explores the performance of NMI, specifically examining its performance in relation to the quantity of communities, and uncovers a significant drawback associated with the metric's behavior as the number of communities increases. Our findings reveal a pronounced bias in the NMI as the number of communities escalates. While previous studies have noted this biased behavior, they have not provided a formal proof and have not addressed the causation of this problem, leaving a gap in the existing literature. In this study, we fill this gap by employing a mathematical approach to formally demonstrate why NMI exhibits biased behavior, thereby establishing its unsuitability as a metric for evaluating clustering and community detection algorithms. Crucially, our study exposes the vulnerability of entropy-based metrics that employ logarithmic functions to similar bias.

Single-Photon Level Dispersive Fourier Transform: Ultrasensitive Characterization of Noise-Driven Nonlinear Dynamics.

Dispersive Fourier transform is a characterization technique that allows directly extracting an optical spectrum from a time domain signal, thus providing access to real-time characterization of the signal spectrum. However, these techniques suffer from sensitivity and dynamic range limitations, hampering their use for special applications in, e.g., high-contrast characterizations and sensing. Here, we report on a novel approach to dispersive Fourier transform-based characterization using single-photon detectors. In particular, we experimentally develop this approach by leveraging mutual information analysis for signal processing and hold a performance comparison with standard dispersive Fourier transform detection and statistical tools. We apply the comparison to the analysis of noise-driven nonlinear dynamics arising from well-known modulation instability processes. We demonstrate that with this dispersive Fourier transform approach, mutual information metrics allow for successfully gaining insight into the fluctuations associated with modulation instability-induced spectral broadening, providing qualitatively similar signatures compared to ultrafast photodetector-based dispersive Fourier transform but with improved signal quality and spectral resolution (down to 53 pm). The technique presents an intrinsically unlimited dynamic range and is extremely sensitive, with a sensitivity reaching below the femtowatt (typically 4 orders of magnitude better than ultrafast dispersive Fourier transform detection). We show that this method can not only be implemented to gain insight into noise-driven (spontaneous) frequency conversion processes but also be leveraged to characterize incoherent dynamics seeded by weak coherent optical fields.

Community Detection On Multi-layer Graph using Intra-layer and Inter-layer Linkage Graphs (CDMIILG)

Effective communication in complex networks is crucial for users and organizations, so graph theory is being used to provide solid groundwork for modeling and evaluating these networks. However, users often need help with group splits and communication issues due to ineffective community models. In addressing these limitations, this study proposed a community discovery technique based on Community Detection on Multi-layer Graph using Intra-layer and Inter-layer Linkage Graphs (CDMIILG) to model multiple user relationships. The CDMIILG technique adopts centrality measures, depth-first search models, recursive and transitivity, with the latter playing significant roles in the network. The practicability and effectiveness of the proposed method were evaluated using modern centrality measures and a graph-based machine-learning approach. The model’s performance was evaluated utilizing notable clustering metrics, including Adjusted Rand Index (ARI), Normalized Mutual Information (NMI), Modularity Approach, and Statistical Metrics, on seven real-world data sets. The study findings revealed that the proposed model is more effective than existing models for multi-layer graphs by yielding a 2%–7% increment.

CSCIM_FS: Cosine similarity coefficient and information measurement criterion-based feature selection method for high-dimensional data

Feature selection (FS) based on mutual information (MI) metrics needs to discretize the data in preprocessing, which is a convenient way to identify correlation between features. However, information loss often occurs in data discretization. In order to solve this information loss problem, this paper proposes a FS algorithm based on cosine similarity coefficient and information measurement criterion (CSCIM_FS). First, the MI between features and tags is calculated, and features are sorted out according to the MI calculated. Then, a feature matrix is constructed to transform the one-dimensional feature sequence into a two-dimensional square matrix. Next, cosine transform is adopted to obtain the high-frequency components of the feature matrix, and sampling is conducted to derive the hash fingerprint of the feature matrix. After that, the similarity between every two features is calculated on the basis of the hash fingerprints of different features. Finally, the feature weight is calculated according to tags, the MI and similarity between features, and a key feature subset is obtained and used to conduct feature selection from the data. The experimental results on several UCI public datasets show that CSCIM_FS algorithm selected a feature subset with high accuracy, and that this algorithm performs better than MIM, CMIM, mRMR and other algorithms.

Atypical speech production of multisyllabic words and phrases by children with developmental dyslexia.

The prevalent "core phonological deficit" model of dyslexia proposes that the reading and spelling difficulties characterizing affected children stem from prior developmental difficulties in processing speech sound structure, for example, perceiving and identifying syllable stress patterns, syllables, rhymes and phonemes. Yet spoken word production appears normal. This suggests an unexpected disconnect between speech input and speech output processes. Here we investigated the output side of this disconnect from a speech rhythm perspective by measuring the speech amplitude envelope (AE) of multisyllabic spoken phrases. The speech AE contains crucial information regarding stress patterns, speech rate, tonal contrasts and intonational information. We created a novel computerized speech copying task in which participants copied aloud familiar spoken targets like "Aladdin." Seventy-five children with and without dyslexia were tested, some of whom were also receiving an oral intervention designed to enhance multi-syllabic processing. Similarity of the child's productions to the target AE was computed using correlation and mutual information metrics. Similarity of pitch contour, another acoustic cue to speech rhythm, was used for control analyses. Children with dyslexia were significantly worse at producing the multi-syllabic targets as indexed by both similarity metrics for computing the AE. However, children with dyslexia were not different from control children in producing pitch contours. Accordingly, the spoken production of multisyllabic phrases by children with dyslexia is atypical regarding the AE. Children with dyslexia may not appear to listeners to exhibit speech production difficulties because their pitch contours are intact. RESEARCH HIGHLIGHTS: Speech production of syllable stress patterns is atypical in children with dyslexia. Children with dyslexia are significantly worse at producing the amplitude envelope of multi-syllabic targets compared to both age-matched and reading-level-matched control children. No group differences were found for pitch contour production between children with dyslexia and age-matched control children. It may be difficult to detect speech output problems in dyslexia as pitch contours are relatively accurate.

Learning Distributed Representations and Deep Embedded Clustering of Texts

Instructors face significant time and effort constraints when grading students’ assessments on a large scale. Clustering similar assessments is a unique and effective technique that has the potential to significantly reduce the workload of instructors in online and large-scale learning environments. By grouping together similar assessments, marking one assessment in a cluster can be scaled to other similar assessments, allowing for a more efficient and streamlined grading process. To address this issue, this paper focuses on text assessments and proposes a method for reducing the workload of instructors by clustering similar assessments. The proposed method involves the use of distributed representation to transform texts into vectors, and contrastive learning to improve the representation that distinguishes the differences among similar texts. The paper presents a general framework for clustering similar texts that includes label representation, K-means, and self-organization map algorithms, with the objective of improving clustering performance using Accuracy (ACC) and Normalized Mutual Information (NMI) metrics. The proposed framework is evaluated experimentally using two real datasets. The results show that self-organization maps and K-means algorithms with Pre-trained language models outperform label representation algorithms for different datasets.

Sparse Depth-Guided Image Enhancement Using Incremental GP with Informative Point Selection.

We propose an online dehazing method with sparse depth priors using an incremental Gaussian Process (iGP). Conventional approaches focus on achieving single image dehazing by using multiple channels. In many robotics platforms, range measurements are directly available, except in a sparse form. This paper exploits direct and possibly sparse depth data in order to achieve efficient and effective dehazing that works for both color and grayscale images. The proposed algorithm is not limited to the channel information and works equally well for both color and gray images. However, efficient depth map estimations (from sparse depth priors) are additionally required. This paper focuses on a highly sparse depth prior for online dehazing. For efficient dehazing, we adopted iGP for incremental depth map estimation and dehazing. Incremental selection of the depth prior was conducted in an information-theoretic way by evaluating mutual information (MI) and other information-based metrics. As per updates, only the most informative depth prior was added, and haze-free images were reconstructed from the atmospheric scattering model with incrementally estimated depth. The proposed method was validated using different scenarios, color images under synthetic fog, real color, and grayscale haze indoors, outdoors, and underwater scenes.

Performance Analysis of Spectrum Sharing Radar in Multipath Environment

Radar based sensing and communication systems sharing a common spectrum have become a potential research problem in recent years due to spectrum scarcity. The spectrum sharing radar (SSR) is a new technology that uses the total available bandwidth (BW) for both radar based sensing and communication. Unlike traditional radar, the SSR divides the total available BW into radar-only and mixed-use bands. In a radar-only band, only radar sensor signals can be transmitted and received. In contrast, radar and communication signals can both be transmitted and received in the mixed-use band. Taking such BW sharing into account, this paper investigates the performance of SSR in an information-theoretic sense. To evaluate performance, mutual information (MI), spectral efficiency (SE) and capacity (C) metrics are used. Initially, this paper considered a clean environment (no multipath) in order to evaluate performance metrics in the mixed-use band with and without successive interference cancellation. Following that, this paper addresses the performance of BW allocation by allocating low to high BW in mixed-band. Furthermore, the performance metrics are extended to account for the multipath environment, and the same analogy as in a clean environment is used. In addition, the MI and SE of traditional radar system is taken into account when comparing the performance of SSR with and without the use of the SIC. Finally, MI and capacity results show that using the SIC scheme in a mixed-use band yields performance comparable to traditional radar and communication system. In terms of SE, the SSR with SIC scheme outperforms traditional radar and communication system.

Evaluating Template Generation Methods for Spatially Normalizing Down Syndrome Brain Magnetic Resonance Images

AbstractBackgroundMagnetic resonance (MR) image registration and segmentation algorithms commonly used in studies of neurodegeneration have a higher failure rate for Down syndrome (DS) populations due to anatomical and morphological differences. For example, the Desikan‐Killiany atlas used by FreeSurfer was defined using MR images from cognitively normal adults and patients with Alzheimer’s disease (AD). The aim of this work was to evaluate three template generation methods (or group‐wise image registration) for spatially matching DS MR images and generating a DS cohort‐specific template.Method138 scans of DS subjects from the NiAD consortium were employed. The three template generation methods evaluated were DARTEL, SHOOT, and SyGN. DARTEL and SHOOT are implemented in the SPM12 (v7771) software package and operate on tissue segmentation maps. SyGN is implemented in ANTs software and operates on MR images. The data processing pipeline is presented in Figure 1. Scan‐specific grey matter (GM) and white matter (WM) tissue probability maps were generated using SPM12’s Unified Segmentation method for DARTEL and SHOOT, and MR images were skull‐stripped for SyGN. To evaluate spatial matching, all MR images were warped to each template space using method‐specific deformation fields and an average MR image was generated specific to each template method. Structural similarity index measure (SSIM) and Normalized Mutual Information (NMI) metrics were calculated on a scan specific basis relative to the average MR image for each template method. Lower SSIM and NMI measures effectively reflect greater spatial variability across DS scans and therefore worse spatial matching performance.ResultThe DARTEL and SHOOT tissue map (GM and WM) templates and SyGN MR template are presented in Figure 2. Average MRs for each template method are presented in Figure 3. Consistent relationships were observed between the 3 template generation methods for the mean (SHOOT > SyGN > DARTEL) and standard deviation (SyGN < DARTEL < SHOOT) of SSIM and NMI (Figures 4 and 5, Table 1).ConclusionWith relatively high means and the lowest standard deviation, SyGN produced robust spatial matching across all scans. Future Down syndrome MRI and PET image processing will make use of the SyGN template.

The Sentinel 2 MSI Spectral Mixing Space

A composite spectral feature space is used to characterize the spectral mixing properties of Sentinel 2 Multispectral Instrument (MSI) spectra over a wide diversity of landscapes. Characterizing the linearity of spectral mixing and identifying bounding spectral endmembers allows the Substrate Vegetation Dark (SVD) spectral mixture model previously developed for the Landsat and MODIS sensors to be extended to the Sentinel 2 MSI sensors. The utility of the SVD model is its ability to represent a wide variety of landscapes in terms of the areal abundance of their most spectrally and physically distinct components. Combining the benefits of location-specific spectral mixture models with standardized spectral indices, the physically based SVD model offers simplicity, consistency, inclusivity and applicability for a wide variety of land cover mapping applications. In this study, a set of 110 image tiles compiled from spectral diversity hotspots worldwide provide a basis for this characterization, and for identification of spectral endmembers that span the feature space. The resulting spectral mixing space of these 13,000,000,000 spectra is effectively 3D, with 99% of variance in 3 low order principal component dimensions. Four physically distinct spectral mixing continua are identified: Snow:Firn:Ice, Reef:Water, Evaporite:Water and Substrate:Vegetation:Dark (water or shadow). The first 3 continua exhibit complex nonlinearities, but the geographically dominant Substrate:Vegetation:Dark (SVD) continuum is conspicuous in the linearity of its spectral mixing. Bounding endmember spectra are identified for the SVD continuum. In a subset of 80 landscapes, excluding the 3 nonlinear mixing continua (reefs, evaporites, cryosphere), a 3 endmember (SVD) linear mixture model produces endmember fraction estimates that represent 99% of modeled spectra with <6% RMS misfit. Two sets of SVD endmembers are identified for the Sentinel 2 MSI sensors, allowing Sentinel 2 spectra to be unmixed globally and compared across time and space. In light of the apparent disparity between the 11D spectral feature space and the statistically 3D spectral mixing space, the relative contribution of 11 Sentinel 2 MSI spectral bands to the information content of this space is quantified using both parametric (Pearson Correlation) and nonparametric (Mutual Information) metrics. Comparison of linear (principal component) and nonlinear (Uniform Manifold Approximation and Projection) projections of the SVD mixing space reveal both physically interpretable spectral mixing continua and geographically distinct spectral properties not resolved in the linear projection.

Web information mining and semantic analysis in heterogeneous unstructured text data using enhanced latent Dirichlet allocation

SummaryInformation mining and semantic analysis have gained significant attention over recent years to obtain appropriate information from unstructured data. Several approaches have been introduced for web information mining. However, the expected accuracy is not reached by these approaches. Therefore, hybrid fuzzy clustering and enhanced latent Dirichlet allocation (ELDA) are proposed for the accuracy increment in this work. The information clustering process is performed using the hybrid fuzzy clustering algorithm called fuzzy‐C‐medoids optimized using improved whale algorithm. The clustering procedure entails grouping data points into more similar clusters than data points from other clusters. Finally, the context of the text is recognized by analyzing the semantic information with ELDA, which offers a suitable index for accurate and fast data extraction. PYTHON tool is used to develop the proposed web mining model, and the simulation analysis of the proposed model is carried out using the BibTex dataset and compared with baseline models. The performance of the proposed method is evaluated using purity, normalized mutual information, accuracy, and precision metrics. Also, it is compared with different existing algorithms. Thus, the analysis of the results proved that the proposed approach achieves better outcomes than the existing approaches.

A quantitative characterization of the spatial distribution of brain metastases from breast cancer and respective molecular subtypes.

Brain metastases (BM) remain a significant cause of morbidity and mortality in breast cancer (BC) patients. Specific factors promoting the process of BM and predilection for selected neuro-anatomical regions remain unknown, yet may have major implications for prevention or treatment. Anatomical spatial distributions of BM from BC suggest a predominance of metastases in the hindbrain and cerebellum. Systematic approaches to quantifying BM location or location-based analyses based on molecular subtypes, however, remain largely unavailable. We analyzed stereotactic Cartesian coordinates derived from 134 patients undergoing gamma- knife radiosurgery (GKRS) for treatment of 407 breast cancer BMs to quantitatively study BM spatial distribution along principal component axes and by intrinsic molecular subtype (ER, PR, Herceptin). We used kernel density estimators (KDE) to highlight clustering and distribution regions in the brain, and we used the metric of mutual information (MI) to tease out subtle differences in the BM distributions associated with different molecular subtypes of BC. BM location maps according to vascular and anatomical distributions using Cartesian coordinates to aid in systematic classification of tumor locations were additionally developed. We corroborated that BC BMs show a consistent propensity to arise posteriorly and caudally, and that Her2+ tumors are relatively more likely to arise medially rather than laterally. To compare the distributions among varying BC molecular subtypes, the mutual information metric reveal that the ER-PR-Her2+ and ER-PR-Her2- subtypes show the smallest amount of mutual information and are most molecularly distinct. The kernel density contour plots show a propensity for triple negative BC to arise in more superiorly or cranially situated BMs. We present a novel and shareable workflow for characterizing and comparing spatial distributions of BM which may aid in identifying therapeutic or diagnostic targets and interactions with the tumor microenvironment. Further characterization of these patterns with larger multi-institutional data-sets may have major impacts on treatment or management of cancer patients.

SAMCOR: A robust and precise co-registration algorithm for brain CT and MR imaging

ObjectiveCurrent methods for co-registration of computed tomography (CT) and magnetic resonance (MR) brain imaging employ optimization-based cost functions of advanced statistical metrics to align datasets (e.g. mutual information, MI). Yet the inherent differences between these modalities still lead to occasional failures of registration. To address these limitations, we developed a novel approached termed sequence adaptive multimodal co-registration (SAMCOR), which is robust to intensity, feature, and scanner-based differences. MethodsSAMCOR was validated using 152 CT and MRI datasets from subjects implanted with intracranial electrodes for evaluation of medically-refractory epilepsy. Alignment outcomes were compared against five current open-source co-registration algorithms and a widely used commercial neurosurgical navigation system and classified as aligned/not-aligned relative to manually co-registered datasets. ResultsSAMCOR was the only method to align all 152 datasets (100%), while success rates for existing algorithms ranged from 3% to 82%. Given the time-intensive nature of manual verification, we further developed a metric to automatically quantify co-registration accuracy using Dice Similarity Coefficients (DSC). We applied a binary classification analysis to evaluate this DSC metric performance against traditional mutual information (MI) metrics for predicting alignment outcomes. The DSC metric demonstrated significantly higher positive and negative predictive values (0.93 and 0.97, respectively) than MI (0.79 and 0.87, respectively). The DSC classifier accuracy was 94.7% (vs 83.2% for MI). ConclusionsOur novel approaches for CT-MRI co-registration and quantitative alignment outcome analysis demonstrate significant advancements of the state-of-the-art for intermodal image co-registration with clinical relevance in patients with implanted intracranial devices.

An improved multi-modal joint segmentation and registration model based on Bhattacharyya distance measure

Joint image segmentation and registration of multi-modality images is a crucial step in the field of image prepossessing. The sensitivity of joint segmentation and registration models to noise is a significant challenge. During the registration process of multi-modal images, the similarity measure plays a vital role in measuring the results as a standard. Accordingly, an improved joint model for registering and segmenting multi-modality images is proposed, by utilising the Bhattacharyya distance measure to achieve improved noise robustness of the proposed model as compared to the existing model using the mutual information metric. The proposed model is applied to various medical and synthetic noisy images of multiple modalities. Moreover, the dataset images used in this study have been obtained from well-known, freely available BRATS 2015 and CHAOS datasets, where the proposed model produces satisfactory results as compared to the existing model. Experimental results show that the proposed model outperforms the existing model in terms of the Bhattacharyya distance measure in noisy images. Statistical analysis and comparison are performed through the relative reduction of the new distance measure, Dice similarity coefficient, Jaccard similarity coefficient and Hausdorff distance.

Biophysical Controls of Ecosystem‐Scale Methane Fluxes From a Subtropical Estuarine Mangrove: Multiscale, Nonlinearity, Asynchrony and Causality

AbstractMangroves are complex ecosystems subject to periodic and irregular changes in meteorology, tides, water quality and biological factors. Methane dynamics in mangroves and their dependence on biogeochemical controls across scales are poorly characterized using traditional parametric statistics. We measured the ecosystem‐scale methane flux (FCH4) using an eddy covariance system in a subtropical estuarine mangrove for two years. A combination of nonparametric statistical approaches, including singular spectrum analysis and information theory, was used to isolate multiscale FCH4 variations and explore their variability and dominant controls at different time scales. FCH4 exhibited the largest variability at the diel scale (57%), followed by seasonal (25%) and multiday (18%) scales. Mutual information metrics suggested that variation in FCH4 was dominantly coupled to plant activities through synchronous processes at the diel scale. Moreover, fluctuations in atmospheric turbulence and soil temperature dominated multiday variation in FCH4 through asynchronous processes. Seasonally, soil temperature and water salinity were dominant variables leading to changes in FCH4 with a time lag of 12 and 17 days, respectively. Weak links between FCH4 and tidal heights were found across all the studied time scales. Transfer entropy metrics indicated that a super typhoon event substantially weakened causal links between biophysical variables and FCH4 by causing severe defoliation and disturbing the soil microbial community. The typhoon also temporarily decoupled the causal link between FCH4 and temperature. These findings call for the inclusion of the properties of scale emergence, nonlinearity, asynchrony and causality for comprehensively understanding and accurately predicting the interactions between mangrove FCH4 and its biophysical controls.

Multi-Server Weakly-Private Information Retrieval

Private information retrieval (PIR) protocols ensure that a user can download a file from a database without revealing any information on the identity of the requested file to the servers storing the database. While existing protocols strictly impose that no information is leaked on the file&#x2019;s identity, this work initiates the study of the tradeoffs that can be achieved by relaxing the perfect privacy requirement. We refer to such protocols as weakly-private information retrieval (WPIR) protocols. In particular, for the case of multiple noncolluding replicated servers, we study how the download rate, the upload cost, and the access complexity can be improved when relaxing the perfect privacy constraint. To quantify the information leakage on the requested file&#x2019;s identity we consider mutual information (MI), worst-case information leakage, and maximal leakage (MaxL). We present two WPIR schemes, denoted by Scheme A and Scheme B, based on two recent PIR protocols and show that the download rate of the former can be optimized by solving a convex optimization problem. We also show that Scheme A achieves an improved download rate compared to the recently proposed scheme by Samy <i>et al.</i> under the so-called <inline-formula> <tex-math notation="LaTeX">$\epsilon $ </tex-math></inline-formula>-privacy metric. Additionally, a family of schemes based on partitioning is presented. Moreover, we provide an information-theoretic converse bound for the maximum possible download rate for the MI and MaxL privacy metrics under a practical restriction on the alphabet size of queries and answers. For two servers and two files, the bound is tight under the MaxL metric, which settles the WPIR capacity in this particular case. Finally, we compare the performance of the proposed schemes and their gap to the converse bound.

Locally Expansion Directed Community Detection Method Considering Node Leadership

In recent years, community structure has become a research hotspot in complex networks, and there are currently many excellent methods in the field of community detection, such as label propagation algorithms, LFM community detection algorithms, etc. These algorithms primarily investigate undirected networks, but the linking relationships between nodes in real complex networks are mostly directed and asymmetric, such as social networks. Therefore it is imperative to perform community detection on directed networks. We proposed a locally expansion community detection algorithm (DLE-NL) that can be applied to directed networks. There are four essential parts of DLE-NL algorithm: Firstly, the node cross-leadership index is proposed to select seed nodes considering local and global factors and directionality; secondly, seed node confidants are chosen to form the initial community using node following degree, and the core of the community is precisely mined; then node aggregation is carried out considering the node affiliation to the community with the initial community as the core. Finally, the initial reconstruction and optimization of the community are carried out by removing communities with less than three nodes, correcting the affiliation of nodes to the community based on the community directed-fitness function and the gain of the derivative, assigning free nodes and merging communities with high similarity. The experimental results show that the method performs well in the directed modularity and normalized mutual information metrics, while the community detection results are more stable.

Iterative Outlier Removal Clustering Based Time-Frequency-Spatial Feature Selection for Binary EEG Motor Imagery Decoding

Electroencephalography (EEG)-based motor imagery (MI) brain-computer interface (BCI) is a bridge in the instruments of rehabilitation and motor assistance field to control external assist devices without external stimulation. Feature extraction (FE) is a key to improving the performance of EEG MI decoding. Existing FE methods usually extract single-domain features such as spatial, frequency features, or dual-domain features such as time–frequency features. However, the multidomains features which describe the intent more comprehensively are not simultaneously extracted. Therefore, inspired by short-time Fourier transform (STFT) and common spatial pattern (CSP) FE methods, a frequency-spatial-temporal multidomain (FSTMD) FE method is proposed. To solve the problem of the high dimensions of the FSTMD features, a feature selection (FS) method is designed. In the proposed FSTMD FE process, the STFT is employed to extract the time–frequency domain features first. Then, the time–frequency features of each channel are cascaded. After that, the cascaded features are divided according to time segments. At last, the covariance transform is performed for each time segment. Since the proposed features have a one-to-one correspondence with frequency, channel, and time, the extracted features have good interpretability. The proposed FS method selects the features in the frequency-spatial domain of each time segment based on mutual information and correlation metrics first and then selects the key time segment based on clustering with outlier removal. The performance of the proposed feature is verified in BCI Competition IV Dataset IIa and IIb. Compared to the CSP feature in the only spatial domain, the proposed feature has significantly lower deviation and slightly higher accuracy. The average recognition accuracy of the proposed MI decoding framework is 85.1% on the dataset IIb session 3, which is competitive compared with the state-of-the-art methods.