Multi-view clustering has been found useful to leverage diverse data sources for accurate and robust underlying data representations. It typically relies on effectively integrating the latent features from different views through allocating weights while simultaneously mining their specificity and consensus information. However, it remains open how to achieve a more fine-grained sample-level weight allocation for promoting view-specific information fusion and view-shared consensus. To address this problem, we propose a novel multi-expert learning framework named Gated Variational Graph AutoEncoder with Competition and Consensus (GVGAE-C2). In particular, it employs multiple view-specific Variational Graph AutoEncoders (VGAEs) as experts to capture the latent features from their own views. Furthermore, we design a fine-grained structure-aware gating network, which dynamically computes sample-level weights based on the proposed structure-aware quality evaluation on each expert, thus facilitating competition among experts. Meanwhile, each expert is trained not only to study its assigned view's specificity features, but also explicitly encouraged to learn consensus-aware features across views. Extensive multi-view clustering experiments on benchmark datasets reveal that GVGAE-C2 significantly outperforms state-of-the-art methods.

Missingness in real-world data often arises in complex patterns, with different subsets of features missing across samples due to diverse factors. Conventional imputation methods mostly learn a single model to cater for all possible missing patterns. However, missing patterns carry specific information which can be leveraged to guide better imputation via specialization. To this end, we propose HynetImpute which leverages a hypernetwork to represent a family of variational autoencoders (VAEs) for missing pattern specialized imputation. Specifically, HynetImpute uses the hypernetwork to generate the encoder parameters of a VAE “specialized” for the input on the fly according to its missing pattern. The encoded input is then fed to a decoder which is shared among all missing patterns to encourage consistent imputation across diverse missing scenarios. Together with the loss function designed for training, HynetImpute can achieve robust imputation results and imputed data distribution consistent with the ground truth distribution and invariant to missing patterns. Through extensive experiments on synthetic, UCI and real-world datasets, HynetImpute is demonstrated to outperform the state-of-the-art baseline methods significantly in terms of imputation accuracy and robustness, effectively capturing the nuances of diverse missing data scenarios.

Time series anomaly detection has garnered significant research attention due to growing demands for temporal data monitoring across diverse domains. Despite the rapid advent of unsupervised anomaly detection models, existing approaches face two critical challenges in understanding the mechanisms of reconstruction-based models when handling diverse temporal dependencies: (1) the insufficient exploration of complex inter-timestamp relationships encompassing both short-term and long-term dependencies, and (2) the lack of integrated frameworks for jointly learning short-term patterns and long-term temporal characteristics. To address these challenges, we propose the novel Multi-Scale Hypergraph Transformer (MSHTrans), which leverages the capacity of hypergraphs for modeling multi-order temporal dependencies. Particularly, our method employs multi-scale downsampling to derive complementary fine-grained and coarse-grained representations, integrated with trainable hypergraph neural networks that can adaptively learn inter-timestamp relationships. The framework further integrates time series decomposition to systematically extract periodic and trend components from multi-granular features, thereby enhancing long-term dependency modeling. Through synergistic integration of learned short-term patterns and long-term temporal structures, the model achieves comprehensive time series reconstruction for effective anomaly detection. Extensive experiments demonstrate that MSHTrans outperforms state-of-the-art competitors with an average performance improvement of 8.21% (without point adjustment) and 3.52% (with point adjustment).

Longitudinal multimodal data, including electronic health records (EHR) and sequential chest X-rays (CXRs), is critical for modeling disease progression, yet remains underutilized due to two key challenges: (1) redundancy in consecutive CXR sequences, where static anatomical regions dominate over clinically-meaningful dynamics, and (2) temporal misalignment between sparse, irregular imaging and continuous EHR data. We introduce $\texttt{DiPro}$, a novel framework that addresses these challenges through region-aware disentanglement and multi-timescale alignment. First, we disentangle static (anatomy) and dynamic (pathology progression) features in sequential CXRs, prioritizing disease-relevant changes. Second, we hierarchically align these static and dynamic CXR features with asynchronous EHR data via local (pairwise interval-level) and global (full-sequence) synchronization to model coherent progression pathways. Extensive experiments on the MIMIC dataset demonstrate that $\texttt{DiPro}$ could effectively extract temporal clinical dynamics and achieve state-of-the-art performance on both disease progression identification and general ICU prediction tasks.

Vision-language models have been explored for radiology report generation with promising results. Yet, uncertainty elaborated in findings and the reasoning process for reaching clinical impressions are seldom explicitly modeled, reducing the clinical accuracy and trustworthiness of the generated reports. We present CURV, a novel framework that alleviates the limitations through integrated awareness of uncertainty and explicit reasoning capabilities. Our approach consists of three key components: (1) an uncertainty modeling mechanism that teaches the model to recognize and express appropriate levels of diagnostic confidence, (2) a structured reasoning framework that generates intermediate explanatory steps connecting visual findings to clinical impressions, and (3) a reasoning coherence reward that ensures logical consistency among findings, reasoning, and impressions. We implement CURV through a three-stage training pipeline that combines uncertainty-aware fine-tuning, reasoning initialization, and reinforcement learning. In particular, we adopt a comprehensive reward function addresses multiple aspects of report quality, incorporating medical term matching, uncertainty expression evaluation, and semantic coherence evaluation. Experimental results demonstrate that CURV generates clinically relevant reports with appropriate uncertainty expressions and transparent reasoning traces, significantly outperforming previous methods. CURV represents a substantial advancement toward interpretable and trustworthy AI-generated radiology reports, with broader implications for the deployment of vision-language models in high-stakes clinical environments where uncertainty awareness and reasoning transparency are essential.

The combination of electronic health records (EHR) and medical images is crucial for clinicians in making diagnoses and forecasting prognoses. Strategically fusing these two data modalities has great potential to improve the accuracy of machine learning models in clinical prediction tasks. However, the asynchronous and complementary nature of EHR and medical images presents unique challenges. Missing modalities due to clinical and administrative factors are inevitable in practice, and the significance of each data modality varies depending on the patient and the prediction target, resulting in inconsistent predictions and suboptimal model performance. To address these challenges, we propose DrFuse to achieve effective clinical multi-modal fusion. It tackles the missing modality issue by disentangling the features shared across modalities and those unique within each modality. Furthermore, we address the modal inconsistency issue via a disease-wise attention layer that produces the patient- and disease-wise weighting for each modality to make the final prediction. We validate the proposed method using real-world large-scale datasets, MIMIC-IV and MIMIC-CXR. Experimental results show that the proposed method significantly outperforms the state-of-the-art models.

Automatic radiology report generation using deep learning models has been recently explored and found promising. Neural decoders are commonly used for the report generation, where irrelevant and unfaithful contents are unavoidable. The retrieval-based approach alleviates the limitation by identifying reports which are relevant to the input to assist the generation. To achieve clinically accurate report retrieval, we make reference to clinicians' diagnostic steps of examining a radiology image where anatomical and diagnostic details are typically focused, and propose a novel hierarchical visual concept representation called anatomy-aware hierarchical vision encoding (AHIVE). To learn AHIVE, we first derive a methodology to extract hierarchical diagnostic descriptions from radiology reports and develop a CLIP-based framework for the model training. Also, the hierarchical architecture of AHIVE is designed to support interactive report retrieval so that report revision made at one layer can be propagated to the subsequent ones to trigger other necessary revisions. We conduct extensive experiments and show that AHIVE can outperform the SOTA vision-language retrieval methods in terms of clinical accuracy by a large margin. We provide also a case study to illustrate how it enables interactive report retrieval.

Workshop courses designed to foster creativity are gaining popularity. However, even experienced faculty teams find it challenging to realize a holistic evaluation that accommodates diverse perspectives. Adequate deliberation is essential to integrate varied assessments, but faculty often lack the time for such exchanges. Deriving an average score without discussion undermines the purpose of a holistic evaluation. Therefore, this paper explores the use of a Large Language Model (LLM) as a facilitator to integrate diverse faculty assessments. Scenario-based experiments were conducted to determine if the LLM could integrate diverse evaluations and explain the underlying pedagogical theories to faculty. The results were noteworthy, showing that the LLM can effectively facilitate faculty discussions. Additionally, the LLM demonstrated the capability to create evaluation criteria by generalizing a single scenario-based experiment, leveraging its already acquired pedagogical domain knowledge.

Integrating multi-modal clinical data, such as electronic health records (EHR) and chest X-ray images (CXR), is particularly beneficial for clinical prediction tasks. However, in a temporal setting, multi-modal data are often inherently asynchronous. EHR can be continuously collected but CXR is generally taken with a much longer interval due to its high cost and radiation dose. When clinical prediction is needed, the last available CXR image might have been outdated, leading to suboptimal predictions. To address this challenge, we propose DDL-CXR, a method that dynamically generates an up-to-date latent representation of the individualized CXR images. Our approach leverages latent diffusion models for patient-specific generation strategically conditioned on a previous CXR image and EHR time series, providing information regarding anatomical structures and disease progressions, respectively. In this way, the interaction across modalities could be better captured by the latent CXR generation process, ultimately improving the prediction performance. Experiments using MIMIC datasets show that the proposed model could effectively address asynchronicity in multimodal fusion and consistently outperform existing methods.

Aligning users across networks using graph representation learning has been found effective where the alignment is accomplished in a low-dimensional embedding space. Yet, highly precise alignment remains challenging, especially for nodes with long-range connectivity to labeled anchors. To alleviate this limitation, we propose WL-Align which employs a regularized representation learning framework to learn distinctive node representations. It extends the Weisfeiler-Lehman Isormorphism Test and learns the alignment in alternating phases of “across-network Weisfeiler-Lehman relabeling” and “proximity-preserving representation learning”. The across-network Weisfeiler-Lehman relabeling is achieved through iterating the anchor-based label propagation and a similarity-based hashing to exploit the known anchors’ connectivity to different nodes in an efficient and robust manner. The representation learning module preserves the second-order proximity within individual networks and is regularized by the across-network Weisfeiler-Lehman hash labels. Extensive experiments on real-world and synthetic datasets have demonstrated that our proposed WL-Align outperforms the state-of-the-art methods, achieving significant performance improvements in the “exact matching” scenario.

Predictive analysis of electronic health records (EHR) is a fundamental task that could provide actionable insights to help clinicians improve the efficiency and quality of care. EHR are commonly recorded in binary format and contain inevitable missing data. The nature of missingness may vary by patients, clinical features, and time, which incurs observation bias. It is essential to account for the binary missingness and observation bias or the predictive performance could be substantially compromised. In this paper, we develop a propensity-adjusted temporal network (PATNet) to conduct data imputation and predictive analysis simultaneously. PATNet contains three subnetworks: 1) an imputation subnetwork that generates the initial imputation based on historical observations, 2) a propensity subnetwork that infers the patient-, feature-, and time-dependent propensity scores, and 3) a prediction subnetwork that produces the missing-informative prediction using the propensity-adjusted imputations and the missing probabilities. To allow the propensity scores to be inferred from data, we use the expectation-maximization (EM) algorithm to learn the imputation and propensity subnetworks and incorporate a low-rank constraint via PARAFAC2 approximation. Extensive evaluation using the MIMIC-III and eICU datasets demonstrates that PATNet outperforms the state-of-the-art methods in terms of binary data imputation, disease progression modeling, and mortality prediction tasks.

Single-cell ribonucleic acid sequencing (scRNA-seq) enables the quantification of gene expression at the transcriptomic level with single-cell resolution, enhancing our understanding of cellular heterogeneity. However, the excessive missing values present in scRNA-seq data hinder downstream analysis. While numerous imputation methods have been proposed to recover scRNA-seq data, high imputation performance often comes with low or no interpretability. Here, we present IGSimpute, an accurate and interpretable imputation method for recovering missing values in scRNA-seq data with an interpretable instance-wise gene selection layer (GSL). IGSimpute outperforms 12 other state-of-the-art imputation methods on 13 out of 17 datasets from different scRNA-seq technologies with the lowest mean squared error as the chosen benchmark metric. We demonstrate that IGSimpute can give unbiased estimates of the missing values compared to other methods, regardless of whether the average gene expression values are small or large. Clustering results of imputed profiles show that IGSimpute offers statistically significant improvement over other imputation methods. By taking the heart-and-aorta and the limb muscle tissues as examples, we show that IGSimpute can also denoise gene expression profiles by removing outlier entries with unexpectedly high expression values via the instance-wise GSL. We also show that genes selected by the instance-wise GSL could indicate the age of B cells from bladder fat tissue of the Tabula Muris Senis atlas. IGSimpute can impute one million cells using 64 min, and thus applicable to large datasets.

Variational autoencoders (VAEs) are a class of effective deep generative models, with the objective to approximate the true, but unknown data distribution. VAEs make use of latent variables to capture high-level semantics so as to reconstruct the data well with the help of informative latent variables. Yet, training VAEs tends to suffer from posterior collapse, when the decoder is parameterized by an autoregressive model for sequence generation. VAEs can be further enhanced by introducing multiple layers of latent variables, but the posterior collapse issue hinders the adoption of such hierarchical VAEs in real-world applications. In this paper, we introduce InfoMaxHVAE, which integrates mutual information estimated via neural networks into hierarchical VAEs to alleviate posterior collapse, when powerful autoregressive models are used for modeling sequences. Experimental results on a number of text and image datasets show that InfoMaxHVAE can outperform the state-of-the-art baselines and exhibits less posterior collapse. We further show that InfoMaxHVAE can shape a coarse-to-fine hierarchical organization of the latent space.

Representation learning has been applied to Electronic Health Records (EHR) for medical concept embedding and the downstream predictive analytics tasks with promising results. Medical ontologies can also be integrated to guide the learning so the embedding space can better align with existing medical knowledge. Yet, properly carrying out the integration is non-trivial. Medical concepts that are similar according to a medical ontology may not be necessarily close in the embedding space learned from the EHR data, as medical ontologies organize medical concepts for their own specific objectives. Any integration methodology without considering the underlying inconsistency will result in sub-optimal medical concept embedding and, in turn, degrade the performance of the downstream tasks. In this article, we propose a novel representation learning framework called ADORE ( AD aptive O ntological RE presentations) that allows the medical ontologies to adapt their structures for more robust integrating with the EHR data. ADORE first learns multiple embeddings for each category in the ontology via an attention mechanism. At the same time, it supports an adaptive integration of categorical and multi-relational ontologies in the embedding space using a category-aware graph attention network. We evaluate the performance of ADORE on a number of predictive analytics tasks using two EHR datasets. Our experimental results show that the medical concept embeddings obtained by ADORE can outperform the state-of-the-art methods for all the tasks. More importantly, it can result in clinically meaningful sub-categorization of the existing ontological categories and yield attention values that can further enhance the model interpretability.

Social network alignment aims at aligning person identities across social networks. Embedding based models have been shown effective for the alignment where the structural proximity preserving objective is typically adopted for the model training. With the observation that “overly-close” user embeddings are unavoidable for such models causing alignment inaccuracy, we propose a novel learning framework which tries to enforce the resulting embeddings to be more widely apart among the users via the introduction of carefully implanted pseudo anchors. We further proposed a meta-learning algorithm to guide the updating of the pseudo anchor embeddings during the learning process. The proposed intervention via the use of pseudo anchors and meta-learning allows the learning framework to be applicable to a wide spectrum of network alignment methods. We have incorporated the proposed learning framework into several state-of-the-art models. Our experimental results demonstrate its efficacy where the methods with the pseudo anchors implanted can outperform their counterparts without pseudo anchors by a fairly large margin, especially when there only exist very few labeled anchors.

Data summarization that presents a small subset of a dataset to users has been widely applied in numerous applications and systems. Many datasets are coded with hierarchical terminologies, e.g., gene ontology, disease ontology, to name a few. In this paper, we study the weighted tree summarization. We motivate and formulate our${\mathsf {kWTS}}$-${\mathsf {problem}}$as selecting a diverse set of$k$nodes tosummarize a hierarchicaltree$T$withweighted terminologies. We first propose an efficient greedy tree summarization algorithm${\mathsf {GTS}}$. It solves the problem with$(1-1/e)$-approximation guarantee. Although${\mathsf {GTS}}$achieves quality-guaranteed answers approximately, but it is still not optimal. To tackle the problem optimally, we further develop a dynamic programming algorithm${\mathsf {OTS}}$to obtain optimal answers for${\mathsf {kWTS}}$-${\mathsf {problem}}$in$O(nhk^3)$time, where$n, h$are the node size and height in tree$T$. The algorithm complexity and correctness of${\mathsf {OTS}}$are theoretically analyzed. In addition, we propose a useful optimization technique of tree reduction to remove useless nodes with zero weights and shrink the tree into a smaller one, which ensures the efficiency acceleration of both${\mathsf {GTS}}$and${\mathsf {OTS}}$in real-world datasets. Moreover, we illustrate one useful application of graph visualization based on the answer of$k$-sized tree summarization and show it in a novel case study. Extensive experimental results on real-world datasets show the effectiveness and efficiency of our proposed approximate and optimal algorithms for tree summarization. Furthermore, we conduct a usability evaluation of attractive topic recommendation on ACM Computing Classification System dataset to validate the usefulness of our model and algorithms.

Despite the recent success of deep learning models for text generation, generating clinically accurate reports remains challenging. More precisely modeling the relationships of the abnormalities revealed in an X-ray image has been found promising to enhance the clinical accuracy. In this paper, we first introduce a novel knowledge graph structure called an attributed abnormality graph (ATAG). It consists of interconnected abnormality nodes and attribute nodes for better capturing more fine-grained abnormality details. In contrast to the existing methods where the abnormality graph are constructed manually, we propose a methodology to automatically construct the fine-grained graph structure based on annotated X-ray reports and the RadLex radiology lexicon. We then learn the ATAG embeddings as part of a deep model with an encoder-decoder architecture for the report generation. In particular, graph attention networks are explored to encode the relationships among the abnormalities and their attributes. A hierarchical attention attention and a gating mechanism are specifically designed to further enhance the generation quality. We carry out extensive experiments based on the benchmark datasets, and show that the proposed ATAG-based deep model outperforms the SOTA methods by a large margin in ensuring the clinical accuracy of the generated reports.

Time-course single-cell RNA sequencing (scRNA-seq) data have been widely used to explore dynamic changes in gene expression of transcription factors (TFs) and their target genes. This information is useful to reconstruct cell-type-specific gene regulatory networks (GRNs). However, the existing tools are commonly designed to analyze either time-course bulk gene expression data or static scRNA-seq data via pseudo-time cell ordering. A few methods successfully utilize the information from multiple time points while also considering the characteristics of scRNA-seq data. We proposed dynDeepDRIM, a novel deep learning model to reconstruct GRNs using time-course scRNA-seq data. It represents the joint expression of a gene pair as an image and utilizes the image of the target TF-gene pair and the ones of the potential neighbors to reconstruct GRNs from time-course scRNA-seq data. dynDeepDRIM can effectively remove the transitive TF-gene interactions by considering neighborhood context and model the gene expression dynamics using high-dimensional tensors. We compared dynDeepDRIM with six GRN reconstruction methods on both simulation and four real time-course scRNA-seq data. dynDeepDRIM achieved substantially better performance than the other methods in inferring TF-gene interactions and eliminated the false positives effectively. We also applied dynDeepDRIM to annotate gene functions and found it achieved evidently better performance than the other tools due to considering the neighbor genes.

Non-negative tensor factorization has been shown a practical solution to automatically discover phenotypes from the electronic health records (EHR) with minimal human supervision. Such methods generally require an input tensor describing the inter-modal interactions to be pre-established; however, the correspondence between different modalities (e.g., correspondence between medications and diagnoses) can often be missing in practice. Although heuristic methods can be applied to estimate them, they inevitably introduce errors, and leads to sub-optimal phenotype quality. This is particularly important for patients with complex health conditions (e.g., in critical care) as multiple diagnoses and medications are simultaneously present in the records. To alleviate this problem and discover phenotypes from EHR with unobserved inter-modal correspondence, we propose the collective hidden interaction tensor factorization (cHITF) to infer the correspondence between multiple modalities jointly with the phenotype discovery. We assume that the observed matrix for each modality is marginalization of the unobserved inter-modal correspondence, which are reconstructed by maximizing the likelihood of the observed matrices. Extensive experiments conducted on the real-world MIMIC-III dataset demonstrate that cHITF effectively infers clinically meaningful inter-modal correspondence, discovers phenotypes that are more clinically relevant and diverse, and achieves better predictive performance compared with a number of state-of-the-art computational phenotyping models.

In few-shot domain adaptation (FDA), classifiers for the target domain are trained with \emph{accessible} labeled data in the source domain (SD) and few labeled data in the target domain (TD). However, data usually contain private information in the current era, e.g., data distributed on personal phones. Thus, the private data will be leaked if we directly access data in SD to train a target-domain classifier (required by FDA methods). In this paper, to prevent privacy leakage in SD, we consider a very challenging problem setting, where the classifier for the TD has to be trained using few labeled target data and a well-trained SD classifier, named few-shot hypothesis adaptation (FHA). In FHA, we cannot access data in SD, as a result, the private information in SD will be protected well. To this end, we propose a target-oriented hypothesis adaptation network (TOHAN) to solve the FHA problem, where we generate highly-compatible unlabeled data (i.e., an intermediate domain) to help train a target-domain classifier. TOHAN maintains two deep networks simultaneously, in which one focuses on learning an intermediate domain and the other takes care of the intermediate-to-target distributional adaptation and the target-risk minimization. Experimental results show that TOHAN outperforms competitive baselines significantly.

PARAFAC2 factorization provides a practical solution to map the temporally irregular electronic health records (EHR) to clinically relevant and interpretable phenotypes.Existing methods ignore the effect of interdependency of diseases over clinical history.Consequently, the crucial temporal information contained in the EHR data cannot be fully utilized and the learned phenotypes can be sub-optimal to characterize patients with progressive conditions.To address this issue, we propose a novel temporally dependent PARAFAC2 (TedPar) factorization in which the temporal dependency among the phenotypes is explicitly modeled.TedPar learns a set of target phenotypes to capture the clinical features relevant to the diseases of interest and a set of background phenotypes to capture irrelevant but frequently co-occurring clinical features.By effectively modeling the temporal dependency and separating relevant and irrelevant features, the discovered target phenotypes can be used to model the progression of the diseases of interest.Empirical evaluations show that TedPar obtains up to 32.4% relative improvement in reconstruction accuracy over the test set, suggesting significantly better generalizability than the baselines for both noise-free and heavily noisy input data.Qualitative analysis also shows that TedPar is capable of discovering clinically meaningful phenotypes and capturing the temporal dependency between them.

Single-cell RNA sequencing has enabled to capture the gene activities at single-cell resolution, thus allowing reconstruction of cell-type-specific gene regulatory networks (GRNs). The available algorithms for reconstructing GRNs are commonly designed for bulk RNA-seq data, and few of them are applicable to analyze scRNA-seq data by dealing with the dropout events and cellular heterogeneity. In this paper, we represent the joint gene expression distribution of a gene pair as an image and propose a novel supervised deep neural network called DeepDRIM which utilizes the image of the target TF-gene pair and the ones of the potential neighbors to reconstruct GRN from scRNA-seq data. Due to the consideration of TF-gene pair's neighborhood context, DeepDRIM can effectively eliminate the false positives caused by transitive gene-gene interactions. We compared DeepDRIM with nine GRN reconstruction algorithms designed for either bulk or single-cell RNA-seq data. It achieves evidently better performance for the scRNA-seq data collected from eight cell lines. The simulated data show that DeepDRIM is robust to the dropout rate, the cell number and the size of the training data. We further applied DeepDRIM to the scRNA-seq gene expression of B cells from the bronchoalveolar lavage fluid of the patients with mild and severe coronavirus disease 2019. We focused on the cell-type-specific GRN alteration and observed targets of TFs that were differentially expressed between the two statuses to be enriched in lysosome, apoptosis, response to decreased oxygen level and microtubule, which had been proved to be associated with coronavirus infection.

Properly analyzing health insurance claims data could lead to significant business insights and benefits for health service providers and insurance companies. Yet, health insurance data is often high dimensional and contains complex interleave sequences of claims. Instead of conducting machine learning tasks directly on the raw data, a better approach is performing the tasks on high-quality embeddings of the raw data. Driven by the real business need of Solution Segic Inc., a Canadian technology company in the group insurance industry, we extract health insurance claims embeddings with neural networks in the context of anomaly detection. We propose and thoroughly examine six embedding components that are customized based on different possible assumptions made on the data. One of our proposed embedding components, EC-ReStepRec, significantly outperforms other candidates on two anomaly detection tasks. This is the first embedding study done on health insurance claims for anomaly detection.

Binary data with one-class missing values are ubiquitous in real-world applications. They can be represented by irregular tensors with varying sizes in one dimension, where value one means presence of a feature while zero means unknown (i.e., either presence or absence of a feature). Learning accurate low-rank approximations from such binary irregular tensors is a challenging task. However, none of the existing models developed for factorizing irregular tensors take the missing values into account, and they assume Gaussian distributions, resulting in a distribution mismatch when applied to binary data. In this paper, we propose Logistic PARAFAC2 (LogPar) by modeling the binary irregular tensor with Bernoulli distribution parameterized by an underlying real-valued tensor. Then we approximate the underlying tensor with a positive-unlabeled learning loss function to account for the missing values. We also incorporate uniqueness and temporal smoothness regularization to enhance the interpretability. Extensive experiments using large-scale real-world datasets show that LogPar outperforms all baselines in both irregular tensor completion and downstream predictive tasks. For the irregular tensor completion, LogPar achieves up to 26% relative improvement compared to the best baseline. Besides, LogPar obtains relative improvement of 13.2% for heart failure prediction and 14% for mortality prediction on average compared to the state-of-the-art PARAFAC2 models.

Aligning users across different social networks has become increasingly studied as an important task to social network analysis. In this paper, we propose a novel representation learning method that mainly exploits social structures for the network alignment. In particular, the proposed network embedding framework models the follower-ship and followee-ship of each user explicitly as input and output context vectors, while preserving the proximity of users with “similar” followers and followees in the embedded space. We incorporate both known and predicted user anchors across the networks as constraints to facilitate the transfer of context information to achieve accurate user alignment. Both network embedding and user alignment are inferred under a unified optimization framework with negative sampling adopted to ensure scalability. Also, variants of the proposed framework, including the incorporation of higher-order structural features, are also explored for further boosting the alignment accuracy. Extensive experiments on large-scale social and academia network datasets demonstrate the efficacy of our proposed model compared with state-of-the-art methods.

Quantization has been widely adopted for large-scale multimedia retrieval due to its effectiveness of coding highdimensional data. Deep quantization models have been demonstrated to achieve the state-of-the-art retrieval accuracy. However, training the deep models given a large-scale database is highly time-consuming as a large amount of parameters are involved. Existing deep quantization methods often sample only a subset from the database for training, which may end up with unsatisfactory retrieval performance as a large portion of label information is discarded. To alleviate this problem, we propose a novel model called Similarity Preserving Deep Asymmetric Quantization (SPDAQ) which can directly learn the compact binary codes and quantization codebooks for all the items in the database efficiently. To do that, SPDAQ makes use of an image subset as well as the label information of all the database items so the image subset items and the database items are mapped to two different but correlated distributions, where the label similarity can be well preserved. An efficient optimization algorithm is proposed for the learning. Extensive experiments conducted on four widely-used benchmark datasets demonstrate the superiority of our proposed SPDAQ model.

Non-negative Tensor Factorization (NTF) has been shown effective to discover clinically relevant and interpretable phenotypes from Electronic Health Records (EHR). Existing NTF based computational phenotyping models aggregate data over the observation window, resulting in the learned phenotypes being mixtures of disease states appearing at different times. We argue that by separating the clinical events happening at different times in the input tensor, the temporal dynamics and the disease progression within the observation window could be modeled and the learned phenotypes will correspond to more specific disease states. Yet how to construct the tensor for data samples with different temporal lengths and properly capture the temporal relationship specific to each individual data sample remains an open challenge. In this paper, we propose a novel Collective Non-negative Tensor Factorization (CNTF) model where each patient is represented by a temporal tensor, and all of the temporal tensors are factorized collectively with the phenotype definitions being shared across all patients. The proposed CNTF model is also flexible to incorporate non-temporal data modality and RNN-based temporal regularization. We validate the proposed model using MIMIC-III dataset, and the empirical results show that the learned phenotypes are clinically interpretable. Moreover, the proposed CNTF model outperforms the state-of-the-art computational phenotyping models for the mortality prediction task.

Dong Qian, William K. Cheung. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Learning representations of medical concepts from the Electronic Health Records (EHR) has been shown effective for predictive analytics in healthcare. Incorporation of medical ontologies has also been explored to further enhance the accuracy and to ensure better alignment with the known medical knowledge. Most of the existing work assumes that medical concepts under the same ontological category should share similar representations, which however does not always hold. In particular, the categorizations in medical ontologies were established with various factors being considered. Medical concepts even under the same ontological category may not follow similar occurrence patterns in the EHR data, leading to contradicting objectives for the representation learning. In this paper, we propose a deep learning model called MMORE which alleviates this conflicting objective issue by allowing multiple representations to be inferred for each ontological category via an attention mechanism. We apply MMORE to diagnosis prediction and our experimental results show that the representations obtained by MMORE can achieve better predictive accuracy and result in clinically meaningful sub-categorization of the existing ontological categories.

With the prevalence of smartphones, many new business opportunities are being spawned by exploiting the locations or routes of the users. We propose a probabilistic Wi-Fi localization approach to meet the special needs of the exhibition industry: tight schedules, large venues and dynamic environment. Unlike the popular fingerprinting approach which requires extensive training time and the need for retraining, our approach has a much lower start-up cost which means that it doesn't need the labor-intensive training phase while having comparable accuracy for localization. This is achieved by employing a real-time calibration of signal degradation model, which also makes our localization model be adaptive to the dynamic environment. We also propose practical optimization techniques which can dynamically control the computational cost during device localization and hence more devices can be tracked. Experiments with our testbed show that our proposed approach can achieve acceptable and stable localization results in different environment based on real-time calibration. From other tests of our system in real-world exhibition event held in a venue having of 4,000 square meters, over half of the localization errors are within 4.5 meters.

Authorship analysis (AA) is the study of unveiling the hidden properties of authors from textual data. It extracts an author's identity and sociolinguistic characteristics based on the reflected writing styles in the text. The process is essential for various areas, such as cybercrime investigation, psycholinguistics, political socialization, etc. However, most of the previous techniques critically depend on the manual feature engineering process. Consequently, the choice of feature set has been shown to be scenario- or dataset-dependent. In this paper, to mimic the human sentence composition process using a neural network approach, we propose to incorporate different categories of linguistic features into distributed representation of words in order to learn simultaneously the writing style representations based on unlabeled texts for AA. In particular, the proposed models allow topical, lexical, syntactical, and character-level feature vectors of each document to be extracted as stylometrics. We evaluate the performance of our approach on the problems of authorship characterization, authorship identification and authorship verification with the Twitter, blog, review, novel, and essay datasets. The experiments suggest that our proposed text representation outperforms the static stylometrics, dynamic n -grams, latent Dirichlet allocation, latent semantic analysis, distributed memory model of paragraph vectors, distributed bag of words version of paragraph vector, word2vec representations, and other baselines.

With the prevalence of WiFi devices (e.g., smartphones), many new business opportunities are being spawned by exploiting the routes or locations of users. Nevertheless, most current approaches based on received signal strength indicator (RSSI) usually assume that the targeted (tracked) devices have the transmission characteristics similar to the devices used for system training. This is far from the reality and may lead to considerable errors. We propose a robust localization approach which automatically infers a customized signal-strength-to-distance function for every device on-the-fly, and simultaneously estimates the location of it. This is achieved by first approximating the function with a set of piecewise linear functions, for each targeted device, and the parameters of the linear functions are updated, with an Expectation Maximum (EM) algorithm, when more and more RSSI data of the device are available. Specifically, during the expectation step of the EM algorithm, the location of the targeted device is estimated. Whereas in the maximization step of the algorithm, the parameters of the linear functions customized for that device are updated. As the approach is capable of learning the parameters during localization, training process or system calibration is unnecessary and thus the system setup time can be shortened. This feature is practical for meeting the special needs of the exhibition industry: extremely tight schedules for setting up the site from scratch and extremely large venues. With our testbeds, experimental results show that the mean localization error can be kept about 1.7 meters for different mobile devices with different transmission characteristics. A real-world test at Hong Kong Convention and Exhibition Centre (HKCEC) was also conducted and various mobile devices can be tracked accurately with little setup time.

In this paper, we propose a novel unsupervised hashing method called Anchor-based Probability Hashing (APHash) to preserve the similarities by exploiting the distribution of data points. In particular, distances are transformed into probabilities in both original and hash code spaces. Our method aims to learn hash codes which minimize the mismatch between probability distributions of these two spaces. To address the high complexity issue, our method randomly selects a set of anchors and constructs asymmetric probability matrices. In this way, APHash can make use of the correlation between anchors and data points to learn hash codes more efficiently. Experimental results on two benchmark datasets demonstrate the effectiveness of the proposed APHash method, outperforming state-of-the-art hashing approaches in the application of image retrieval.

Hashing is an efficient approximate nearest neighbor search method and has been widely adopted for large-scale multimedia retrieval. While supervised learning is more popular for the data-dependent hashing, deep unsupervised hashing methods have recently been developed to learn non-linear transformations for converting multimedia inputs to binary codes. Most of existing deep unsupervised hashing methods make use of a quadratic constraint for minimizing the difference between the compact representations and the target binary codes, which inevitably causes severe information loss. In this paper, we propose a novel deep unsupervised method called DeepQuan for hashing. The DeepQuan model utilizes a deep autoencoder network, where the encoder is used to learn compact representations and the decoder is for manifold preservation. To contrast with the existing unsupervised methods, DeepQuan learns the binary codes by minimizing the quantization error through product quantization technique. Furthermore, a weighted triplet loss is proposed to avoid trivial solution and poor generalization. Extensive experimental results on standard datasets show that the proposed DeepQuan model outperforms the state-of-the-art unsupervised hashing methods for image retrieval tasks.

Non-negative tensor factorization has been shown effective for discovering phenotypes from the EHR data with minimal human supervision. In most cases, an interaction tensor of the elements in the EHR (e.g., diagnoses and medications) has to be first established before the factorization can be applied. Such correspondence information however is often missing. While different heuristics can be used to estimate the missing correspondence, any errors introduced will in turn cause inaccuracy for the subsequent phenotype discovery task. This is especially true for patients with multiple diseases diagnosed (e.g., under critical care). To alleviate this limitation, we propose the hidden interaction tensor factorization (HITF) where the diagnosis-medication correspondence and the underlying phenotypes are inferred simultaneously. We formulate it under a Poisson non-negative tensor factorization framework and learn the HITF model via maximum likelihood estimation. For performance evaluation, we applied HITF to the MIMIC III dataset. Our empirical results show that both the phenotypes and the correspondence inferred are clinically meaningful. In addition, the inferred HITF model outperforms a number of state-of-the-art methods for mortality prediction.

Scientific workflows define computational processes needed for carrying out scientific experiments. Existing workflow repositories contain hundreds of scientific workflows, where scientists can find materials and knowledge to facilitate workflow design for running related experiments. Identifying reusable fragments in growing workflow repositories has become increasingly important. In this paper, we present PSM-Flow, a probabilistic subgraph mining algorithm designed to discover commonly occurring fragments in a workflow corpus using a modified version of the Latent Dirichlet Allocation algorithm. The proposed model encodes the geodesic distance between workflow steps into the model for implicitly modeling fragments. PSM-Flow captures variations of frequent fragments while maintaining its space complexity bounded polynomially, as it requires no candidate generation. We applied PSM-Flow to three real-world scientific workflow datasets containing more than 750 workflows for neuroimaging analysis. Our results show that PSM-Flow outperforms three state of the art frequent subgraph mining techniques. We also discuss other potential future improvements of the proposed method.

The elderly living in smart homes can have their daily movement recorded and analyzed. As different elders can have their own living habits, a methodology that can automatically identify their daily activities and discover their daily routines will be useful for better elderly care and support. In this article, we focus on automatic detection of behavioral patterns from the trajectory data of an individual for activity identification as well as daily routine discovery. The underlying challenges lie in the need to consider longer-range dependency of the sensor triggering events and spatiotemporal variations of the behavioral patterns exhibited by humans. We propose to represent the trajectory data using a behavior-aware flow graph that is a probabilistic finite state automaton with its nodes and edges attributed with some local behavior-aware features. We identify the underlying subflows as the behavioral patterns using the kernel k -means algorithm. Given the identified activities, we propose a novel nominal matrix factorization method under a Bayesian framework with Lasso to extract highly interpretable daily routines. For empirical evaluation, the proposed methodology has been compared with a number of existing methods based on both synthetic and publicly available real smart home datasets with promising results obtained. We also discuss how the proposed unsupervised methodology can be used to support exploratory behavior analysis for elderly care.

Data summarization that presents a small subset of a dataset to users has been widely applied in numerous applications and systems. Many datasets are coded with hierarchical terminologies, e.g., the international classification of Diseases-9, Medical Subject Heading, and Gene Ontology, to name a few. In this paper, we study the problem of selecting a diverse set of k elements to summarize an input dataset with hierarchical terminologies, and visualize the summary in an ontology structure. We propose an efficient greedy algorithm to solve the problem with (1-1/e)≈ 62%-approximation guarantee. Preliminary experimental results on real-world datasets show the effectiveness and efficiency of the proposed algorithm for data summarization.

Most of the existing multi-relational network embedding methods, e.g., TransE, are formulated to preserve pair-wise connectivity structures in the networks. With the observations that significant triangular connectivity structures and parallelogram connectivity structures found in many real multi-relational networks are often ignored and that a hard-constraint commonly adopted by most of the network embedding methods is inaccurate by design, we propose a novel representation learning model for multi-relational networks which can alleviate both fundamental limitations. Scalable learning algorithms are derived using the stochastic gradient descent algorithm and negative sampling. Extensive experiments on real multi-relational network datasets of WordNet and Freebase demonstrate the efficacy of the proposed model when compared with the state-of-the-art embedding methods.

Diffusion on social networks refers to the process where opinions are spread via the connected nodes. Given a set of observed information cascades, one can infer the underlying diffusion process for social network analysis. The independent cascade model (IC model) is a widely adopted diffusion model where a node is assumed to be activated independently by any one of its neighbors. In reality, how a node will be activated also depends on how its neighbors are connected and activated. For instance, the opinions from the neighbors of the same social group are often similar and thus redundant. In this paper, we extend the IC model by considering that: 1) the information coming from the connected neighbors are similar and 2) the underlying redundancy can be modeled using a dynamic structural diversity measure of the neighbors. Our proposed model assumes each node to be activated independently by different communities (or components) of its parent nodes, each weighted by its effective size. An expectation maximization algorithm is derived to infer the model parameters. We compare the performance of the proposed model with the basic IC model and its variants using both synthetic data sets and a real-world data set containing news stories and Web blogs. Our empirical results show that incorporating the community structure of neighbors and the structural diversity measure into the diffusion model significantly improves the accuracy of the model, at the expense of only a reasonable increase in run-time.

In this paper, we address the problem of personalized next Point-of-interest (POI) recommendation which has become an important and very challenging task in location-based social networks (LBSNs), but not well studied yet. With the conjecture that, under different contextual scenario, human exhibits distinct mobility patterns, we attempt here to jointly model the next POI recommendation under the influence of user's latent behavior pattern. We propose to adopt a third-rank tensor to model the successive check-in behaviors. By incorporating softmax function to fuse the personalized Markov chain with latent pattern, we furnish a Bayesian Personalized Ranking (BPR) approach and derive the optimization criterion accordingly. Expectation Maximization (EM) is then used to estimate the model parameters. Extensive experiments on two large-scale LBSNs datasets demonstrate the significant improvements of our model over several state-of-the-art methods.

With the advent of the Internet of things (IoT) and smart sensor technologies, the data-driven paradigm has been found promising to support human behavioral analysis in a smart home for better healthcare and well-being of senior adults. This work focuses on discovering daily activity routines from sensor data collected in a smart home. By representing the sensor data as a matrix, daily activity routines can be identified using matrix factorization methods. The key challenge rests on the fact that the matrix contains discrete labels as its elements, and decomposing the nominal data matrix into basis vectors of the labels is nontrivial. We propose a novel principled methodology to tackle the nominal matrix factorization problem. Assuming that the similarity matrix of the labels is known, the discrete labels are first projected onto a continuous space with the interlabel distance preserving the given similarity matrix of the labels as far as possible. Then, we extend a hierarchical probabilistic model for ordinal matrix factorization with Bayesian Lasso that the factorization can be more robust to noise and more sparse to ease human interpretation. Our experimental results based on a synthetic data set shows that the factorization results obtained using the proposed methodology outperform those obtained using a number of the state-of-the-art factorization methods in terms of the basis vector reconstruction accuracy. We also applied our model to a publicly available smart home data set to illustrate how the proposed methodology can be used to support daily activity routine analysis.

Location-based social networks have been increasingly used to experience users new possibilities, including personalized point-of-interest (POI) recommendation services which leverages on the overlapping of user trajectories to recommend POI collaboratively. POI recommendation is challenging as it does not just suffers from the problems known for collaborative filtering such as data sparsity and cold-start, but to a much greater extent. Most of the related works apply the conventional recommendation approaches to POI recommendation while overlooking the personalized time-variant human behavioral tendency. In this paper, we put forward a tensor factorization-based ranking methodology to recommend users their interested locations by considering their time-varying behavioral trends. We also propose to categorize the locations to address data sparsity and cold-start issues, and accordingly new locations the user have not been visited can thus be bubbled up during ranking the location candidates. The tensor factorization is carefully studied to prune the irrelevant factors to the ranking results to achieve efficient POI recommendation. The experimental results validate the effectiveness of our proposed mechanism which outperforms the state-of-the-art approaches by over 8% for precision.

The Knowledge Discovery in Databases (KDD) comprises a number of stages, such as understanding its goals, acquiring, cleaning and integrating relevant data sources, modeling the obtained data sets by means of feature extraction and selection in order to prepare them as inputs to intelligent data analysis algorithms, and finally interpreting the results of those algorithms in practice.The success of KDD depends, among many other factors, on the ability to interact with the domain experts and the availability of effective data and information sharing.The domain knowledge is crucial for the design and tuning of practically all KDD stages.Privacy-preserving is crucial to run intelligent systems without violation of individual privacy by exposing potentially sensitive person-specific information.This special issue aims at elaborating on the state of the art in the areas of privacy-aware intelligent systems and the utilization of the domain knowledge in KDD.It is inspired by two workshops held at the 20th International Symposium on Methodologies for Intelligent Systems in Macau, China, 4-7 December 2012.The issue contains six carefully extended and thoroughly peer-reviewed articles, which were originally presented at the workshops.Their topics refer to data clustering and similarity, association rule induction and multi-label classification, as well as database engines, search systems and web services.Their results can contribute to a number of areas in our everyday lives, such as medical and healthcare systems, social and scientific networks, administrative services and so on.The first paper, by Jan Rauch, is titled "Formal Framework for Data Mining with Association Rules and Domain Knowledge -Overview of an Approach".The author outlines a methodology for data mining with association rules.The framework is based on a logical, appropriately enhanced calculus of rules.It allows to describe the whole knowledge discovery process, including formulation of analytical questions, application of the proposed analytical procedure and interpretation of its results.The domain knowledge is expressed using a formalized language aimed at deriving truly relevant rules.The second paper, by Nicolas Anciaux, Danae Boutara, Benjamin Nguyen and Michalis Vazirgiannis, is titled "Limiting Data Exposure in Multi-Label Classification Processes".The authors consider decision making processes based on multi-label classification models in the area of administrative services.They continue their previous research on minimizing data required from applicants as an input to online forms.They develop an experimental platform able to transform arbitrary multi-label data sets into collection rules and to measure the gain obtained in terms of minimizing data exposure.

Epidemic prediction is an important problem in epidemic control. Poisson regression methods are often adopted in existing works, mostly with only the (intra-)regional environmental factors considered. As the diffusion of epidemics is affected by not only the intra-regional factors but also inter-regional and external ones, a unified framework based on Poisson regression with the three types of factors incorporated is proposed for the prediction. Specifically, we propose a Poisson-regression-based model first with the intra-regional and inter-regional factors included. The intra-regional factor in a particular time interval is represented by one feature vector with the regionally environmental and social factors considered. The inter-regional factor is modeled by a diffusion matrix which describes the possibilities that the epidemics can spread from one region to another, which in turn accounts for the propagating effects of the infected cases. To learn the structure of the diffusion matrix, we propose two approaches-utilizing some a priori knowledge (e.g., transportation network) and estimating it from scratch via a sparse structure assumption. The resulting optimization problem of the maximum a posterior solution is a convex one and can be efficiently solved by the alternating direction method of multipliers (ADMM). In addition, we incorporate also the external factor, i.e., the imported cases. With one fact that the distribution of the number of infected cases over a year is (approximately) unimodal for most epidemics and one assumption that the importing rate has a small variance over the year, we can approximate the effect of the external factor with a parametric function (e.g., a quadratic function) over time. The resulting optimization problem is still convex and can be also solved by the ADMM algorithm. Empirical evaluations are conducted based on a real data set which records the 16-days-reported cases in the Yunnan province of China for seven years, from 2005 to 2011. The experimental results demonstrate the effectiveness of our proposed models.

Many real-world problems exhibit phenomena which are best represented as complex networks with dynamic structures (e.g., human communication networks). Network motifs have been shown effective for characterizing the structural properties of such complex networks. Nevertheless, related motif models typically do not consider stochastic structural and sequential variations, hinting their limitations on dynamic network analysis. In this paper, we consider networks with time-stamped edges and model their local structural and temporal variations using a mixture of Markov chains for stochastic temporal network motif detection. The optimal number of motifs is automatically estimated in a Bayesian framework. We evaluated the proposed method using synthetic networks and found to be robust against noise compared to the deterministic approach. Also, we applied it to a mobile phone usage data set to demonstrate how the human communication patterns embedded in the data set can be detected. In addition, we make use of a hidden Markov model with different distributions for the mixing proportions of the motifs defining its states, and demonstrated how the evolution of the communication patterns can also be identified.

Diffusion is known to be an important process governing the behaviours observed in network environments like social networks, contact networks, etc. For modeling the diffusion process, the Independent Cascade Model (IC Model) is commonly adopted and algorithms have been proposed for recovering the hidden diffusion network based on observed cascades. However, the IC Model assumes the effects of multiple neighbors on a node to be independent and does not consider the structural diversity of nodes' neighbourhood. In this paper, we propose an extension of the IC Model with the community structure of node neighbours incorporated. We derive an expectation maximization (EM) algorithm to infer the model parameters. To evaluate the effectiveness and efficiency of the proposed method, we compared it with the IC model and its variants that do not consider the structural properties. Our empirical results based on the MemeTracker dataset, shows that after incorporating the structural diversity, there is a significant improvement in the modelling accuracy, with reasonable increase in run-time.

Latent Dirichlet allocation (LDA) is an important hierarchical Bayesian model for probabilistic topic modeling, which attracts worldwide interest and touches on many important applications in text mining, computer vision and computational biology. This paper represents the collapsed LDA as a factor graph, which enables the classic loopy belief propagation (BP) algorithm for approximate inference and parameter estimation. Although two commonly used approximate inference methods, such as variational Bayes (VB) and collapsed Gibbs sampling (GS), have gained great success in learning LDA, the proposed BP is competitive in both speed and accuracy, as validated by encouraging experimental results on four large-scale document datasets. Furthermore, the BP algorithm has the potential to become a generic scheme for learning variants of LDA-based topic models in the collapsed space. To this end, we show how to learn two typical variants of LDA-based topic models, such as author-topic models (ATM) and relational topic models (RTM), using BP based on the factor graph representations.

Ubiquitous computing aims to enable and support anywhere, anytime, context-aware applications. Sensing, interpretation and integration of events, behaviors and environmental states have been keys to the success of such ubiquitous systems. Over the past two decades, there has been a constant shift of sensor observation modeling, representation, interpretation and usage, namely from low-level raw observation data and their direct/hardwired usage, data aggregation and fusion, to high-level formal context modeling and context-based computing. It is envisioned that this trend will continue towards a further higher level of abstraction, allowing situation, activity and goal modeling, representation and inference, thus realizing the vision of ubiquitous computing. The proposed "mini-track" workshop intends to bring together researchers and practitioners from relevant fields to present and disseminate the latest accomplished and/or ongoing research on Situation, Activity and Situation Awareness (SAGAware) and their novel application in ubiquitous computing. It aims to facilitate knowledge transfer and synergy, bridge gaps between different research communities/groups, lay down foundation for common purposes, and help identify opportunities and challenges for interested researchers and technology and system developers.

This report summarizes the International Workshop on Situation, Activity and Goal Awareness held at the 13th ACM International Conference on Ubiquitous Computing, on September 18 in Beijing, China.

The increasing availability of large quantities of user contributed images with labels has provided opportunities to develop automatic tools to tag images to facilitate image search and retrieval. In this paper, we present a novel hybrid probabilistic model (HPM) which integrates low-level image features and high-level user provided tags to automatically tag images. For images without any tags, HPM predicts new tags based solely on the low-level image features. For images with user provided tags, HPM jointly exploits both the image features and the tags in a unified probabilistic framework to recommend additional tags to label the images. The HPM framework makes use of the tag-image association matrix (TIAM). However, since the number of images is usually very large and user-provided tags are diverse, TIAM is very sparse, thus making it difficult to reliably estimate tag-to-tag co-occurrence probabilities. We developed a collaborative filtering method based on nonnegative matrix factorization (NMF) for tackling this data sparsity issue. Also, an L1 norm kernel method is used to estimate the correlations between image features and semantic concepts. The effectiveness of the proposed approach has been evaluated using three databases containing 5,000 images with 371 tags, 31,695 images with 5,587 tags, and 269,648 images with 5,018 tags, respectively.

This paper presents the coauthor network topic (CNT) model constructed based on Markov random fields (MRFs) with higher-order cliques. Regularized by the complex coauthor network structures, the CNT can simultaneously learn topic distributions as well as expertise of authors from large document collections. Besides modeling the pairwise relations, we model also higher-order coauthor relations and investigate their effects on topic and expertise modeling. We derive efficient inference and learning algorithms from the Gibbs sampling procedure. To confirm the effectiveness, we apply the CNT to the expert finding problem on a DBLP corpus of titles from six different computer science conferences. Experiments show that the higher-order relations among coauthors can improve the topic and expertise modeling performance over the case with pairwise relations, and thus can find more relevant experts given a query topic or document.

Partially observable Markov decision process (POMDP) is a commonly adopted mathematical framework for solving planning problems in stochastic environments. However, computing the optimal policy of POMDP for large-scale problems is known to be intractable, where the high dimensionality of the underlying belief space is one of the major causes. In this paper, we propose a hybrid approach that integrates two different approaches for reducing the dimensionality of the belief space: 1) belief compression and 2) value-directed compression. In particular, a novel orthogonal nonnegative matrix factorization is derived for the belief compression, which is then integrated in a value-directed framework for computing the policy. In addition, with the conjecture that a properly partitioned belief space can have its per-cluster intrinsic dimension further reduced, we propose to apply a k-means-like clustering technique to partition the belief space to form a set of sub-POMDPs before applying the dimension reduction techniques to each of them. We have evaluated the proposed belief compression and clustering approaches based on a set of benchmark problems and demonstrated their effectiveness in reducing the cost for computing policies, with the quality of the policies being retained.

In this paper we propose the multirelational topic model (MRTM) for multiple types of link modeling such as citation and coauthor links in document networks. In the citation network, the MRTM models the citation link between each pair of documents as a binary variable conditioned on their topic distributions. In the coauthor network, the MRTM models the coauthor link between each pair of authors as a binary variable conditioned on their expertise distributions. The topic discovery is collectively regularized by multiple relations in both citation and coauthor networks. This model can summarize topics from the document network, predict citation links between documents and coauthor links between authors. Efficient inference and learning algorithms are derived based on Gibbs sampling. Experiments demonstrate that the MRTM significantly outperforms other state-of-the-art single-relational link modeling methods for large scientific document networks.

Real-world social networks (e.g., blogosphere) often evolve over time and thus poses challenges on conventional social network analysis techniques which model the underlying networks as static graphs. In this paper, we are interested in detecting dynamic communities and their trend of evolution in a social network by examining the structural and dynamic patterns of interactions. In doing so, we propose an iterative mining algorithm for computing the intensities and bursts of some hidden communities over time. Our method is probabilistic in nature and can be applied to both undirected graphs and directed graphs. Quantitative and qualitative performance comparisons between the proposed method and some representative methods for social network analysis are provided. Evaluation results based on three benchmark datasets, including Reuters terror news network, political blogosphere, and Enron emails, show that the proposed method is both effective and efficient.

Many on-line photo sharing systems allow users to tag their images so as to support semantic image search. In this paper, we study how one can take advantages of the already-tagged images to (semi-)automate the labeling of newly uploaded ones. In particular, we propose a hybrid approach for the prediction where user-provided tags and image visual contents are fused under a unified probabilistic framework. Kernel smoothing and collaborative filtering techniques are explored for improving the accuracy of the probabilistic models estimation. By comparing with some state-of-the-art content-based image labeling methods, we have empirically shown that 1) the proposed method can achieve comparable tag prediction accuracy when there is no user-provided tag, and that 2) it can significantly boost the prediction accuracy if the user can provide just a few tags.

High dimensionality of POMDP's belief state space is one major cause that makes the underlying optimal policy computation intractable. Belief compression refers to the methodology that projects the belief state space to a low-dimensional one to alleviate the problem. In this paper, we propose a novel orthogonal non-negative matrix factorization (O-NMF) for the projection. The proposed O-NMF not only factors the belief state space by minimizing the reconstruction error, but also allows the compressed POMDP formulation to be efficiently computed (due to its orthogonality) in a value-directed manner so that the value function will take same values for corresponding belief states in the original and compressed state spaces. We have tested the proposed approach using a number of benchmark problems and the empirical results confirms its effectiveness in achieving substantial computational cost saving in policy computation.

Many complex systems in the real world can be modeled as signed social networks that contain both positive and negative relations. Algorithms for mining social networks have been developed in the past, however most of them were designed primarily for networks containing only positive relations and thus not suitable for signed networks. In this work, we propose a new algorithm, called FEC, to mine signed social networks so that both positive within-group relations and negative between-group relations are dense. FEC considers both the sign and the density of relations as the clustering attributes, making itself effective for not only signed networks but also conventional social networks including only positive relations. Also, FEC adopts an agent-based heuristic that makes the algorithm efficient (in linear time with respect to the size of a network) and capable of giving nearly optimal solutions. FEC depends on only one parameter whose value can easily be set, and requires no prior knowledge on hidden community structures. The effectiveness and efficacy of FEC have been demonstrated through a set of rigorous experiments involving both benchmark and randomly-generated signed networks.

With the recent adoption of marketing activities outsourcing, there have been increasing demands and concerns for privacy control. The traditional approach of a bulk transmission of the customers' information to a marketing company cannot meet such demands, especially in the finance and healthcare businesses. Therefore, we propose a layered architecture and a development methodology for end-to-end privacy control over the export of each individual customer's records through a Web services platform, according to the corresponding enterprise's privacy control policies. A Web services system, with up-dated security and privacy facilities, can provide a suitable interoperation platform for required application-to-application interactions over the Internet. We further develop a conceptual model and an interaction protocol to send only the required part of a customer's records at a time. We illustrate our approach for end-to-end privacy control with a tele-marketing case study and show how the software of the outsourced call center can be integrated effectively with the Web services of a bank to protect privacy. Copyright 2005 ACM.

Mining distributed data for global knowledge is getting more attention recently. The problem is especially challenging when data sharing is prohibited due to local constraints like limited bandwidth and data privacy. In this paper, we investigate how to derive the embedded manifold (as a 2-D map) for a horizontally partitioned data set, where data cannot be shared among the partitions directly. We propose a model-based approach which computes hierarchical local data abstractions, aggregates the abstractions, and finally learns a global generative model - generative topographic mapping (GTM) based on the aggregated data abstraction. We applied the proposed method to two benchmarking data sets and demonstrated that the accuracy of the derived manifold can effectively be controlled by adjusting the data granularity level of the adopted local abstraction.

In order to meet the on-demand requirement of service composition at a large scale, one should go beyond the use of static domain ontologies but allow different focused aspects of the ontologies to be distributed as sub-ontologies. To demonstrate the feasibility of the sub-ontology idea, we suggest a possible implementation using the semantic Web technology and apply it to distributed e-learning.

Structured link vector model (SLVM) is a recently proposed document representation that takes into account both structural and semantic information for measuring XML document similarity. Its formulation includes an element similarity matrix for capturing the semantic similarity between XML elements - the structural components of XML documents. In this paper, instead of applying heuristics to define the similarity matrix, we proposed to learn the matrix using pair wise similar training data in an iterative manner. In addition, we extended SLVM to SLVM-LSI by incorporating term semantics into SLVM using latent semantic indexing, with the element similarity related properties of the original SLVM preserved. For performance evaluation, we applied SLVM-LSI to similarity-based clustering of two XML datasets and the proposed SLVM-LSI was found to significantly outperform the conventional vector space model and the edit-distance based methods. The similarity matrix, obtained as a byproduct via the learning, can provide higher level knowledge about the semantic relationship between the XML elements.

The next generation Web Intelligence (WI) aims at enabling users to go beyond the existing online information search and knowledge queries functionalities and to gain, from the Web, practical wisdom for problem solving. To support such a Wisdom Web, we envision that a grid‐like computing infrastructure with intelligent service agencies is needed, where these agencies can interact, self‐organize, learn, and evolve their course of actions, identities, and interrelationships for new knowledge creation, as well as scientific and social evolution. In this paper, we first provide an overview of recent development in WI and Semantic/Knowledge Grid. Then, the fundamental capabilities of the Wisdom Web as well as the conceptual architecture of an intelligent Grid for supporting it are described. Technical challenges for realizing Grid Intelligence are highlighted and the recent advancements in related research areas are reviewed.

Web services are becoming important in applications from electronic commerce to application interoperation. While numerous efforts have focused on service composition, service selection among similar services from multiple providers has not been addressed. Such issue is more serious when services are embraced in Grid platforms, which are usually resource-conscious. Experimental results show that our considerations are valid and our preliminary solution works well in our Globus grid network.

While numerous efforts have focused on service composition in the Grid environment, service selection among similar services from multiple providers has not been addressed. In particular, all service composition work done so far are based on a given selection of services under a well set environment. As a result, uncertainty (e.g., server load, network traffic, computation time of the services due to changing memory and other unexpected conditions) under a real, dynamic environment has never been considered. This paper prototypes the service selection under a Grid environment and proposes an uncertainty framework to address the issue. Experimental results show that our considerations are valid and our preliminary solution works well in our Globus Grid network.

Distributed data mining has been a topic getting more important nowadays as there are many cases where physically sharing of data is probibited, e.g., due to huge data volume or data privacy. In this paper, we are interested in learning a global cluster model by exploring data in distributed sources. A methodology based on periodic model exchange and merge is proposed and applied to hyperlinked Web pages analysis. In addition, we have tested a number of variations of the basic idea, including putting more emphasis on the privacy concern and testing the effect of having different numbers of distributed sources. Experimental results show that the proposed distributed learning scheme is effective with accuracy close to the case with all the data physically shared for the learning.

The Centre for e-Transformation Research (CTR), also an affiliated Centre of Web Intelligence Consortium (WIC), is established under the Science Faculty of Hong Kong Baptist University, currently funded by Hong Kong Research Grant Council Central Allocation and FRG Strategic Research Grant, Hong Kong Baptist University for developing an Area of Strength in e-transformation research, making high impact to various sectors of the society, from e-business, e-learning, to e-government, to name a few.

With the advent of the World Wide Web, providing just-in-time personalized product recommendations to customers now becomes possible. Collaborative recommender systems utilize correlation between customer preference ratings to identify "like-minded" customers and predict their product preference. One factor determining the success of the recommender systems is the prediction accuracy, which in many cases is limited by lacking adequate ratings (the sparsity problem). Recently, the use of latent class model (LCM) has been proposed to alleviate this problem. In this paper, we first study how the LCM can be extended to handle customers and products outside the training set. In addition, we propose the use of a pair of LCMs (called dual latent class model-DLCM), instead of a single LCM, to model customers' likes and dislikes separately for enhancing the prediction accuracy. Experimental results based on the EachMovie dataset show that DLCM outperforms both LCM and the conventional correlation-based method when the available ratings are sparse.

Foraging information in large and complex Web sites simply using keyword search usually results in unpleasant experience due to the overloaded search results. To support more effective information search, some descriptive abstractions of the Web sites (e.g., sitemaps) are mostly needed. However, their creation and maintenance normally requires recurrent manual effort due to the fast-changing Web contents. We extend the HITS algorithm and integrate hyperlink topology and Web site hierarchy to identify a hierarchy of Web page clusters as the abstraction of a Web site. As the algorithm is based on HITS, each identified cluster follows the bipartite graph structure, with an authority and hub pair as the cluster summary. The effectiveness of the algorithm has been evaluated using three different Web sites (containing /spl sim/6000-14000 Web pages) with promising results. Detailed interpretation of the experimental results as well as qualitative comparison with other related works are also included.

To achieve integrated segmentation and recognition in complex scenes, the model-based approach has widely been accepted as a promising paradigm. However, the performance is still far from satisfactory when the target object is highly deformed and the level of outlier contamination is high. In this paper, we first describe two Bayesian frameworks, one for classifying input patterns and another for detecting target patterns in complex scenes using deformable models. Then, we show that the two frameworks are similar to the forward-reverse setting of Hausdorff matching and that their matching and discriminating properties are complementary to each other. By properly combining the two frameworks, we propose a new matching scheme called bidirectional matching. This combined approach inherits the advantages of the two Bayesian frameworks. In particular, we have obtained encouraging empirical results on shape-based pattern extraction, using a subset of the CEDAR handwriting database containing handwritten words of highly varying shape.

A Bayesian framework for deformable pattern classification was proposed by K.W. Cheung et al. (1998), with promising results for isolated handwritten character recognition. Its performance, however degrades significantly when it is applied to detect deformable patterns in complex scenes, where the amount of outliers due to other neighboring objects or the background is usually large. Also, the fact that the associated evidence measure does not penalize models resting on white space results in a high false alarm rate. Another Bayesian framework for deformable pattern detection is proposed. The framework possesses the intrinsic property of matching with only part of an image (segmentation) and its associated evidence measure can penalize white space implicitly. However, limited data exploration capability is the major trade-off. By properly combining the two frameworks, a new matching algorithm called bidirectional matching is proposed. This combined approach possesses the advantages of the two frameworks and gives robust results for non-rigid shape extraction. To evaluate the performance of the proposed approach, we have applied it to shape-based handwritten word retrieval. Using a subset of the bb dataset in the CEDAR database, we can achieve a recall rate of 59% and a precision rate of 43%.

Deformable models have recently been proposed for many pattern recognition applications due to their ability to handle large shape variations. These proposed approaches represent patterns or shapes as deformable models, which deform themselves to match with the input image, and subsequently feed the extracted information into a classifier. The three components-modeling, matching, and classification-are often treated as independent tasks. In this paper, we study how to integrate deformable models into a Bayesian framework as a unified approach for modeling, matching, and classifying shapes. Handwritten character recognition serves as a testbed for evaluating the approach. With the use of our system, recognition is invariant to affine transformation as well as other handwriting variations. In addition, no preprocessing or manual setting of hyperparameters (e.g., regularization parameter and character width) is required. Besides, issues on the incorporation of constraints on model flexibility, detection of subparts, and speed-up are investigated. Using a model set with only 23 prototypes without any discriminative training, we can achieve an accuracy of 94.7 percent with no rejection on a subset (11,791 images by 100 writers) of handwritten digits from the NIST SD-1 dataset.

Following the success of applying deformable models to feature extraction, a natural next step is to apply such models to pattern classification. Recently, we have cast a deformable model under a Bayesian framework for classification, giving promising results. However, deformable model methods are computationally expensive due to the required iterative optimization process. The problem is even more severe when there are a large number of models (e.g., for character recognition), because each of them has to deform and match with the input data before a final classification can be derived. In this paper, we propose to combine the deformable models into a mixture, in which the individual models compete with each other to survive the matching process during classification. Models that do not compete well are eliminated early, thus allowing substantial savings in computation. This process of competition-elimination has been applied to handwritten digit recognition in which significant speedup can be achieved without sacrificing recognition accuracy.

A new approach to indoor propagation prediction which is based on the Rayleigh-Gans approximation is proposed. The approach has been motivated by the observation that the indoor environment is different from the outdoor environment in that many of the structures within buildings are approximately the size of a wavelength making propagation prediction based on high frequency techniques not strictly applicable. In the paper details of the Rayleigh-Gans approximation are provided. Actual indoor propagation measurements are also provided to allow preliminary discussions on the accuracy of the Rayleigh-Gans approximation to be given. Comparisons with simulations of empirical and ray tracing models are also provided. The results indicate that further development would be worthwhile.>