Grapheme-to-phoneme (G2P) conversion is critical in speech processing, particularly for applications like speech synthesis. G2P systems must possess linguistic understanding and contextual awareness of languages with homograph words and context-dependent phonemes. Large language models (LLMs) have recently demonstrated significant potential in various language tasks, suggesting that their phonetic knowledge could be leveraged for G2P. In this paper, we evaluate the performance of LLMs in G2P conversion and introduce prompting and post-processing methods that enhance LLM outputs without additional training or labeled data. We also present a benchmarking dataset designed to assess G2P performance on sentence-level phonetic challenges of the Persian language. Our results show that by applying the proposed methods, LLMs can outperform traditional G2P tools, even in an underrepresented language like Persian, highlighting the potential of developing LLM-aided G2P systems.

Off-policy learning and evaluation leverage logged bandit feedback datasets, which contain context, action, propensity score, and feedback for each data point. These scenarios face significant challenges due to high variance and poor performance with low-quality propensity scores and heavy-tailed reward distributions. We address these issues by introducing a novel estimator based on the log-sum-exponential (LSE) operator, which outperforms traditional inverse propensity score estimators. Our LSE estimator demonstrates variance reduction and robustness under heavy-tailed conditions. For off-policy evaluation, we derive upper bounds on the estimator's bias and variance. In the off-policy learning scenario, we establish bounds on the regret—the performance gap between our LSE estimator and the optimal policy—assuming bounded $(1+\epsilon)$-th moment of weighted reward. Notably, we achieve a convergence rate of $O(n^{-\epsilon/(1+\epsilon)})$ for the regret bounds, where $\epsilon\in[0,1]$ and $n$ is the size of logged bandit feedback dataset. Theoretical analysis is complemented by comprehensive empirical evaluations in both off-policy learning and evaluation scenarios, confirming the practical advantages of our approach. The code for our estimator is available at the following link: https://github.com/armin-behnamnia/lse-offpolicy-learning .

Mahta Fetrat Qharabagh, Zahra Dehghanian, Hamid R. Rabiee. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

This paper studies how to integrate historical control data with experimental data to enhance A/B testing, while addressing the distributional shift between historical and experimental datasets. We propose a pessimistic data integration method that combines two causal effect estimators constructed based on experimental and historical datasets. Our main idea is to conceptualize the weight function for this combination as a policy so that existing pessimistic policy learning algorithms are applicable to learn the optimal weight that minimizes the resulting weighted estimator's mean squared error. Additionally, we conduct comprehensive theoretical and empirical analyses to compare our method against various baseline estimators across five scenarios. Both our theoretical and numerical findings demonstrate that the proposed estimator achieves near-optimal performance across all scenarios.

Decentralized strategies are of interest for learning from large-scale data over networks. This paper studies learning over a network of geographically distributed nodes/agents subject to quantization. Each node possesses a private local cost function, collectively contributing to a global cost function, which the considered methodology aims to minimize. In contrast to many existing papers, the information exchange among nodes is log-quantized to address limited network-bandwidth in practical situations. We consider a first-order computationally efficient distributed optimization algorithm (with no extra inner consensus loop) that leverages node-level gradient correction based on local data and network-level gradient aggregation only over nearby nodes. This method only requires balanced networks with no need for stochastic weight design. It can handle log-scale quantized data exchange over possibly time-varying and switching network setups. We study convergence over both structured networks (for example, training over data-centers) and ad-hoc multi-agent networks (for example, training over dynamic robotic networks). Through experimental validation, we show that (i) structured networks generally result in a smaller optimality gap, and (ii) log-scale quantization leads to a smaller optimality gap compared to uniform quantization.

This paper proposes a novel interpretation technique to explain the behavior of structured output models, which simultaneously learn mappings between an input vector and a set of output variables. As a result of the complex relationships between the computational path of output variables in structured models, a feature may impact an output value via other output variables. We focus on one of the outputs as the target and try to find the most important features adopted by the structured model to decide on the target in each locality of the input space. We consider an arbitrary structured output model available as a black-box and argue that considering correlations among output variables can improve explanation quality. The goal is to train a function as an interpreter for the target output variable over the input space. We introduce an energy-based training process for the interpreter function, which effectively considers the structural information incorporated into the model to be explained. The proposed method's effectiveness is confirmed using various simulated and real data sets.

MOTIVATION: Drug repurposing is a viable solution for reducing the time and cost associated with drug development. However, thus far, the proposed drug repurposing approaches still need to meet expectations. Therefore, it is crucial to offer a systematic approach for drug repurposing to achieve cost savings and enhance human lives. In recent years, using biological network-based methods for drug repurposing has generated promising results. Nevertheless, these methods have limitations. Primarily, the scope of these methods is generally limited concerning the size and variety of data they can effectively handle. Another issue arises from the treatment of heterogeneous data, which needs to be addressed or converted into homogeneous data, leading to a loss of information. A significant drawback is that most of these approaches lack end-to-end functionality, necessitating manual implementation and expert knowledge in certain stages. RESULTS: We propose a new solution, Heterogeneous Graph Transformer for Drug Repurposing (HGTDR), to address the challenges associated with drug repurposing. HGTDR is a three-step approach for knowledge graph-based drug repurposing: (1) constructing a heterogeneous knowledge graph, (2) utilizing a heterogeneous graph transformer network, and (3) computing relationship scores using a fully connected network. By leveraging HGTDR, users gain the ability to manipulate input graphs, extract information from diverse entities, and obtain their desired output. In the evaluation step, we demonstrate that HGTDR performs comparably to previous methods. Furthermore, we review medical studies to validate our method's top 10 drug repurposing suggestions, which have exhibited promising results. We also demonstrated HGTDR's capability to predict other types of relations through numerical and experimental validation, such as drug-protein and disease-protein inter-relations. AVAILABILITY AND IMPLEMENTATION: The source code and data are available at https://github.com/bcb-sut/HGTDR and http://git.dml.ir/BCB/HGTDR.

BACKGROUND: Copy number variants (CNVs) have become increasingly instrumental in understanding the etiology of all diseases and phenotypes, including Neurocognitive Disorders (NDs). Among the well-established regions associated with ND are small parts of chromosome 16 deletions (16p11.2) and chromosome 15 duplications (15q3). Various methods have been developed to identify associations between CNVs and diseases of interest. The majority of methods are based on statistical inference techniques. However, due to the multi-dimensional nature of the features of the CNVs, these methods are still immature. The other aspect is that regions discovered by different methods are large, while the causative regions may be much smaller. RESULTS: In this study, we propose a regularized deep learning model to select causal regions for the target disease. With the help of the proximal [20] gradient descent algorithm, the model utilizes the group LASSO concept and embraces a deep learning model in a sparsity framework. We perform the CNV analysis for 74,811 individuals with three types of brain disorders, autism spectrum disorder (ASD), schizophrenia (SCZ), and developmental delay (DD), and also perform cumulative analysis to discover the regions that are common among the NDs. The brain expression of genes associated with diseases has increased by an average of 20 percent, and genes with homologs in mice that cause nervous system phenotypes have increased by 18 percent (on average). The DECIPHER data source also seeks other phenotypes connected to the detected regions alongside gene ontology analysis. The target diseases are correlated with some unexplored regions, such as deletions on 1q21.1 and 1q21.2 (for ASD), deletions on 20q12 (for SCZ), and duplications on 8p23.3 (for DD). Furthermore, our method is compared with other machine learning algorithms. CONCLUSIONS: Our model effectively identifies regions associated with phenotypic traits using regularized deep learning. Rather than attempting to analyze the whole genome, CNVDeep allows us to focus only on the causative regions of disease.

The genetic etiology of brain disorders is highly heterogeneous, characterized by abnormalities in the development of the central nervous system that lead to diminished physical or intellectual capabilities. The process of determining which gene drives disease, known as "gene prioritization," is not entirely understood. Genome-wide searches for gene-disease associations are still underdeveloped due to reliance on previous discoveries and evidence sources with false positive or negative relations. This paper introduces DeepGenePrior, a model based on deep neural networks that prioritizes candidate genes in genetic diseases. Using the well-studied Variational AutoEncoder (VAE), we developed a score to measure the impact of genes on target diseases. Unlike other methods that use prior data to select candidate genes, based on the "guilt by association" principle and auxiliary data sources like protein networks, our study exclusively employs copy number variants (CNVs) for gene prioritization. By analyzing CNVs from 74,811 individuals with autism, schizophrenia, and developmental delay, we identified genes that best distinguish cases from controls. Our findings indicate a 12% increase in fold enrichment in brain-expressed genes compared to previous studies and a 15% increase in genes associated with mouse nervous system phenotypes. Furthermore, we identified common deletions in ZDHHC8, DGCR5, and CATG00000022283 among the top genes related to all three disorders, suggesting a common etiology among these clinically distinct conditions. DeepGenePrior is publicly available online at http://git.dml.ir/z_rahaie/DGP to address obstacles in existing gene prioritization studies identifying candidate genes.

Access to complete data in large-scale networks is often infeasible. Therefore, the problem of missing data is a crucial and unavoidable issue in the analysis and modeling of real-world social networks. However, most of the research on different aspects of social networks does not consider this limitation. One effective way to solve this problem is to recover the missing data as a pre-processing step. In this paper, a model is learned from partially observed data to infer unobserved diffusion and structure networks. To jointly discover omitted diffusion activities and hidden network structures, we develop a probabilistic generative model called “DiffStru.” The interrelations among links of nodes and cascade processes are utilized in the proposed method via learning coupled with low-dimensional latent factors. Besides inferring unseen data, latent factors such as community detection may also aid in network classification problems. We tested different missing data scenarios on simulated independent cascades over LFR networks and real datasets, including Twitter and Memetracker. Experiments on these synthetic and real-world datasets show that the proposed method successfully detects invisible social behaviors, predicts links, and identifies latent features.

Missing data is unavoidable in graphs, which can significantly affect the accuracy of downstream tasks. Many methods have been proposed to mitigate missing data in partially observed graphs. Most of these approaches assume they have complete access to graph nodes and only focus on recovering missing links, while in practice a part of the graph nodes can also be out of access. This work presents Deep Missing Node Predictor (DMNP), a novel deep learning-based approach to recovering missing nodes in partly observed graphs. Our proposed approach does not rely on additional information that in many cases does not exist. We compare our model with graph completion and deep graph generation baselines. The experimental results show that the DMNP model outperforms previous state-of-the-art approaches.

In this paper we propose a novel hierarchical Bayesian model for sparse regression problem to use in semi-supervised hyperspectral unmixing which assumes the signal recorded in each hyperspectral pixel is a linear combination of members of the spectral library contaminated by an additive Gaussian noise. To effectively utilizing the spatial correlation between neighboring pixels during the unmixing process, we exploit a Markov random field to simultaneously group pixels to clusters which are associated to regions with homogeneous mixtures in a natural scene. We assume Sparse fractional abundances of members of a cluster to be generated from an exponential distribution with the same rate parameter. We show that our method is able to detect unconnected regions which have similar mixtures. Experiments on synthetic and real hyperspectral images confirm the superiority of the proposed method compared to alternatives.

BACKGROUND: Colorectal cancer (CRC) is one of the leading causes of cancer-related deaths worldwide. Recent studies have observed causative mutations in susceptible genes related to colorectal cancer in 10 to 15% of the patients. This highlights the importance of identifying mutations for early detection of this cancer for more effective treatments among high risk individuals. Mutation is considered as the key point in cancer research. Many studies have performed cancer subtyping based on the type of frequently mutated genes, or the proportion of mutational processes. However, to the best of our knowledge, combination of these features has never been used together for this task. This highlights the potential to introduce better and more inclusive subtype classification approaches using wider range of related features to enable biomarker discovery and thus inform drug development for CRC. RESULTS: In this study, we develop a new pipeline based on a novel concept called 'gene-motif', which merges mutated gene information with tri-nucleotide motif of mutated sites, for colorectal cancer subtype identification. We apply our pipeline to the International Cancer Genome Consortium (ICGC) CRC samples and identify, for the first time, 3131 gene-motif combinations that are significantly mutated in 536 ICGC colorectal cancer samples. Using these features, we identify seven CRC subtypes with distinguishable phenotypes and biomarkers, including unique cancer related signaling pathways, in which for most of them targeted treatment options are currently available. Interestingly, we also identify several genes that are mutated in multiple subtypes but with unique sequence contexts. CONCLUSION: Our results highlight the importance of considering both the mutation type and mutated genes in identification of cancer subtypes and cancer biomarkers. The new CRC subtypes presented in this study demonstrates distinguished phenotypic properties which can be effectively used to develop new treatments. By knowing the genes and phenotypes associated with the subtypes, a personalized treatment plan can be developed that considers the specific phenotypes associated with their genomic lesion.

BACKGROUND: The advent of high throughput sequencing has enabled researchers to systematically evaluate the genetic variations in cancer, identifying many cancer-associated genes. Although cancers in the same tissue are widely categorized in the same group, they demonstrate many differences concerning their mutational profiles. Hence, there is no definitive treatment for most cancer types. This reveals the importance of developing new pipelines to identify cancer-associated genes accurately and re-classify patients with similar mutational profiles. Classification of cancer patients with similar mutational profiles may help discover subtypes of cancer patients who might benefit from specific treatment types. RESULTS: In this study, we propose a new machine learning pipeline to identify protein-coding genes mutated in many samples to identify cancer subtypes. We apply our pipeline to 12,270 samples collected from the international cancer genome consortium, covering 19 cancer types. As a result, we identify 17 different cancer subtypes. Comprehensive phenotypic and genotypic analysis indicates distinguishable properties, including unique cancer-related signaling pathways. CONCLUSIONS: This new subtyping approach offers a novel opportunity for cancer drug development based on the mutational profile of patients. Additionally, we analyze the mutational signatures for samples in each subtype, which provides important insight into their active molecular mechanisms. Some of the pathways we identified in most subtypes, including the cell cycle and the Axon guidance pathways, are frequently observed in cancer disease. Interestingly, we also identified several mutated genes and different rates of mutation in multiple cancer subtypes. In addition, our study on "gene-motif" suggests the importance of considering both the context of the mutations and mutational processes in identifying cancer-associated genes. The source codes for our proposed clustering pipeline and analysis are publicly available at: https://github.com/bcb-sut/Pan-Cancer .

BACKGROUND: Several types of RNA in the cell are usually involved in biological processes with multiple functions. Coding RNAs code for proteins while non-coding RNAs regulate gene expression. Some single-strand RNAs can create a circular shape via the back splicing process and convert into a new type called circular RNA (circRNA). circRNAs are among the essential non-coding RNAs in the cell that involve multiple disorders. One of the critical functions of circRNAs is to regulate the expression of other genes through sponging micro RNAs (miRNAs) in diseases. This mechanism, known as the competing endogenous RNA (ceRNA) hypothesis, and additional information obtained from biological datasets can be used by computational approaches to predict novel associations between disease and circRNAs. RESULTS: We applied multiple classifiers to validate the extracted features from the heterogeneous network and selected the most appropriate one based on some evaluation criteria. Then, the XGBoost is utilized in our pipeline to generate a novel approach, called CircWalk, to predict CircRNA-Disease associations. Our results demonstrate that CircWalk has reasonable accuracy and AUC compared with other state-of-the-art algorithms. We also use CircWalk to predict novel circRNAs associated with lung, gastric, and colorectal cancers as a case study. The results show that our approach can accurately detect novel circRNAs related to these diseases. CONCLUSIONS: Considering the ceRNA hypothesis, we integrate multiple resources to construct a heterogeneous network from circRNAs, mRNAs, miRNAs, and diseases. Next, the DeepWalk algorithm is applied to the network to extract feature vectors for circRNAs and diseases. The extracted features are used to learn a classifier and generate a model to predict novel CircRNA-Disease associations. Our approach uses the concept of the ceRNA hypothesis and the miRNA sponge effect of circRNAs to predict their associations with diseases. Our results show that this outlook could help identify CircRNA-Disease associations more accurately.

Hi-C is a genome-wide chromosome conformation capture technology that detects interactions between pairs of genomic regions and exploits higher order chromatin structures. Conceptually Hi-C data counts interaction frequencies between every position in the genome and every other position. Biologically functional interactions are expected to occur more frequently than transient background and artefactual interactions. To identify biologically relevant interactions, several background models that take biases such as distance, GC content and mappability into account have been proposed. Here we introduce MaxHiC, a background correction tool that deals with these complex biases and robustly identifies statistically significant interactions in both Hi-C and capture Hi-C experiments. MaxHiC uses a negative binomial distribution model and a maximum likelihood technique to correct biases in both Hi-C and capture Hi-C libraries. We systematically benchmark MaxHiC against major Hi-C background correction tools including Hi-C significant interaction callers (SIC) and Hi-C loop callers using published Hi-C, capture Hi-C, and Micro-C datasets. Our results demonstrate that 1) Interacting regions identified by MaxHiC have significantly greater levels of overlap with known regulatory features (e.g. active chromatin histone marks, CTCF binding sites, DNase sensitivity) and also disease-associated genome-wide association SNPs than those identified by currently existing models, 2) the pairs of interacting regions are more likely to be linked by eQTL pairs and 3) more likely to link known regulatory features including known functional enhancer-promoter pairs validated by CRISPRi than any of the existing methods. We also demonstrate that interactions between different genomic region types have distinct distance distributions only revealed by MaxHiC. MaxHiC is publicly available as a python package for the analysis of Hi-C, capture Hi-C and Micro-C data.

[This corrects the article DOI: 10.1371/journal.pcbi.1010241.].

User churn is an important issue in online services that threatens the health and profitability of services. Most of the previous works on churn prediction convert the problem into a binary classification task where the users are labeled as churned and non-churned. More recently, some works have tried to convert the user churn prediction problem into the prediction of user return time. In this approach which is more realistic in real world online services, at each time-step the model predicts the user return time instead of predicting a churn label. However, the previous works in this category suffer from lack of generality and require high computational complexity. In this paper, we introduceChOracle, an oracle that predicts the user churn by modeling the user return times to service by utilizing a combination of Temporal Point Processes and Recurrent Neural Networks. Moreover, we incorporate latent variables into the proposed recurrent neural network to model the latent user loyalty to the system. We also develop an efficient approximate variational inference algorithm for learning parameters of the proposed RNN by using back propagation through time. Finally, we demonstrate the superior performance of ChOracle on a wide variety of real world datasets.

Unsupervised representation learning has proved to be a critical component of anomaly detection/localization in images. The challenges to learn such a representation are two-fold. Firstly, the sample size is not often large enough to learn a rich generalizable representation through conventional techniques. Secondly, while only normal samples are available at training, the learned features should be discriminative of normal and anomalous samples. Here, we propose to use the "distillation" of features at various layers of an expert network, which is pre-trained on ImageNet, into a simpler cloner network to tackle both issues. We detect and localize anomalies using the discrepancy between the expert and cloner networks’ intermediate activation values given an input sample. We show that considering multiple intermediate hints in distillation leads to better exploitation of the expert’s knowledge and a more distinctive discrepancy between the two networks, compared to utilizing only the last layer activation values. Notably, previous methods either fail in precise anomaly localization or need expensive region-based training. In contrast, with no need for any special or intensive training procedure, we incorporate interpretability algorithms in our novel framework to localize anomalous regions. Despite the striking difference between some test datasets and ImageNet, we achieve competitive or significantly superior results compared to SOTA on MNIST, F-MNIST, CIFAR-10, MVTecAD, Retinal-OCT, and two other medical datasets on both anomaly detection and localization.

MOTIVATION: Single-cell RNA-sequencing (scRNA-seq) offers the opportunity to dissect heterogeneous cellular compositions and interrogate the cell-type-specific gene expression patterns across diverse conditions. However, batch effects such as laboratory conditions and individual-variability hinder their usage in cross-condition designs. RESULTS: Here, we present a single-cell Generative Adversarial Network (scGAN) to simultaneously acquire patterns from raw data while minimizing the confounding effect driven by technical artifacts or other factors inherent to the data. Specifically, scGAN models the data likelihood of the raw scRNA-seq counts by projecting each cell onto a latent embedding. Meanwhile, scGAN attempts to minimize the correlation between the latent embeddings and the batch labels across all cells. We demonstrate scGAN on three public scRNA-seq datasets and show that our method confers superior performance over the state-of-the-art methods in forming clusters of known cell types and identifying known psychiatric genes that are associated with major depressive disorder. AVAILABILITYAND IMPLEMENTATION: The scGAN code and the information for the public scRNA-seq datasets are available at https://github.com/li-lab-mcgill/singlecell-deepfeature. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Internet-of-Things (IoT) devices and applications are being deployed in our homes and workplaces. These devices often rely on continuous data collection to feed machine learning models. However, this approach introduces several privacy and efficiency challenges, as the service operator can perform unwanted inferences on the available data. Recently, advances in edge processing have paved the way for more efficient, and private, data processing at the source for simple tasks and lighter models, though they remain a challenge for larger and more complicated models. In this article, we present a hybrid approach for breaking down large, complex deep neural networks for cooperative, and privacy-preserving analytics. To this end, instead of performing the whole operation on the cloud, we let an IoT device to run the initial layers of the neural network, and then send the output to the cloud to feed the remaining layers and produce the final result. In order to ensure that the user's device contains no extra information except what is necessary for the main task and preventing any secondary inference on the data, we introduce Siamese fine-tuning. We evaluate the privacy benefits of this approach based on the information exposed to the cloud service. We also assess the local inference cost of different layers on a modern handset. Our evaluations show that by using Siamese fine-tuning and at a small processing cost, we can greatly reduce the level of unnecessary, potentially sensitive information in the personal data, thus achieving the desired tradeoff between utility, privacy, and performance.

Poisson Factorization (PF) is the gold standard framework for recommendation systems with implicit feedback whose variants show state-of-the-art performance on real-world recommendation tasks. However, they do not explicitly take into account the temporal behavior of users which is essential to recommend the right item to the right user at the right time. In this paper, we introduce Recurrent Poisson Factorization (RPF) framework that generalizes the classical PF methods by utilizing a Poisson process for modeling the implicit feedback. RPF treats time as a natural constituent of the model, and takes important factors for recommendation into consideration to provide a rich family of time-sensitive factorization models. They include Hierarchical RPFthat captures the consumption heterogeneity among users and items, Dynamic RPF that handles dynamic user preferences and item specifications, Social RPF that models the social-aspect of product adoption, Item-Item RPFthat considers the inter-item correlations, and eXtended Item-Item RPF that utilizes items' metadata to better infer the correlation among engagement patterns of users with items. We also develop an efficient variational algorithm for approximate inference that scales up to massive datasets. We demonstrate RPF's superior performance over many state-of-the-art methods on synthetic dataset, and wide variety of large scale real-world datasets.

We present and evaluate Deep Private-Feature Extractor (DPFE), a deep model which is trained and evaluated based on information theoretic constraints. Using the selective exchange of information between a user's device and a service provider, DPFE enables the user to prevent certain sensitive information from being shared with a service provider, while allowing them to extract approved information using their model. We introduce and utilize the log-rank privacy, a novel measure to assess the effectiveness of DPFE in removing sensitive information and compare different models based on their accuracy-privacy trade-off. We then implement and evaluate the performance of DPFEon smartphones to understand its complexity, resource demands, and efficiency trade-offs. Our results on benchmark image datasets demonstrate that under moderate resource utilization, DPFE can achieve high accuracy for primary tasks while preserving the privacy of sensitive information.

Graph embedding is an important approach for graph analysis tasks such as node classification and link prediction. The goal of graph embedding is to find a low dimensional representation of graph nodes that preserves the graph information. Recent methods like Graph Convolutional Network (GCN) try to consider node attributes (if available) besides node relations and learn node embeddings for unsupervised and semi-supervised tasks on graphs. On the other hand, multi-layer graph analysis has been received attention recently. However, the existing methods for multi-layer graph embedding cannot incorporate all available information (like node attributes). Moreover, most of them consider either type of nodes or type of edges, and they do not treat within and between layer edges differently. In this paper, we propose a method called MGCN that utilizes the GCN for multi-layer graphs. MGCN embeds nodes of multi-layer graphs using both within and between layers relations and nodes attributes. We evaluate our method on the semi-supervised node classification task. Experimental results demonstrate the superiority of the proposed method to other multi-layer and single-layer competitors and also show the positive effect of using cross-layer edges.

Differentiating between real and fake news propagation through online social networks is an important issue in many applications. The time gap between the news release time and detection of its label is a significant step towards broadcasting the real information and avoiding the fake. Therefore, one of the challenging tasks in this area is to identify fake and real news in early stages of propagation. However, there is a tradeoff between minimizing the time gap and maximizing accuracy. Despite recent efforts in detection of fake news, there has been no significant work that explicitly incorporates early detection in its model. The proposed method utilizes recurrent neural networks with a novel loss function, and a new stopping rule. Experiments on real datasets demonstrate the effectiveness of our model both in terms of early labelling and accuracy, compared to the state of the art baseline and models.

Privacy-preserving data release is about disclosing information about useful data while retaining the privacy of sensitive data. Assuming that the sensitive data is threatened by a brute-force adversary, we define Guessing Leakage as a measure of privacy, based on the concept of guessing. After investigating the properties of this measure, we derive the optimal utility-privacy trade-off via a linear program with any f-information adopted as the utility measure, and show that the optimal utility is a concave and piece-wise linear function of the privacy-leakage budget.

Online social networks, World Wide Web, media, and technological networks, and other types of so-called information networks are ubiquitous nowadays. These information networks are inherently heterogeneous and dynamic. They are heterogeneous as they consist of multi-typed objects and relations, and they are dynamic as they are constantly evolving over time. One of the challenging issues in such heterogeneous and dynamic environments is to forecast those relationships in the network that will appear in the future. In this article, we try to solve the problem of continuous-time relationship prediction in dynamic and heterogeneous information networks. This implies predicting the time it takes for a relationship to appear in the future, given its features that have been extracted by considering both heterogeneity and temporal dynamics of the underlying network. To this end, we first introduce a feature extraction framework that combines the power of meta-path-based modeling and recurrent neural networks to effectively extract features suitable for relationship prediction regarding heterogeneity and dynamicity of the networks. Next, we propose a supervised non-parametric approach, called Non-Parametric Generalized Linear Model (Np-Glm), which infers the hidden underlying probability distribution of the relationship building time given its features. We then present a learning algorithm to train Np-Glm and an inference method to answer time-related queries. Extensive experiments conducted on synthetic data and three real-world datasets, namely Delicious, MovieLens, and DBLP, demonstrate the effectiveness of Np-Glm in solving continuous-time relationship prediction problem vis-à-vis competitive baselines.

In this letter, we discuss cost optimization of sensor networks monitoring structurally full-rank systems under distributed observability constraint. Using structured systems theory, the problem is relaxed into two subproblems: first, sensing cost optimization; and second, networking cost optimization. Both problems are reformulated as combinatorial optimization problems. The sensing cost optimization is shown to have a polynomial-order solution. The networking cost optimization is shown to be NP-hard in general, but has a polynomial-order solution under specific conditions. A 2-approximation polynomial-order relaxation is provided for general networking cost optimization, which is applicable in large-scale system monitoring.

User modeling plays an important role in delivering customized web services to the users and improving their engagement. However, most user models in the literature do not explicitly consider the temporal behavior of users. More recently, continuous-time user modeling has gained considerable attention and many user behavior models have been proposed based on temporal point processes. However, typical point process-based models often considered the impact of peer influence and content on the user participation and neglected other factors. Gamification elements are among those factors that are neglected, while they have a strong impact on user participation in online services. In this article, we propose interdependent multi-dimensional temporal point processes that capture the impact of badges on user participation besides the peer influence and content factors. We extend the proposed processes to model user actions over the community-based question and answering websites, and propose an inference algorithm based on Variational-Expectation Maximization that can efficiently learn the model parameters. Extensive experiments on both synthetic and real data gathered from Stack Overflow show that our inference algorithm learns the parameters efficiently and the proposed method can better predict the user behavior compared to the alternatives.

Community detection in social networks has become a popular topic of research during the last decade. There exist a variety of algorithms for modularizing the network graph into different communities. However, they mostly assume that partial or complete information of the network graphs are available that is not feasible in many cases. In this article, we focus on detecting communities by exploiting their diffusion information. To this end, we utilize the Conditional Random Fields (CRF) to discover the community structures. The proposed method, community diffusion (CoDi), does not require any prior knowledge about the network structure or specific properties of communities. Furthermore, in contrast to the structure-based community detection methods, this method is able to identify the hidden communities. The experimental results indicate considerable improvements in detecting communities based on accuracy, scalability, and real cascade information measures.

In real world social networks, there are multiple cascades which are rarely independent. They usually compete or cooperate with each other. Motivated by the reinforcement theory in sociology we leverage the fact that adoption of a user to any behavior is modeled by the aggregation of behaviors of its neighbors. We use a multidimensional marked Hawkes process to model users product adoption and consequently spread of cascades in social networks. The resulting inference problem is proved to be convex and is solved in parallel by using the barrier method. The advantage of the proposed model is twofold; it models correlated cascades and also learns the latent diffusion network. Experimental results on synthetic and two real datasets gathered from Twitter, URL shortening and music streaming services, illustrate the superior performance of the proposed model over the alternatives.

Over-the-top (OTT) video streaming services have been growing rapidly in the last decade, and thus increasing Quality-of-Experience (QoE) of end users is of great interest in emerging services. Although, numerous subjective studies have been conducted to investigate the impact of video presentation quality and playback interruptions, understanding the interactions between impairment types is still an open problem. In this work, we develop a streaming video dataset that contains compressed videos with different distortion levels as well as playback stalling events. Then, a subjective user study is performed to measure the QoE of these videos. The results of our experiment reveal strong dependency between video presentation quality and playback interruptions that provides useful insight for designing QoE models in video streaming.

Poisson factorization is a probabilistic model of users and items for recommendation systems, where the so-called implicit consumer data is modeled by a factorized Poisson distribution. There are many variants of Poisson factorization methods who show state-of-the-art performance on real-world recommendation tasks. However, most of them do not explicitly take into account the temporal behavior and the recurrent activities of users which is essential to recommend the right item to the right user at the right time. In this paper, we introduce Recurrent Poisson Factorization (RPF) framework that generalizes the classical PF methods by utilizing a Poisson process for modeling the implicit feedback. RPF treats time as a natural constituent of the model and brings to the table a rich family of time-sensitive factorization models. To elaborate, we instantiate several variants of RPF who are capable of handling dynamic user preferences and item specification (DRPF), modeling the social-aspect of product adoption (SRPF), and capturing the consumption heterogeneity among users and items (HRPF). We also develop a variational algorithm for approximate posterior inference that scales up to massive data sets. Furthermore, we demonstrate RPF's superior performance over many state-of-the-art methods on synthetic dataset, and large scale real-world datasets on music streaming logs, and user-item interactions in M-Commerce platforms.

Users in social networks whose posts stay at the top of their followers' feeds the longest time are more likely to be noticed. Can we design an online algorithm to help them decide when to post to stay at the top? In this paper, we address this question as a novel optimal control problem for jump stochastic differential equations. For a wide variety of feed dynamics, we show that the optimal broadcasting intensity for any user is surprisingly simple ? it is given by the position of her most recent post on each of her follower's feeds. As a consequence, we are able to develop a simple and highly efficient online algorithm, RedQueen, to sample the optimal times for the user to post. Experiments on both synthetic and real data gathered from Twitter show that our algorithm is able to consistently make a user's posts more visible over time, is robust to volume changes on her followers' feeds, and significantly outperforms the state of the art.

Semisupervised hyperspectral unmixing finds the ratio of spectral library members in the mixture of hyperspectral pixels to find the proportion of pure materials in a natural scene. The two main challenges are noise in observed spectral vectors and high mutual coherence of spectral libraries. To tackle these challenges, we propose a probabilistic sparse regression method for linear hyperspectral unmixing, which utilizes the implicit relations of neighboring pixels. We partition the hyperspectral image into rectangular patches. The sparse coefficients of pixels in each patch are assumed to be generated from a Laplacian scale mixture model with the same latent variables. These latent variables specify the probability of existence of endmembers in the mixture of each pixel. Experiments on synthetic and real hyperspectral images illustrate the superior performance of the proposed method over alternatives.

Single-time-scale distributed estimation of dynamic systems via a network of sensors/estimators is addressed in this letter. In single-time-scale distributed estimation, the two fusion steps, consensus and measurement exchange, are implemented only once, in contrast to, e.g., a large number of consensus iterations at every step of the system dynamics. We particularly discuss the problem of failure in the sensor/estimator network and how to recover for distributed estimation by adding new sensor measurements from equivalent states. We separately discuss the recovery for two types of sensors, namely α and β sensors. We propose polynomial-order algorithms to find equivalent state nodes in graph representation of the system to recover for distributed observability. The polynomial-order solution is particularly significant for large-scale systems.

People usually get involved in multiple social networks to enjoy new services or to fulfill their needs. Many new social networks try to attract users of other existing networks to increase the number of their users. Once a user (called source user) of a social network (called source network) joins a new social network (called target network), a new inter-network link (called anchor link) is formed between the source and target networks. In this paper, we concentrated on predicting the formation of such anchor links between heterogeneous social networks. Unlike conventional link prediction problems in which the formation of a link between two existing users within a single network is predicted, in anchor link prediction, the target user is missing and will be added to the target network once the anchor link is created. To solve this problem, we propose an effective general meta-path-based approach called Connector and Recursive Meta-Paths (CRMP). By using those two different categories of meta-paths, we model different aspects of social factors that may affect a source user to join the target network, resulting in the formation of a new anchor link. Extensive experiments on real-world heterogeneous social networks demonstrate the effectiveness of the proposed method against the recent methods.

Deep learning has received much attention as of the most powerful approaches for multimodal representation learning in recent years. An ideal model for multimodal data can reason about missing modalities using the available ones, and usually provides more information when multiple modalities are being considered. All the previous deep models contain separate modality-specific networks and find a shared representation on top of those networks. Therefore, they only consider high level interactions between modalities to find a joint representation for them. In this paper, we propose a multimodal deep learning framework (MDLCW) that exploits the cross weights between representation of modalities, and try to gradually learn interactions of the modalities in a deep network manner (from low to high level interactions). Moreover, we theoretically show that considering these interactions provide more intra-modality information, and introduce a multi-stage pre-training method that is based on the properties of multi-modal data. In the proposed framework, as opposed to the existing deep methods for multi-modal data, we try to reconstruct the representation of each modality at a given level, with representation of other modalities in the previous layer. Extensive experimental results show that the proposed model outperforms state-of-the-art information retrieval methods for both image and text queries on the PASCAL-sentence and SUN-Attribute databases.

This paper introduces a novel framework for modeling temporal events with complex longitudinal dependency that are generated by dependent sources. This framework takes advantage of multidimensional point processes for modeling time of events. The intensity function of the proposed process is a mixture of intensities, and its complexity grows with the complexity of temporal patterns of data. Moreover, it utilizes a hierarchical dependent nonparametric approach to model marks of events. These capabilities allow the proposed model to adapt its temporal and topical complexity according to the complexity of data, which makes it a suitable candidate for real world scenarios. An online inference algorithm is also proposed that makes the framework applicable to a vast range of applications. The framework is applied to a real world application, modeling the diffusion of contents over networks. Extensive experiments reveal the effectiveness of the proposed framework in comparison with state-of-the-art methods.

The peer to peer (P2P) file sharing applications have allocated a significant amount of today's Internet traffic. Among various P2P file sharing protocols, BitTorrent is the most common and popular one that attracts monthly a quarter of a billion users from all over the world. Similar to other P2P file sharing protocols, BitTorrent is mostly used for illegal sharing of copyright protected files such as movies, music and TV series. To impede this huge amount of illegal file distributions, anti-P2P companies have arisen to stand against these applications (specially the BitTorrent). To this end, they have begun to fire large-scale Internet attacks against BitTorrent networks. In this paper, we are going to actively measure the impact of the piece-attack against BitTorrent networks. Our measurement is divided into five scenarios in order to figure out the constraint factors that influence the success of the attack. To be able to evaluate the attack in different experiments, we defined attack effectiveness to quantitatively verify the success of the attack. Based on the measurement results, we discovered how it is possible to achieve significant outcome with modest amount of resources used by the attackers in hampering the illegal distribution of files in BitTorrent networks.

Online media play an important role in information societies by providing a convenient infrastructure for different processes. Information diffusion that is a fundamental process taking place on social and information networks has been investigated in many studies. Research on information diffusion in these networks faces two main challenges: (1) In most cases, diffusion takes place on an underlying network, which is latent and its structure is unknown. (2) This latent network is not fixed and changes over time. In this article, we investigate the diffusion network extraction (DNE) problem when the underlying network is dynamic and latent. We model the diffusion behavior (existence probability) of each edge as a stochastic process and utilize the Hidden Markov Model (HMM) to discover the most probable diffusion links according to the current observation of the diffusion process, which is the infection time of nodes and the past diffusion behavior of links. We evaluate the performance of our Dynamic Diffusion Network Extraction (DDNE) method, on both synthetic and real datasets. Experimental results show that the performance of the proposed method is independent of the cascade transmission model and outperforms the state of art method in terms of F-measure.

In this paper, a semi-supervised graph-based method for estimating 3D body pose from a sequence of silhouettes, is presented. The performance of graph-based methods is highly dependent on the quality of the constructed graph. In the case of the human pose estimation problem, the missing depth information from silhouettes intensifies the occurrence of shortcut edges within the graph. To identify and remove these shortcut edges, we measure the similarity of each pair of connected vertices through the use of sliding temporal windows. Furthermore, by exploiting the relationships between labeled and unlabeled data, the proposed method can estimate the 3D body poses, with a small set of labeled data. We evaluated the proposed method on several activities and compared the results with other recent methods. Our method significantly reduced the mean squared error, showing the positive effect of removing shortcut edges.

One of the most relevant characteristics of social networks is community structure, in which network nodes are joined together in densely connected groups between which there are only sparser links. Uncovering these sparse links (i.e. intercommunity links) has a significant role in community detection problem which has been of great importance in sociology, biology, and computer science. In this paper, we propose a novel approach, called CS-ComDet, to efficiently detect the inter-community links based on a newly emerged paradigm in sparse signal recovery, called compressive sensing. We test our method on real-world networks of various kinds whose community structures are already known, and illustrate that the proposed method detects the inter-community links accurately even with low number of measurements (i.e. when the number of measurements is less than half of the number of existing links in the network).

In this letter, a spatially weighted sparse unmixing approach is proposed as a front-end for hyperspectral image classification using a linear SVM. The idea is to partition the pixels of a hyperspectral image into a number of disjoint spatial neighborhoods. Since neighboring pixels are often composed of similar materials, their sparse codes are encouraged to have similar sparsity patterns. This is accomplished by means of a reweighted ℓ1framework where it is assumed that fractional abundances of neighboring pixels are distributed according to a common Laplacian Scale Mixture (LSM) prior with a shared scale parameter. This shared parameter determines which endmembers contribute to the group of pixels. Experiments on the AVIRIS Indian Pines show that the model is very effective in finding discriminative representations for HSI pixels, especially when the training data is limited.

This paper presents a structured dictionary-based model for hyperspectral data that incorporates both spectral and contextual characteristics of spectral samples. The idea is to partition the pixels of a hyperspectral image into a number of spatial neighborhoods called contextual groups and to model the pixels inside a group as members of a common subspace. That is, each pixel is represented using a linear combination of a few dictionary elements learned from the data, but since pixels inside a contextual group are often made up of the same materials, their linear combinations are constrained to use common elements from the dictionary. To this end, dictionary learning is carried out with a joint sparse regularizer to induce a common sparsity pattern in the sparse coefficients of a contextual group. The sparse coefficients are then used for classification using a linear support vector machine. Experimental results on a number of real hyperspectral images confirm the effectiveness of the proposed representation for hyperspectral image classification. Moreover, experiments with simulated multispectral data show that the proposed model is capable of finding representations that may effectively be used for classification of multispectral resolution samples.

Feature coding has received great attention in recent years as a building block of many image processing algorithms. In particular, the importance of the locality assumption in coding approaches has been studied in many previous works. We review this assumption and claim that using the similarity of data points to a more global set of anchor points does not necessarily weaken the coding method, as long as the underlying structure of the anchor points is considered. We propose to capture the underlying structure by assuming a random walker over the anchor points. We also show that our method is a fast approximation to the diffusion map kernel. Experiments on various data sets show that with a knowledge of the underlying structure of anchor points, different state-of-the-art coding algorithms may boost their performance in different learning tasks by utilizing the proposed method.

In this paper, a novel discriminative dictionary learning approach is proposed that attempts to preserve the local structure of the data while encouraging discriminability. The reconstruction error and sparsity inducing ℓ1-penalty of dictionary learning are minimized alongside a locality preserving and discriminative term. In this setting, each data point is represented by a sparse linear combination of dictionary atoms with the goal that its k-nearest same-label neighbors are preserved. Since the class of a new data point is unknown, its sparse representation is found once for each class. The class that produces the lowest error is associated with that point. Experimental results on five common classification datasets, show that this method outperforms state-of-the-art classifiers, especially when the training data is limited.

Sparse representation techniques for visual tracking have rarely taken advantage of the similarity between target objects in consecutive frames. In this paper, the target is divided into disjoint patches, and the sparse representation of corresponding consecutive target patches is assumed to be distributed according to a common Laplacian Scale Mixture (LSM) with a shared scale parameter. The target patches collaborate to determine this shared parameter, which in turn encourages smooth temporal variation in their representations. The target's appearance is modeled using a dictionary composed of patch templates. This patchwise treatment allows occluded patches to be detected and excluded when updating the dictionary. Experimental results on 6 challenging video sequences, show superior performance, especially in scenarios with considerable appearance change.

This paper presents a robust tracking approach to handle challenges such as occlusion and appearance change. Here, the target is partitioned into a number of patches. Then, the appearance of each patch is modeled using a dictionary composed of corresponding target patches in previous frames. In each frame, the target is found among a set of candidates generated by a particle filter, via a likelihood measure that is shown to be proportional to the sum of patch-reconstruction errors of each candidate. Since the target's appearance often changes slowly in a video sequence, it is assumed that the target in the current frame and the best candidates of a small number of previous frames, belong to a common subspace. This is imposed using joint sparse representation to enforce the target and previous best candidates to have a common sparsity pattern. Moreover, an occlusion detection scheme is proposed that uses patch-reconstruction errors and a prior probability of occlusion, extracted from an adaptive Markov chain, to calculate the probability of occlusion per patch. In each frame, occluded patches are excluded when updating the dictionary. Extensive experimental results on several challenging sequences shows that the proposed method outperforms state-of-the-art trackers.

The current multi-channel P2P video streaming architectures still suffer from several performance problems such as low Quality of Service (QoS) in unpopular channels. The P2P systems are inherently dynamic, and their performance problems could be categorized into four groups, peer churn, channel churn, uncooperative peers, and geographical distribution of peers. In this paper, for the first time, we develop a novel locality-incentive framework for multi-channel live video streaming. We propose a hierarchical overlay network architecture by utilizing a dual-mode locality-awareness method (spatial and temporal). Moreover, an incentive mechanism for encouraging peers to dedicate their upload bandwidth is introduced. Finally, an efficient buffer-map structure is proposed to predict the validation time of each video chunk request as well as the receiving time of video chunks. We have evaluated the performance of our framework via extensive simulations and compared it with the state-of-the-art method in multi-channel systems. The simulation results demonstrate that the quality of unpopular channels is improved by up to 38%. Moreover, the proposed method improved the quality of video by reducing the playback delay (up to 27%), distortion (up to 39%), and reducing the redundant traffic into the Internet backbones (up to 43%).

Bag of words models for feature extraction have demonstrated top-notch performance in image classification. These representations are usually accompanied by a coding method. Recently, methods that code a descriptor giving regard to its nearby bases have proved efficacious. These methods take into account the nonlinear structure of descriptors, since local similarities are a good approximation of global similarities. However, they confine their usage of the global similarities to nearby bases. In this paper, we propose a coding scheme that brings into focus the manifold structure of descriptors, and devise a method to compute the global similarities of descriptors to the bases. Given a local similarity measure between bases, a global measure is computed. Exploiting the local similarity of a descriptor and its nearby bases, a global measure of association of a descriptor to all the bases is computed. Unlike the locality-based and sparse coding methods, the proposed coding varies smoothly with respect to the underlying manifold. Experiments on benchmark image classification datasets substantiate the superiority of the proposed method over its locality and sparsity based rivals.

This paper addresses the problem of recovering sparse link vectors with network topological constraints that is motivated by network inference and tomography applications. We propose a novel framework called UCS-NT in the context of compressive sensing for sparse recovery in networks. In order to efficiently recover sparse specification of link vectors, we construct a feasible measurement matrix using this framework through connected paths. It is theoretically shown that, only O(k log(n)) path measurements are sufficient for uniquely recovering any k-sparse link vector. Moreover, extensive simulations demonstrate that this framework would converge to an accurate solution for a wide class of networks.

This paper presents a new Compressive Sensing (CS) scheme for detecting network congested links. We focus on decreasing the required number of measurements to detect all congested links in the context of network tomography. We have expanded the LASSO objective function by adding a new term corresponding to the prior knowledge based on the relationship between the congested links and the corresponding link Betweenness Centrality (BC). The accuracy of the proposed model is verified by simulations on two real datasets. The results demonstrate that our model outperformed the state-of-the-art CS based method with significant improvements in terms of F-Score.

BACKGROUND: The abundance of gene expression microarray data has led to the development of machine learning algorithms applicable for tackling disease diagnosis, disease prognosis, and treatment selection problems. However, these algorithms often produce classifiers with weaknesses in terms of accuracy, robustness, and interpretability. This paper introduces fuzzy support vector machine which is a learning algorithm based on combination of fuzzy classifiers and kernel machines for microarray classification. RESULTS: Experimental results on public leukemia, prostate, and colon cancer datasets show that fuzzy support vector machine applied in combination with filter or wrapper feature selection methods develops a robust model with higher accuracy than the conventional microarray classification models such as support vector machine, artificial neural network, decision trees, k nearest neighbors, and diagonal linear discriminant analysis. Furthermore, the interpretable rule-base inferred from fuzzy support vector machine helps extracting biological knowledge from microarray data. CONCLUSIONS: Fuzzy support vector machine as a new classification model with high generalization power, robustness, and good interpretability seems to be a promising tool for gene expression microarray classification.

Partially-observed network data collected by link-tracing based sampling methods is often being studied to obtain the characteristics of a large complex network. However, little attention has been paid to sampling from directed networks such as WWW and Peer-to-Peer networks. In this paper, we propose a novel two-step (sampling/estimation) framework to measure nodal characteristics which can be defined by an average target function in an arbitrary directed network. To this end, we propose a personalized PageRank-based algorithm to visit and sample nodes. This algorithm only uses already visited nodes as local information without any prior knowledge about the latent structure of the network. Moreover, we introduce a new estimator based on the approximate importance sampling to estimate average target functions. The proposed estimator utilizes calculated PageRank value of each sampled node as an approximation for the exact visiting probability. To the best of our knowledge, this is the first study on correcting the bias of a sampling method by re-weighting of measured values that considers the effect of approximation of visiting probabilities. Comprehensive theoretical and empirical analysis of the estimator demonstrate that it is asymptotically unbiased even in situations where stationary distribution of PageRank is poorly approximated.

In this paper, we address the problem of data description using a Bayesian framework. The goal of data description is to draw a boundary around objects of a certain class of interest to discriminate that class from the rest of the feature space. Data description is also known as one-class learning and has a wide range of applications. The proposed approach uses a Bayesian framework to precisely compute the class boundary and therefore can utilize domain information in form of prior knowledge in the framework. It can also operate in the kernel space and therefore recognize arbitrary boundary shapes. Moreover, the proposed method can utilize unlabeled data in order to improve accuracy of discrimination. We evaluate our method using various real-world datasets and compare it with other state of the art approaches of data description. Experiments show promising results and improved performance over other data description and one-class learning algorithms.

The quality of service inWireless Multimedia Sensor Networks (WMSN) is related to packet loss rate. Recently different studies have been done on developing efficient protocols in the transport layer for controlling packet loss in WMSN. However, all of these protocols are independent of the characteristics of multimedia content. In this paper, a novel transport layer protocol, called User Datagram Dispatcher Protocol (UDDP), is proposed to minimize the packet loss ratio in WMSN by considering traffic characteristics, the inter-arrival pattern of packets and packet priority. UDDP is a new cross-layer transport layer that uses information of MAC and application layers to distribute packet arrivals in a GOP. The experimental results show that our protocol provides performance improvement in terms of high video quality, packet loss, energy conservation and delay compared to other protocols such as CCF and PCCP.

Using Peer-to-Peer (P2P) overlay networks have become a progressively popular approach for streaming live media over the Internet due to their deployment simplicity and scalability. In this paper, we propose a new hybrid push-pull live P2P video streaming protocol called PPM that combines the benefits of pull and push mechanisms for video delivery. Our main goal is to minimize the network end-to-end delay compared to the pure mesh networks. The PPM consists of two phases; Pull-based and Push-based. In the first phase, a new peer joins to the network based on a pull-based mechanism. In the second phase, a parent node based on the peers' overlay hop count in the mesh topology is selected. Then, a dynamic tree is constructed to push the high priority video frames to the children of the selected parent. Using OMNET++ as the simulation platform, we show that beside significant improvement on the end-to-end delay, PPM achieves lower visual distortion compared to the pure mesh networks. Moreover, the simulation results confirm superiority of the PPM in comparison with the popular mesh-based P2P streaming systems.

In this letter, a new adaptive Gaussian process (GP) frame work for signal extrapolation is proposed. Signal extrapolation is an essential task in many applications such as concealment of corrupted data in image and video communications. While possessing many interesting properties, Gaussian process priors with inappropriate stationary kernels may create extremely blurred edges in concealed areas of the image. To address this problem, we propose adaptive non-stationary kernels in a Gaussian process framework. The proposed adaptive kernel functions are defined based on the hypothesized edges of the missing areas. Experimental results verify the effectiveness of the proposed method compared to the existing state of the art algorithms, based on objective and subjective evaluations.

Twitter as one of the most important microblogging online social networks has attracted more than 200 million users in recent years. Although there have been several attempts on characterizing the Twitter by using incomplete sampled data, they have not been very successful to estimate the characteristics of the whole network. In this paper, we characterize Twitter by sampling from its social graph and user behaviors through a random walk based sampling technique called Respondent-Driven Sampling (RDS). To the best of our knowledge, for the first time RDS method and its estimator are used in order to obtain uniform unbiased estimation of several key structural and behavioral properties of Twitter. We compared the performance of the proposed method with other sampling methods such as Metropolis-Hasting Random Walk (MHRW) and sampling from active users (Timeline) against the uniform sampling (UNI). In order to gather the required data, we have implemented four independent crawlers. Our experimental results indicate that the RDS method exhibits lower estimation errors to the sample in- and out-degree distribution compared to MHRW and Timeline. We also show that RDS is more suitable to sample the followers vs. followings ratio, and the correlation between followers/followings vs. tweets.

In this paper, we propose a Gaussian Process Regression (GPR) framework for concealment of corrupted motion vectors in predictive video coding of packet video systems. The problem of estimating the lost motion vectors is modelled as a kernel construction problem in a Bayesian framework. First, to describe the similarity between the neighboring motion vectors, a kernel function is defined. Then the parameters of the kernel function is estimated as the coefficients of a linear Bayesian estimator. The experimental results verify the superiority of the proposed algorithm over the conventional and state of the art motion vector concealment methods. Moreover, noticeable improvements on both objective and subjective measures, on videos with heavy packet loss rates have been achieved.

Semi-supervised learning based on manifolds has been the focus of extensive research in recent years. Convenient neighbourhood graph construction is a key component of a successful semi-supervised classification method. Previous graph construction methods fail when there are pairs of data points that have small Euclidean distance, but are far apart over the manifold. To overcome this problem, we start with an arbitrary neighbourhood graph and iteratively update the edge weights by using the estimates of the geodesic distances between points. Moreover, we provide theoretical bounds on the values of estimated geodesic distances. Experimental results on real-world data show significant improvement compared to the previous graph construction methods.

In this paper, we propose an ensemble-based approach to boost performance of Tied Factor Analysis(TFA) to overcome some of the challenges in face recognition across large pose variations. We use Adaboost. m1 to boost TFA which has shown to possess state-of-the-art face recognition performance under large pose variations. To this end, we have employed boosting as a discriminative training in the TFA as a generative model. In this model, TFA is used as a base classifier for the boosting algorithm and a weighted likelihood model for TFA is proposed to adjust the importance of each training data. Moreover, a modified weighting and a diversity criterion are used to generate more diverse classifiers in the boosting process. Experimental results on the FERET data set demonstrated the improved performance of the Boosted Tied Factor Analysis(BTFA) in comparison with TFA for lower dimensions when a holistic approach is being used.

This paper presents a novel algorithm for matching image interest points. Potential interest points are identified by searching for local peaks in Difference-of-Gaussian (DoG) images. We refine and assign rotation, scale and location for each keypoint by using the SIFT algorithm . Pseudo log-polar sampling grid is then applied to properly scaled image patches around each keypoint, and a weighted adaptive lifting scheme transform is designed for each ring of the log-polar grid. The designed adaptive transform for a ring in the reference keypoint and the general non-adaptive transform are applied to the corresponding ring in a test keypoint. Similarity measure is calculated by comparing the corresponding transform domain coefficients of the adaptive and non-adaptive transforms. We refer to the proposed versatile system of Rotation And Scale Invariant Matching as RASIM. Our experiments show that the accuracy of RASIM is more than SIFT, which is the most widely used interest point matching algorithm in the literature. RASIM is also more robust to image deformations while its computation time is comparable to SIFT.

We have developed a Gaussian Process Regression method with adaptive kernels for concealment of the missing macro-blocks of block-based video compression schemes in a packet video system. Despite promising results, the proposed algorithm introduces a solid framework for further improvements. In this paper, the problem of estimating lost macro-blocks will be solved by estimating the proper covariance function of the Gaussian process defined over a region around the missing macro-blocks (i.e. its kernel function). In order to preserve block edges, the kernel is constructed adaptively by using the local edge related information. Moreover, we can achieve more improvement by local estimation of the kernel parameters. While restoring the prominent edges of the missing macro-blocks, the proposed method produces perceptually smooth concealed frames. Objective and subjective evaluations verify the effectiveness of the proposed method.

Searching large image databases is a time consuming process when done manually. Current CBIR methods mostly rely on training data in specific domains. When source and domain of images are unknown, unsupervised methods provide better solutions. In this work, we use a hierarchical clustering scheme to group images in an unknown and large image database. In addition, the user should provide the current class assignment of a small number of images as a feedback to the system. The proposed method uses this feedback to guess the number of required clusters, and optimizes the weight vector in an iterative manner. In each step, after modification of the weight vector, the images are reclustered. We compared our method with a similar approach (but without users feedback) named CLUE. Our experimental results show that by considering the user feedback, the accuracy of clustering is considerably improved.

Power and bandwidth constraints are two major challenges in wireless sensor networks. Since a considerable amount of energy in sensor networks is consumed for data transmission, compression techniques may prolong the life of such networks. Moreover, with fewer bits to transmit, the network can cope better with the problem of inadequate bandwidth. In this paper, we consider an image sensor network and propose a paradigm based on the principles of Distributed Source Coding (DSC) for efficient compression. Our method relies on high correlation between the sensor nodes. The algorithm consists of two phases: the Training Phase and the Main Phase. In the Training Phase an aggregation node or a cluster head determines the correlation of the sources, and in the Main Phase the image is coded based on the computed correlations. We compare our method with JPEG and show its superiority in terms of compression ratio as the correlation increases.

Error control is an important mechanism for providing robust multimedia communication in wireless sensor networks. Although there have been several research works in analysis of error control mechanisms in wireless multimedia networks and wireless sensor networks, but none of them are directly applicable to the wireless multimedia sensor networks (WMSNs) which has resource and performance constraints of WSNs as well as QoS requirements of multimedia communications. In this paper, we comprehensively evaluate the performance of several error control mechanisms in WMSNs. The results of our analysis provide an extensive comparison between automatic repeat request (ARQ), forward error correction (FEC), and hybrid FEC/ARQ error control mechanisms in terms of frame loss rate, frame peak signal-to-noise ratio (PSNR), and energy efficiency.

Mobility is one of the most challenging issues in mobile ad-hoc networks and has a significant impact on performance of network protocols. Different kinds of mobility models have been proposed to represent the movement pattern of mobile nodes in mobile ad-hoc networks. These models attempt to capture various mobility characteristics existing in real movement of mobile nodes. In this paper, a new framework called Fuzzy Mobility Analyzer has been proposed to evaluate mobility models. At first, our framework categorizes mobility models based on their mobility characteristics into five classes; subsequently it uses mobility metrics to capture the mobility and graph connectivity characteristics of mobile nodes in each class; finally it specifies the similarity degree between mobility classes and real world movements of mobile nodes by using the fuzzy set theory. Experimental results on well known mobility models with real world mobility traces is presented to verify our claims.

The increasing use of multiparty Web conferencing applications demands for suitable multicast protocols. Limited bandwidth of typical Internet user requires the underlying multicast routing to be efficient. In this paper we present HOMA; an application layer multicast protocol which has been tailored for small scale multiparty videoconferencing applications. Considering the requirements of such applications, HOMA uses a heuristic routing to construct efficient multicast trees on the application layer. We have evaluated the performance of the proposed routing algorithm through simulations. Experimental results indicate that the proposed algorithm performs better than the related algorithms in terms of rejection rate, while satisfying QoS constraints of conferencing applications such as end-to-end delay and bandwidth.

In this paper a new solution for the single sample problem in low resolution face recognition is proposed. The proposed solution uses an enhanced virtual pose generation method to extend the number of face images of each identity. Using a top-right face image of an identity in the gallery, the method generates other poses of the same identity. Face images are represented as a set of local patches. In order to avoid image alignment problems, patches in the first and second pose are clustered. For each cluster the mapping between the patches of the two poses is learned. Experimental results show superior subjective and objective performance of the proposed method on CASPEAL database compared to earlier local and global methods.

In this paper, we propose a novel OCR system which can recognize and calculate handwritten Persian arithmetic expressions without using a keyboard or a memory to store the intermediate results. Our system is composed of two major phases: character recognition and calculation. The recognition phase is based on a new approach for feature extraction followed by a Fuzzy Support Vector Machines (FSVMs) as the classifier. In calculation phase a simple algorithm is used for calculating the recognized arithmetic expressions. The performance of the system was evaluated on a database consisting of 3400 digits and symbols written by 20 different people. 92 percent accuracy in recognition proves the good performance of the proposed system.

Topology control algorithms in mobile ad-hoc networks aim to reduce the power consumption while keeping the topology connected. These algorithms can preserve network resources and increase network capacity. However, few efforts have focused on the issue of topology control in presence of node mobility. One of the notable mobility aware topology control protocols is the ldquomobility aware distributed topology control protocolrdquo. The main drawback of this protocol is on its mobility prediction method. This prediction method assumes linear movements and is unable to cope with sudden changes in the mobile node movements. In this paper, we propose a pattern matching based mobility prediction method in which every mobile node predicts its future location through finding similar patterns in its history of movements. Simulation results show significant improvements in terms of prediction accuracy and power consumption compared to the other known algorithms.

Overlay multicast has gained much attention in recent years as an alternative method to network layer multicast, especially for mobile ad hoc networks(MANETs). In this paper, we propose a new overlay multicast protocol to achieve simplicity of deployment, rapid adaptation of overlay structure when nodes move, and reduced delay. In our algorithm, to join or leave a multicast group, it is only sufficient for a member node to inform its first upstream member node. This updates the tree structure more rapidly when nodes move. In addition, join and leave delays are reduced and this makes the protocol suitable for multimedia multicasting in MANETs. Simulation results compared to that of ODMRP and ODOMP show that when the number of member nodes increases, our protocol outperforms ODOMP and when speed of nodes is high, it outperforms both ODOMP and ODMRP in terms of delivery ratio.

Transmission of video over error prone and still bandwidth limited wireless channels demand high compression efficiency and resilience to packet losses and errors. Scalable or layered video coding applied to highly compression efficient codecs is an ideal solution to the problem. However, scalability reduces compression efficiency of the coders. In this paper we show how compression efficiency of two-layer SNR scalable video coders can be retained via joint base-enhancement layer optimization. Simulation results show that joint base-enhancement layer optimization significantly outperforms separate optimization of the layers, and it closely follows the compression performance of the single-layer optimized codec.

In this work we have examined the problem of consensus information fusion from a novel point of view, challenging the fundamental assumption of mutual trust among the fusion parties. In quest for a method to make information fusion possible while preserving the mutual confidentiality and anonymity of the fused information even in case of collusion of the malicious nodes, we propose the Blind Information Fusion Framework (BIFF). In BIFF, which is a secure information fusion framework, the nodes are not aware of the actual information they are processing, yet converging to the intended result(s). We formulate BIFF according to the anonymization transform and discuss its robustness against collusions for privacy violation. As an example, two secure consensus averaging methods are formulated according to BIFF.

In this paper, we propose a new algorithm for occlusion handling for object tracking in the crowded video scenes. The algorithm exploits the properties of undecimated wavelet packet transform (UWPT) coefficients and texture analysis to track arbitrary objects. The algorithm is initialized by the user through specifying a region around the object of interest at the reference frame. Then, coefficients of the UWPT of the region construct a Feature Vector (FV) for every pixel in that region. Optimal search for the best match is then performed by using the generated FVs inside an adaptive search window. Adaptation of the search window is achieved by inter- frame texture analysis to find the direction and speed of the object motion. This temporal texture analysis also assists in tracking of the object under partial or short-term full occlusion. Experimental results show a good performance for occlusion handling for object tracking in crowded scenes, in particular crowds on stairs in airports or train stations.

This paper presents a lossless data hiding method using coefficients of integer wavelet domain. The modification of selected small coefficients of the high frequency subbands are used to embed data. We use the histogram modification to intelligently select the proper coefficients for data hiding. Data embedding is done by processing these selected coefficients. We show that at low payload data hiding our method has comparable PSNR than the best known reversible data hiding techniques, while at higher payloads it has significant superiority on image quality.

This paper presents a new adaptive lifting scheme transform for detecting user-selected objects in a sequence of images. In our algorithm, we first select a set of object features in the wavelet transform domain and then build an adaptive transform by using the selected features. The adaptive transform is constructed based on adaptive prediction in a lifting scheme procedure. Adaptive prediction is performed such that, the large coefficients in the high-pass component of the non-adaptive transform vanishes in the high-pass component of the adaptive transform. Finally, both the non-adaptive and adaptive transforms are applied to a given test image and the transform domain coefficients are compared for detecting the object of interest. It is shown that the presented algorithm is robust to the noisy environments with reasonable signal-to-noise ratio. We have verified our claims with experimental results on noisy 1-D signals and images

This paper presents a new algorithm for detecting user-selected objects in a sequence of images based on a new weighted adaptive lifting scheme transform. In our algorithm, we first select a set of coefficients as object features in the wavelet transform domain and then build an adaptive transform considering the selected features. The goal of the designed adaptive transform is to "vanish" the selected features as much as possible in the transform domain. After applying both non-adaptive and adaptive transforms to a given test image, the corresponding transform domain coefficients are compared for detecting the object of interest. We have verified our claim with experimental results on 1-D signals and real images

This paper introduces adaptive consensus, a spatio-temporal adaptive method to improve convergence behavior of the current consensus fusion schemes. This is achieved by introducing a time adaptive weighting method for updating each sensor data in each iteration. Adaptive consensus method will improve node convergence rate, average convergence rate and the variance of error over the network. A mathematical formulation of the method according to the adaptive filter theory as well as derivation of the time adaptive weights and convergence conditions are presented. The analytical results are verified by simulation as well

This paper presents a shape prediction algorithm in a noisy video sequence based on pixel representation in the undecimated wavelet domain. In our algorithm for tracking of user-defined shapes in a noisy sequence of images, the amplitude of coefficients in the best basis tree expansion of the undecimated wavelet packet transform are used as feature vectors (FVs). FVs robustness against noise has been achieved through inherent denoising and edge component separation in the best basis selection algorithm. The algorithm uses these FVs to track the pixels of small square blocks located at the vicinity of the object boundary. Searching for the best-matched block has been performed using conventional block matching algorithm in the wavelet domain. Our experimental results show that the algorithm is robust to noise in case of object's shape translation, rotation and/or scaling and can be used to track both rigid and non-rigid shapes in image sequences

Feature vector extraction, based on local image texture, is a primitive algorithm for many other applications, like segmentation, clustering and identification. If these feature vectors are a good match to the human visual system (HVS), we can expect to get the appropriate results by using them. Gabor filters have been used for this purpose successfully. In this paper we introduce a novel refinement, with the use of matching pursuits (MP) to improve the Gabor based texture feature extractor. With this improvement, we show that the separability of different textures increases. Another consideration in this work is computation complexity. Therefore, we limit the basis function set to reduce MP computation time.

In this paper we propose a rate-efficient peer-to-peer architecture for video multicasting over the Internet. The limited capacity of the Internet hosts and the heterogeneous property of their access links are the main challenges of the peer-to-peer video multicasting over the Internet. Although, the rate-optimized overlay tree construction is a NP-hard problem, we propose a number of distributed and efficient protocols for rate-efficient overlay tree construction. Our proposed protocols include efficient join, improvement, overlay tree refinement and an optimum rate allocation protocol. The simulation results show the efficiency of the proposed protocols in rate-efficient overlay tree construction. We show that through combination of the efficient join protocol, tree refinement operations and an optimum rate allocation algorithm we can achieve a suboptimum overlay tree.

In this paper, a new on-line handwritten signature verification system using Hidden Markov Model (HMM) is presented. The proposed system segments each signature based on its perceptually important points and then computes for each segment a number of features that are scale and displacement invariant. The resulted sequence is then used for training an HMM to achieve signature verification. Our database includes 622 genuine signatures and 1010 forgery signatures that were collected from a population of 69 human subjects. Our verification system has achieved a false acceptance rate (FAR) of 4% and a false rejection rate (FRR) of 12%.

This paper describes a new real-time algorithm for tracking user-selected objects in a sequence of images based on a new block matching algorithm in wavelet domain. In our algorithm, the amplitude of coefficients in the best basis tree expansion of undecimated wavelet packet transform is used as the feature vectors (FVs). Real-time object tracking have been achieved using a new search technique for finding the best match among FVs of the reference block and FVs of the search area in the wavelet domain. Our experimental results show that the algorithm is robust to various object deformations and noisy video sequences.

We present here a novel multiresolution-based image segmentation algorithm. The proposed method extends and improves the Gaussian mixture model (GMM) paradigm by incorporating a multiscale correlation model of pixel dependence into the standard approach. In particular, the standard GMM is modified by introducing a multiscale neighborhood clique that incorporates the correlation between pixels in space and scale. We modify the log likelihood function of the image field by a penalization term that is derived from a multiscale neighborhood clique. Maximum likelihood (ML) estimation via the expectation-maximization (EM) algorithm is used to estimate the parameters of the new model. Then, utilizing the parameter estimates, the image field is segmented with a MAP classifier. It is demonstrated that the proposed algorithm provides superior segmentations of synthetic images, yet is computationally efficient.

In this paper a novel algorithm for adaptive signal expansion is presented. Here the main concern is to efficiently represent the natural images and audio signals. These signals are one or two dimensional signals with unknown or time-varying characteristics. For this type of signal, linear expansion with a fixed set of basis functions is not flexible enough to represent the data with the desired degree of sparseness. We introduce a new algorithm called segmented orthogonal matching pursuit (SOMP). Our experimental results show that the SOMP algorithm is more suitable than the existing signal expansion algorithms for efficient representation of audio and visual information.

In this paper, a novel algorithm for low bit-rate image compression is presented. In this technique, we use a new image representation algorithm called segmented orthogonal matching pursuit (SOMP) (Rabiee and Kashyap, 1998) to encode the subbands of an image. Our preliminary results show that our algorithm performs better than the segmentation based matching pursuit (QTMP) (Rabiee et al. 1996) and EZW (Shapiro 1993) encoders at lower bit rates, based on subjective image quality and peak signal-to-noise ratio (PSNR).

A novel approach for de-blocking of block based image and video compression algorithms is presented. In our algorithm a blocky image is first decomposed into approximation and detail subspaces with a J-level multiresolution analysis (MRA). In the next step, the low-pass approximation image and the high-pass detail images are processed independently with spatially adaptive order statistic (OS) filters. The proposed algorithm can effectively eliminate or reduce the blocking artifacts and is computationally efficient.

In this paper, we demonstrate that a window-based Gaussian mixture model can be applied in the development of a robust nonlinear filter for image restoration. Via the EM algorithm, we utilize ML estimation of the spatially-varying model parameters to achieve the desired noise suppression and detail preservation. We demonstrate that this approach is a powerful tool which gives us information about the local statistics of noisy images. We demonstrate that the estimated local statistics can be efficiently utilized for outlier detection and edge detection. The advantage of our algorithm is that it can simultaneously suppress additive Gaussian and impulsive noise, while preserving fine details and edges.

A novel multiresolution algorithm for lossy gray-scale image compression is presented. High-quality low bit rate image compression is achieved first by segmenting an image into regions of different sizes based on perceptual variation in each region and then constructing a distinct code for each block by using the theory of projection pursuit (PP). Projection pursuit allows one to adaptively construct a better approximation for each block by optimally selecting basis functions. The process is stopped when the desired peak signal-to-noise ratio (PSNR) or bit rate (b/pixel) is achieved. At rates below 0.5 b/pixel, our algorithm shows superior performance, both in terms of PSNR and subjective image quality, over the Joint Photographers Expert Group (JPEG) algorithm, and comparable performance to the embedded zerotree wavelet (EZW) algorithm.

This paper presents two multiresolution segmentation-based algorithms for low bit rate image compression using hierarchical data structures. The segmentation is achieved with quadtree and binary space partitioning tree hierarchical data structures and the encoding is performed by using the projection pursuit (matching pursuit) with a finite dictionary of spline functions with various degrees of smoothness. Comparison with JPEG at rates below 0.5 bit/pixel shows superior performance both in terms of peak signal-to-noise ratio (PSNR) and subjective image quality.

There has been a growing interest in representation and compression of signals by using dictionaries of basis functions other than the traditional dictionary of sinusoids. These new set of dictionaries include cosine packets, chirplets, Gabor functions, wavelets, and wavelet packets. In this paper matching pursuit and basis pursuit with finite dictionaries of convolutional splines are used for adaptive multiresolution image compression. At the cost of computational complexity these algorithms outperform the DCT based JPEG both in terms of PSNR and subjective image quality at lower bit rates.

A novel approach for error concealment in block-based image and video coding systems over the ATM networks is presented. This new approach aims at reconstructing the lost pixels in the intra-coded frames by spatial interpolation of the nearest undamaged pixels with a recursive multi-directional nonlinear filtering scheme. The lost interframe information are also reconstructed by using robust nonlinear filtering of the adjacent and previous motion vector blocks. We assume that the important synchronization and header information of the compressed bit streams are packed into high priority ATM cells and an integral number of macroblocks are packed into normal cells for efficient transmission. Finally, simulation results are provided to illustrate the effectiveness of the proposed method.

This paper presents a new multiresolution segmentation-based algorithm for image compression. High quality low bit rate image compression is achieved by recursively coding the binary space partitioning (BSP) tree representation of images with multilevel block truncation coding (BTC). Comparison with JPEG at rates below 0.25 bit/pixel shows superior performance both in terms of power signal-to-noise ratio (PSNR) and subjective image quality.

The pre- and post-processing units for digital image filtering, are an essential part of any integrated vision or imaging system which uses an intensity image as input. These kinds of processing are normally multiple criteria optimization problems that may involve both restoration and enhancement of the degraded images. The most commonly used figures of merit for evaluating these filters are noise attenuation edge preservation, detail presentation and edge enhancement properties. In recent years, nonlinear techniques have been extensively used to achieve these tasks. However, none of these filters have shown to satisfy all of the above requirements. In this work we introduce a new nonparametric robust nonlinear filter based on generalized maximum likelihood reasoning and order statistics (GMLOS). A qualitative and quantitative comparison of GMLOS and other efficient nonlinear filters is presented to illustrate the capability of this filter in satisfying the desired requirements.>