Graph anomaly detection (GAD), which aims to detect outliers in graph-structured data, has received increasing research attention recently. However, existing GAD methods assume identical training and testing distributions, which is rarely valid in practice. In real-world scenarios, unseen but normal samples may emerge during deployment, leading to a normality shift that degrades the performance of GAD models trained on the original data. Through empirical analysis, we reveal that the degradation arises from (1) semantic confusion, where unseen normal samples are misinterpreted as anomalies due to their novel patterns, and (2) aggregation contamination, where the representations of seen normal nodes are distorted by unseen normals through message aggregation. While retraining or fine-tuning GAD models could be a potential solution to the above challenges, the high cost of model retraining and the difficulty of obtaining labeled data often render this approach impractical in real-world applications. To bridge the gap, we proposed a lightweight and plug-and-play Test-time adaptation framework for correcting Unseen Normal pattErns (TUNE) in GAD. To address semantic confusion, a graph aligner is employed to align the shifted data to the original one at the graph attribute level. Moreover, we utilize the minimization of representation-level shift as a supervision signal to train the aligner, which leverages the estimated aggregation contamination as a key indicator of normality shift. Extensive experiments on 10 real-world datasets demonstrate that TUNE significantly enhances the generalizability of pre-trained GAD models to both synthetic and real unseen normal patterns.

Junjun Pan, Yixin Liu, Rui Miao, Kaize Ding, Yu Zheng, Quoc Viet Hung Nguyen, Alan Wee-Chung Liew, Shirui Pan. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Model transferability estimation is a task-adaptive pre-trained model selection problem, aiming to determine the optimal model for target dataset from a model hub pre-trained on source dataset without fine-tuning. Although existing model transferability evaluation methods have made some progress, they mainly focus on image or text data in CV and NLP. In contrast, the graph structural data with GNNs models is still underexplored, due to the complexity of the graph structure and the limitations of the generalization ability of GNN models under distribution shift. To fill this blank, we first propose a Graph Neural Network Model Transferability Estimation method via decomposition-augmented discriminant analysis, named GNNMTE, to evaluate the transferability of GNN models on target graph dataset without fine-tuning. It only calculates the GNNMTE score to determine whether it can be effectively transferred to target graph dataset and better select the optimal model for the target graph dataset. Specifically, our proposed \method contains three core components: (1) Dual-block SVD fusion for obtaining the corresponding principal component information; (2) Adaptive weighting by singular value ratio for guiding the extraction of important principal component information on graph data; (3) Graph discriminant analysis for finding the optimal projection direction that separates the classes of graph data. Extensive experimental results on cross-domain graph datasets achieve excellent results, demonstrating powerful superiority.

Unsupervised graph domain adaptation (UGDA) aims to transfer knowledge from a labeled source graph to an unlabeled target graph, addressing the performance degradation caused by distributional shifts in node attributes and graph structures across domains. Despite recent progress, existing UGDA approaches still face two key challenges: (C1) Data-level: Most methods rely on a single source domain, overlooking the complementary knowledge that could be leveraged from multiple sources. (C2) Model-level: Many UGDA models emphasize complex, handcrafted Graph neural network (GNN) architectures, while simpler yet effective designs with propagation (P) & transformation (T) pipeline remain underexplored. To address these challenges, in this paper, we propose a novel approach, which leverages Concise Propagation–Transformation pipeline for multi-source unsupervised Graph Domain Adaptation, dubbed as CPT-GDA, to better capture complementary knowledge from multiple sources in an efficient manner. Specifically, the proposed CPT-GDA adopts a dual-branch GNN architecture with different depths of propagation but the same P-T patterns, which enables the model to efficiently learn node representations to mitigate domain discrepancy. Meanwhile, to facilitate effective knowledge transfer across graphs, we derive three optimization objectives: (1) the classifier loss to learn discriminative representations; (2) the alignment loss weighted by the graph Wasserstein distance to align the structure and feature distribution; and (3) the pseudo-label loss to refine target node representations. Extensive experiments on real-world datasets confirm that the proposed method outperforms recent state-of-the-art baselines, demonstrating its effectiveness.

Traditional Federated Learning (FL) focuses on collaborative global model training while ensuring privacy and personalization. Decentralized Federated Learning (DFL), a variant of FL, allows clients to independently manage and optimize local models without a central server. DFL reduces centralized communication bottlenecks and vulnerability to server failures or attacks. However, because the optimization dynamics change and there is no global model, generalization can suffer, making effective learning under data and model heterogeneity a critical challenge in DFL. Despite growing interest in DFL, the lack of distributional and uncertainty modeling in the literature limits reliability and effective generalization in non-IID settings. In this work, we propose DFedDG2, a personalized federated learning framework that operates within a peer-to-peer protocol. DFedDG2 offers the technical advantage of modeling each client’s local data distribution and exchanging this information with one-hop neighbors. In the decentralized network, clients perform distribution-aware gossip, where statistically similar clients exert greater influence to drive global alignment. This likelihood-weighted mixing fuses only a handful of vectors and scalars, significantly reducing communication costs while aligning semantic spaces across the network and enabling personalized training at the edge. In addition, theoretically, we prove that DFedDG2 achieves a sublinear convergence rate while the consensus error decays at a geometric rate under well-principled properties of gossip. Unlike the label-only non-IID experiments in DFL literature, we conduct extensive experiments on multiple data non-IID scenarios, topology variations, model heterogeneity, and uncertainty quantification, demonstrating the practical advantages of DFedDG2. Our results show that DFedDG2 not only achieves communication efficiency but also provides better generalization and improved reliability compared to state-of-the-art approaches.

Graph fraud detection (GFD) has rapidly advanced in protecting online services by identifying malicious fraudsters. Recent supervised GFD research highlights that heterophilic connections between fraudster and user greatly impacts detection performance, where the fraudsters tend to camouflage themselves by building more connections to benign users. Despite their promising performance, their label reliance limits its application in unsupervised scenarios; Additionally, accurately capturing complex and diverse heterophily patterns without labels poses a further challenge. Therefore, we propose a Heterophily-guided Unsupervised Graph fraud dEtection approach (HUGE) for unsupervised GFD, which contains two essential components: a heterophily estimation module and an alignment-based fraud detection module. In the heterophily estimation module, we design a novel unsupervised heterophily metric called HALO, which captures the critical graph properties for GFD, enabling its outstanding ability to estimate heterophily with attributes. In the alignment-based fraud detection module, we develop a joint MLP-GNN architecture with ranking loss and asymmetric alignment loss. The ranking loss aligns the predicted fraud score with the relative order of HALO, providing an extra robustness guarantee by comparing heterophily between non-adjacent nodes. Moreover, the asymmetric alignment loss effectively utilizes structural information to alleviate the feature-smooth effects. Extensive experiments on six datasets demonstrate that HUGE consistently outperforms competitors, showcasing its effectiveness and robustness.

Ocean salinity plays a vital role in circulation, climate, and marine ecosystems, yet its measurement is often sparse, irregular, and noisy, especially in drifter-based datasets. Traditional approaches, such as remote sensing and optimal interpolation, rely on linearity and stationarity, and are limited by cloud cover, sensor drift, and low satellite revisit rates. While machine learning models offer flexibility, they often fail under severe sparsity and lack principled ways to incorporate physical covariates without specialized sensors. In this paper, we introduce the OceAn Salinity Imputation System, a novel diffusion adversarial framework designed to address these challenges by: (1) employing a transformer-based global dependency capturing module to learn long-range spatio-temporal correlations from sparse trajectories; (2) constructing a generative imputation model that conditions on easily observed tidal covariates to progressively refine imputed salinity fields; and (3) using a scheduler diffusion method to enhance the model's robustness. This unified architecture exploits the periodic nature of tidal signals as a proxy for unmeasured physical drivers, without the need for additional equipment. We evaluate OASIS on four benchmark datasets, including one real-world measurement from Fort Pierce Inlet and three simulated Gulf of Mexico trajectories. Results show consistent improvements over both traditional and neural baselines, achieving up to 52.5% reduction in MAE compared to Kriging. We also develop a lightweight, web-based deployment system that enables salinity imputation through interactive and batch interfaces, available at: https://github.com/yfeng77/OASIS.

Conversational Recommender Systems (CRSs) aim to provide personalized recommendations by capturing user preferences through interactive dialogues. Explainability in CRSs is crucial as it enables users to understand the reasoning behind recommendations, increasing system transparency and trustworthiness. However, current CRSs often leverage knowledge graphs (KGs) or language models to extract and represent user preferences as latent vectors, which limits their explainability. Large language models (LLMs) offer powerful reasoning capabilities that can bridge this gap by generating human-understandable preference summaries. However, effectively reasoning over user preferences in CRSs remains challenging as LLMs pre-trained on large-scale corpora may not be well-suited for analyzing user preferences, which requires domain-specific knowledge. While KGs provide rich domain knowledge, integrating them with LLMs encounters a significant modality gap between structured KG information and unstructured conversations. In this paper, we propose COMPASS (Compact Preference Analyzer and Summarization System), a plug-and-play framework that synergizes LLMs and KGs to reason over user preferences, enhancing the performance and explainability of existing CRSs. COMPASS employs a two-stage training approach: first, it bridges the gap between the structured KG and natural language through novel graph entity captioning pre-training. This enables the LLM to transform KG entities into natural language descriptions, allowing it to comprehend domain-specific knowledge. Next, COMPASS optimizes user preference reasoning via knowledgeaware instruction fine-tuning, where the LLM learns to reason and summarize user preferences from dialogue histories and KG-augmented context. This enables COMPASS to perform knowledge-aware reasoning and generate interpretable user preferences that can seamlessly integrate with existing CRS models for improving recommendation performance and explainability. Our experiments on benchmark datasets demonstrate the effectiveness of COMPASS in improving various CRS models.

Recent advancements in test-time adaptation (TTA) offer promising solutions for mitigating performance degradation caused by distribution shifts. However, they may fall short in practical deployment of GNNs during test time, due to the significant reliance on impractical retraining or modifications to online GNN models. To address such challenges, in this work, we propose a novel method, i.e., Test-Time Graph REBirth, dubbed TT-GREB, to effectively generalize the well-trained GNN models to the test-time graphs under distribution shifts by directly manipulating the test graph data. Concretely, we develop an overall framework designed by two principles for obtaining newly reborn test graphs, corresponding to two sub-modules: (1) prototype extractor for re-extracting the environment-invariant features of the test-time graph; and (2) environment refiner for re-fining the environment-varying features to explore the potential shifts. Furthermore, we propose a dual test-time graph contrastive learning objective with an effective iterative optimization strategy to obtain optimal prototype components and environmental components of the test graph. Extensive experiments on real-world graphs under diverse test-time distribution shifts verify the effectiveness of our proposed method, showcasing its superior ability to manipulate test-time graphs for better GNN generalization ability.

Short-term electric load forecasting is essential for efficient power system operation, but existing deep learning models struggle to capture the multi-scale features and cyclical fluctuations inherent in short-term load data. This paper introduces a novel deep learning model, Wavelet Transform Convolution-inverted ProbSparse Transformer (WTC-iPST), specifically designed for short-term load forecasting. Unlike existing deep learning models, WTC-iPST leverages Wavelet Transform Convolution (WTConv) for multi-scale feature extraction and integrates Wavelet Kolmogorov-Arnold Networks (Wav-KAN) to enhance the ProbSparse self-attention mechanism, significantly improving the model's ability to capture multi-scale features and cyclical fluctuations inherent in short-term load data. This design addresses the challenge of extracting multi-scale and cyclical features from short-term load data, which existing models struggle with, and strengthens the model's capacity to handle long series. Additionally, WTC-iPST incorporates quantile regression to quantify uncertainty and provide confidence intervals, further enhancing the prediction's reliability and accuracy. Experimental results on real-world datasets demonstrate that WTC-iPST outperforms state-of-the-art forecasting models, with significant improvements over the baseline iTransformer, achieving reductions of up to 16.84% in RMSE, 18.09% in MAPE, and 17.65% in RRMSE, as well as an increase of up to 2.96% in R². In terms of probabilistic prediction, WTC-iPST consistently maintains a narrow confidence interval with high interval coverage. Moreover, WTC-iPST shows strong performance across various prediction horizons and different distribution substations, highlighting its robustness and adaptability. These results confirm that WTC-iPST provides more accurate and reliable forecasts, making it a valuable tool for power system dispatch and operational planning.

The rapid proliferation of Internet of Things (IoT) devices has exposed us to a growing threat landscape. Advancements in artificial intelligence (AI), particularly deep learning (DL) techniques, offer innovative solutions for trustworthy IoT attack detection. Despite the resource-constrained nature of IoT devices, most existing AI-based IoT attack detection methods rely on complex deep learning models for automated feature extraction and generalization. However, DL models are inherently overconfident and tend to produce unreliable predictions both in-distribution and under distribution shifts when deployed. The large model size, coupled with deterministic and overconfident predictions, and limited focus on estimating the uncertainty, hinders the trustworthy deployment of DL models in applications where reliability is critical, not just high accuracy. In this paper, we present a framework for confidence calibration and uncertainty estimation of DL models in the context of IoT attack detection. Specifically, we propose a novel framework for the trustworthy design of deep IoT attack detection models, along with a lightweight DL model that achieves state-of-the-art (SoTA) attack detection performance. We demonstrate how both our DL model and other SoTA DL models can be effectively integrated into the proposed framework to ensure trustworthy IoT attack detection. Evaluated on the large-scale benchmark IoT-23 network traffic dataset in terms of calibration, uncertainty, and distribution shift, our approach delivers superior and trustworthy IoT attack detection performance.

Today, computer systems hold large amounts of personal data. Yet while such an abundance of data allows breakthroughs in AI, and especially machine learning, its existence can be a threat to user privacy, and it can weaken the bonds of trust between humans and AI. Recent regulations now require that, on request, private information about a user must be removed both from computer systems and from machine learning models—this legislation is more colloquially called “the right to be forgotten.” While removing data from back-end databases should be straightforward, it is not sufficient in the AI context as machine learning models often “remember” the old data. Contemporary adversarial attacks on trained models have proven that we can learn whether an instance or an attribute belonged to the training data. This phenomenon calls for a new paradigm, namely machine unlearning , to make machine learning models forget about particular data. It turns out that recent works on machine unlearning have not been able to completely solve the problem due to the lack of common frameworks and resources. Therefore, this article aspires to present a comprehensive examination of machine unlearning’s concepts, designs, methods, and applications. Specifically, as a category collection of cutting-edge studies, the intention behind this article is to serve as a comprehensive resource for researchers and practitioners seeking an introduction to machine unlearning and its formulations, design criteria, removal requests, algorithms, and applications. In addition, we aim to highlight the key findings, current trends, and new research areas that have not yet featured the use of machine unlearning but could benefit greatly from it. We hope that this survey serves as a valuable resource for machine learning researchers and those seeking to innovate privacy technologies. Our resources are publicly available at https://github.com/tamlhp/awesome-machine-unlearning .

Modern recommender systems derive predictions from an interaction graph that links users and items. To this end, many of today's state-of-the-art systems use graph neural networks (GNNs) to learn effective representations of these graphs under the assumption of homophily, i.e., the idea that similar users will sit close to each other in the graph. However, recent studies have revealed that real-world recommendation graphs are often heterophilous, i.e., dissimilar users will also often sit close to each other. One of the reasons for this heterophilia is shilling attacks that obscure the inherent characteristics of the graph and make the derived recommendations less accurate as a consequence. Hence, to cope with low homophily in recommender systems, we propose a recommendation model called PGT4Rec that is based on a Partitioned Graph Transformer. The model integrates label information into the learning process, which allows discriminative neighbourhoods of users to be generated. As such, the framework can both detect shilling attacks and predict user ratings for items. Extensive experiments on real and synthetic datasets show PGT4Rec as not only providing superior performance in these two tasks but also significant robustness to a range of adversarial conditions.

For incomplete data classification, missing attribute values are often estimated by imputation methods before building classifiers. The estimated attribute values are not actual attribute values. Thus, the distributions of data will be changed after imputing, and this phenomenon often results in degradation of classification performance. Here, we propose a new framework called integration of multikinds imputation with covariance adaptation (MICA) based on evidence theory (ET) to effectively deal with the classification problem with incomplete training data and complete test data. In MICA, we first employ different kinds of imputation methods to obtain multiple imputed training datasets. In general, the distributions of each imputed training dataset and test dataset will be different. A covariance adaptation module (CAM) is then developed to reduce the distribution difference of each imputed training dataset and test dataset. Then, multiple classifiers can be learned on the multiple imputed training datasets, and they are complementary to each other. For a test pattern, we can combine the multiple pieces of soft classification results yielded by these classifiers based on ET to obtain better classification performance. However, the reliabilities/weights of different imputed training datasets are usually different, so the soft classification results cannot be treated equally during fusion. We propose to use covariance difference across datasets and accuracy of imputed training data to estimate the weights. Finally, the soft classification results discounted by the estimated weights are combined by ET to make the final class decision. MICA was compared with a variety of related methods on several datasets, and the experimental results demonstrate that this new method can significantly improve the classification performance.

Conversational recommender systems (CRSs) utilize natural language interactions and dialog history to infer user preferences and provide accurate recommendations. Due to the limited conversation context and background knowledge, existing CRSs rely on external sources such as knowledge graphs (KGs) to enrich the context and model entities based on their interrelations. However, these methods ignore the rich intrinsic information within entities. To address this, we introduce the knowledge-enhanced entity representation learning (KERL) framework, which leverages both the KG and a pretrained language model (PLM) to improve the semantic understanding of entities for CRS. In our KERL framework, entity textual descriptions are encoded via a PLM, while a KG helps reinforce the representation of these entities. We also employ positional encoding to effectively capture the temporal information of entities in a conversation. The enhanced entity representation is then used to develop a recommender component that fuses both entity and contextual representations for more informed recommendations, as well as a dialog component that generates informative entity-related information in the response text. A high-quality KG with aligned entity descriptions is constructed to facilitate this study, namely, the Wiki Movie Knowledge Graph (WikiMKG). The experimental results show that KERL achieves state-of-the-art results in both recommendation and response generation tasks. Our code is publicly available at the link: https://github.com/icedpanda/KERL.

Deep neural networks (DNN) has achieved great successes across multiple domains. In recent years, a number of approaches have emerged on automatically finding the optimal DNN configurations. A technique among these approaches which show great promise is Evolutionary Algorithms (EA), which are based on observations from natural, biological processes. However, since the EA needs to evaluate multiple DNN candidates, and if the training time for a DNN is large, then the required time would be very large. A potential solution is to use Surrogate Assisted Evolutionary Algorithm (SAEA), in which a surrogate model is used to predict performance of DNNs without training. It is noted that all popular surrogate models in the literature require a fixed-length input, while encodings of a DNN are usually variable-length, since a DNN structure is very complex and its depths, sizes, etc. cannot be known beforehand. In this paper, we propose a novel surrogate model for variable-length encoding to optimise deep learning architecture. An encoder-decoder model is used to convert the variable-length encoding into a fixed-length representation, which is used as inputs to the surrogate model to predict the DNN performance without training. The weights of the encoder-decoder model are found via training on the variable-length data, with the targets being the same as the inputs, while the surrogate model is trained on the encoder output in the encoder-decoder model. In this study, a Long Short-Term Memory (LSTM) model is used as the encoder and decoder. Our proposed variable-length encoding based surrogate model is tested on a well-known method which evolves optimal Convolutional Neural Networks (CNNs). The experimental results show that our proposed method has competitive performance while significantly reducing the time of optimisation process.

We proposed a novel ensemble selection method called VISTA for multiple layers ensemble systems (MLES). Our ensemble model consists of multiple layers of ensemble of classifiers (EoC) in which the EoC in each layer is trained on the data generated by a concatenation of the original training data and the predictions by classifiers of the previous layer. The predictions of the EoC in the final layer are aggregated to obtain the final prediction. To enhance the accuracy of the MLES, we used the Variable-Length Genetic Algorithm (VLGA) to search for the optimal configuration of EoC in each layer. Since the optimisation process is computationally intensive, we use Surrogate-Assisted Evolutionary Algorithms (SAEA) to reduce the training time. Most surrogate models developed in the literature require a fixed-length input, which limits their applications when the encoding is of variable length. In this paper, we proposed to use a Long Short-Term Memory (LSTM)-based surrogate model, in which the LSTM transforms the variable-length encoding to a fixed-size representation which will then be used by the surrogate model to predict the fitness values in VLGA. For the surrogate model, we adopted Radial Basis Function (RBF) for surrogation. We first conducted experiments in comparing two types of LSTM converters, and the results suggest that the proposed chunk-based LSTM converter provides better results compared to the normal LSTM converter. Our experiments on 15 datasets show that VISTA outperforms several benchmark algorithms.

Temporal Knowledge Graph Reasoning (TKGR) is the process of utilizing temporal information to capture complex relations within a Temporal Knowledge Graph (TKG) to infer new knowledge. Conventional methods in TKGR typically depend on deep learning algorithms or temporal logical rules. However, deep learning-based TKGRs often lack interpretability, whereas rule-based TKGRs struggle to effectively learn temporal rules that capture temporal patterns. Recently, Large Language Models (LLMs) have demonstrated extensive knowledge and remarkable proficiency in temporal reasoning. Consequently, the employment of LLMs for Temporal Knowledge Graph Reasoning (TKGR) has sparked increasing interest among researchers. Nonetheless, LLMs are known to function as black boxes, making it challenging to comprehend their reasoning process. Additionally, due to the resource-intensive nature of fine-tuning, promptly updating LLMs to integrate evolving knowledge within TKGs for reasoning is impractical. To address these challenges, in this paper, we propose a Large Language Models-guided Dynamic Adaptation (LLM-DA) method for reasoning on TKGs. Specifically, LLM-DA harnesses the capabilities of LLMs to analyze historical data and extract temporal logical rules. These rules unveil temporal patterns and facilitate interpretable reasoning. To account for the evolving nature of TKGs, a dynamic adaptation strategy is proposed to update the LLM-generated rules with the latest events. This ensures that the extracted rules always incorporate the most recent knowledge and better generalize to the predictions on future events. Experimental results show that without the need of fine-tuning, LLM-DA significantly improves the accuracy of reasoning over several common datasets, providing a robust framework for TKGR tasks.

• A multi-layer ensemble connects each classifier to multiple ones in prior layer. • Connections signify the use of previous-layer-classifiers’ outputs as new inputs. • A binary encoding scheme is proposed to encode the topology of proposed ensemble. • The optimal topology of proposed ensemble is found by using Differential Evolution. • Our ensemble performs better than benchmark algorithms on experimental datasets. Ensemble learning is a powerful machine learning strategy that combines multiple models e.g. classifiers to improve predictions beyond what any single model can achieve. Until recently, traditional ensemble methods typically use only one layer of models which limits the exploration of different aspects in the classifiers’ predictions. On the other hand, the rise of deep learning has introduced multi-layer architectures that can learn complex functions by transforming data into multiple levels of representation. This characteristic of deep learning suggests that multi-layer ensembles may potentially provide better performance compared to single-layer ensembles. However, a problem which might arise is that in the subsequent layers, not all the inputs to a classifier are desirable, leading to lower performance. In this paper, we introduce a novel multi-layer ensemble of classifiers named COME in which each classifier at a specific layer is connected to multiple classifiers in the previous layer. These connections signify the use of the previous-layer-classifiers’ outputs as inputs for training the current layer's classifier. Each classifier can be connected to different classifiers in the previous layer, which allows inputs in each layer to be optimally selected. We propose a binary encoding scheme to encode the topology of the proposed multi-layer ensemble with defined connections between layers. Differential Evolution, a popular evolutionary computation method, is used as the optimisation algorithm to search for the optimal set of connections. Experimental results on 30 datasets from the UCI Machine Learning Repository and OpenML demonstrate that our proposed ensemble outperforms many state-of-the-art ensemble learning algorithms.

Recent studies have shown that lip shape and movement can be used as an effective biometric feature for speaker authentication. By using random prompt text scheme, lip-based authentication system can also achieve good liveness detection performance in laboratory scenarios. However, due to the increasingly widespread mobile application, the authentication system may face additional practical difficulties such as complex background, limited user samples, etc., which will degrade the authentication performance derived by current methods. To confront the above problems, a new deep neural network, i.e. the Triple-feature Disentanglement Network for Visual Speaker Authentication (TDVSA-Net), is proposed in this paper to extract discriminative and disentangled lip features for visual speaker authentication in the random prompt text scenario. Three decoupled lip features, including the content feature inferring the speech content, the physiological lip feature describing the static lip shape and appearance and the behavioral lip feature depicting the unique pattern in lip movements during utterance, are extracted by TDVSA-Net and fed into corresponding modules to authenticate both the prompt text and the speaker’s identity. Experiment results have demonstrated that compared with several SOTA visual speaker authentication methods, the proposed TDVSA-Net can extract more discriminative and robust lip features which boost the content recognition and identity authentication performance against both human imposters and DeepFake attacks.

Fine-grained lip image segmentation plays a critical role in downstream tasks, such as automatic lipreading, as it enables the accurate identification of inner mouth components, such as teeth and tongue, which are essential for comprehending spoken utterances. However, achieving accurate and robust lip image segmentation in natural scenes is still challenging due to significant variations in lighting condition, head pose, and background. This article proposes a novel deep neural network-based method for fine-grained lip image segmentation that exploits fuzzy and graph theories to handle these variations. A fuzzy learning module is designed to deal with the uncertainties in color and edge information and enhance feature maps at various scales. The fuzzy graph reasoning module with fuzzy projection models the relationship among semantics components and achieves a global receptive field. In our experiments, a fine-grained lip region segmentation dataset, i.e., FLRSeg, is built for evaluation, and experimental results have shown that the proposed method can achieve superior segmentation performance (94.36% in pixel accuracy and 74.89% in mIoU) compared with several state-of-the-art (SOTA) lip image segmentation methods.

The fusion of hyperspectral and LiDAR data has been an active research topic. Existing fusion methods have ignored the high-dimensionality and redundancy challenges in hyperspectral images, despite that band selection methods have been intensively studied for hyperspectral image (HSI) processing. This paper addresses this significant gap by introducing a cross-attention mechanism from the transformer architecture for the selection of HSI bands guided by LiDAR data. LiDAR provides high-resolution vertical structural information, which can be useful in distinguishing different types of land cover that may have similar spectral signatures but different structural profiles. In our approach, the LiDAR data are used as the “query” to search and identify the “key” from the HSI to choose the most pertinent bands for LiDAR. This method ensures that the selected HSI bands drastically reduce redundancy and computational requirements while working optimally with the LiDAR data. Extensive experiments have been undertaken on three paired HSI and LiDAR data sets: Houston 2013, Trento and MUUFL. The results highlight the superiority of the cross-attention mechanism, underlining the enhanced classification accuracy of the identified HSI bands when fused with the LiDAR features. The results also show that the use of fewer bands combined with LiDAR surpasses the performance of state-of-the-art fusion models.

The motion cueing algorithm (MCA) enables lifelike motion in simulators resembling real driving. Regenerated motions must adhere to workspace constraints. Vehicle motion signals (linear acceleration, angular velocity) are generated in a simulated vehicle environment utilised in MCA for motion cues. These signals are categorised into levels (slow, medium, fast) based on frequency and amplitude. The commonly used MCA, the classical washout filter, is typically fine-tuned using worst-case (fast-driving) scenarios to meet the simulator’s requirements across various situations. However, this approach reduces the MCA’s effectiveness in handling slower driving scenarios, resulting in conservatism in platform workspace usage for slow and medium driving. Consequently, a noticeable motion sensation error arises between real vehicle drivers and motion simulator users. To rectify this issue, a novel neural network-based MCA is developed in this study. Three distinct classical washout filters are meticulously tuned to cater to slow, medium, and fast driving scenarios. These filters generate precise motion cues for simulator users at corresponding levels of driving scenarios. The neural network-based MCA is constructed using the synthesised signals from these classical washout filters. This proposed method is thoroughly validated through the utilisation of MATLAB software. In direct comparison with the standard classical washout filter, the proposed MCA significantly reduces the motion sensation error, enriches motion fidelity, and optimises the utilisation of the simulator’s workspace.

For cross-domain pattern classification, the supervised information (i.e., labeled patterns) in the source domain is often employed to help classify the unlabeled target domain patterns. In practice, multiple target domains are usually available. The unlabeled patterns (in different target domains) which have high-confidence predictions, can also provide some pseudo-supervised information for the downstream classification task. The performance in each target domain would be further improved if the pseudo-supervised information in different target domains can be effectively used. To this end, we propose an evidential multi-target domain adaptation (EMDA) method to take full advantage of the useful information in the single-source and multiple target domains. In EMDA, we first align distributions of the source and target domains by reducing maximum mean discrepancy (MMD) and covariance difference across domains. After that, we use the classifier learned by the labeled source domain data to classify query patterns in the target domains. The query patterns with high-confidence predictions are then selected to train a new classifier for yielding an extra piece of soft classification results of query patterns. The two pieces of soft classification results are then combined by evidence theory. In practice, their reliabilities/weights are usually diverse, and an equal treatment of them often yields the unreliable combination result. Thus, we propose to use the distribution discrepancy across domains to estimate their weighting factors, and discount them before fusing. The evidential combination of the two pieces of discounted soft classification results is employed to make the final class decision. The effectiveness of EMDA was verified by comparing with many advanced domain adaptation methods on several cross-domain pattern classification benchmark datasets.

Federated learning (FL) is vulnerable to poisoning attacks, where adversaries corrupt the global aggregation results and cause denial-of-service (DoS). Unlike recent model poisoning attacks that optimize the amplitude of malicious perturbations along certain prescribed directions to cause DoS, we propose a flexible model poisoning attack (FMPA) that can achieve versatile attack goals. We consider a practical threat scenario where no extra knowledge about the FL system (e.g., aggregation rules or updates on benign devices) is available to adversaries. FMPA exploits the global historical information to construct an estimator that predicts the next round of the global model as a benign reference. It then fine-tunes the reference model to obtain the desired poisoned model with low accuracy and small perturbations. Besides the goal of causing DoS, FMPA can be naturally extended to launch a fine-grained controllable attack, making it possible to precisely reduce the global accuracy. Armed with precise control, malicious FL service providers can gain advantages over their competitors without getting noticed, hence opening a new attack surface in FL other than DoS. Even for the purpose of DoS, experiments show that FMPA significantly decreases the global accuracy, outperforming six state-of-the-art attacks.

With the significant breakthrough of Deep Neural Networks in recent years, multi-layer architecture has influenced other sub-fields of machine learning including ensemble learning. In 2017, Zhou and Feng introduced a deep random forest called gcForest that involves several layers of Random Forest-based classifiers. Although gcForest has outperformed several benchmark algorithms on specific datasets in terms of classification accuracy and model complexity, its input features do not ensure better performance when going deeply through layer-by-layer architecture. We address this limitation by introducing a deep ensemble model with a novel feature generation module. Unlike gcForest where the original features are concatenated to the outputs of classifiers to generate the input features for the subsequent layer, we integrate weights on the classifiers’ outputs as augmented features to grow the deep model. The usage of weights in the feature generation process can adjust the input data of each layer, leading the better results for the deep model. We encode the weights using variable-length encoding and develop a variable-length Particle Swarm Optimization method to search for the optimal values of the weights by maximizing the classification accuracy on the validation data. Experiments on a number of UCI datasets confirm the benefit of the proposed method compared to some well-known benchmark algorithms.

Interdependent dimensions including categorical and continuous variables can be seen commonly as heterogeneous behavioral data in the real world. Mixed-type objects are more or less associated in terms of certain coupling relationships. The usual representation of such behavioral data is an information table with explicit behavior exterior dimensions (i.e. the original attributes to describe data heterogeneity), assuming the independence of dimensions and the independence of objects. However, both variables and objects are actually very often interdependent on one another either explicitly or implicitly in functional and semantic manners. Limited research has been done in analyzing such interactions among dimensions and those relationships among objects, leading to the learning results to be more local than global. This paper proposes the interdependence analysis to capture the functional multifarious relationships among attributes and among objects in heterogeneous data by addressing the coupling context and coupling weights in unsupervised learning. Such global couplings consider the interactions within discrete dimensions, within numerical attributes and across them, as well as the relationships within an individual object and between multiple objects, to form the attribute-based and object-based coupled data representation schemes based on feature conversion and neighborhood calculation. In addition, we interpret both the representation models via implicit behavior interior dimensions (i.e. the newly defined attributes to model data interdependence) to explain the intrinsic rationales for the superiority of our proposed methods. This work explicitly models the coupling of multiple attributes and the coupling of multiple objects for heterogeneous data sets, demonstrated by various data mining and machine learning applications, such as cluster structure analysis, data clustering evaluation, and data density comparison. Moreover, the sensitivity study is carried out to tune the neighborhood parameter and weight parameter, and the scalability analysis is explored to test the robustness of both models. Extensive experiments on a series of synthetic data sets and multiple UCI data sets show that our proposed framework can effectively capture the global couplings of both heterogeneous variables and mixed-type objects, and is superior to the traditional way as well as the state-of-the-art approaches, which is also verified by statistical analysis.

As an ownership verification technique for deep neural networks, the white-box neural network watermark is being challenged by the functionality equivalence attack. By leveraging the structural symmetry within a deep neural network and manipulating the parameters accordingly, an adversary can invalidate almost all white-box watermarks without affecting the network’s performance. This paper introduces the linear functionality equivalence attack, which can adapt to different network architectures without requiring knowledge of either the watermark or data. We also propose NeuronMap, a framework that can efficiently neutralize linear functionality equivalence attacks and can be easily combined with existing white-box watermarks to enhance their robustness. Experiments conducted on several deep neural networks and state-of-the-art white-box watermarking schemes have demonstrated not only the destructive power of linear functionality equivalence attacks but also the defense capability of NeuronMap. Our result shows that the threat of basic linear functionality equivalence attacks against deep neural network watermarks can be effectively solved using NeuronMap.

Electrocardiogram (ECG) signal analysis is widely used to diagnose various cardiac and non-cardiac diseases. Detecting abnormalities on ECG is critical for preventing the onset of life-threatening cardiac arrhythmias. This paper proposed a method based on deep convolutional neural network (DCNN) to detect abnormal heartbeats such as ventricular ectopic beats (VEB) and supraventricular ectopic beats (SVEB). The proposed model was trained and validated on two large-sample PhysioNet’s MIT-BIH datasets. A separate test result showed overall accuracy of 96% on distinguishing three types of heartbeats VEB, SVEB, and other heartbeats which are not ectopic beat (NOTEB).

Deep Neural Networks (DNNs) have created a breakthrough in medical image analysis in recent years. Because clinical applications of automated medical analysis are required to be reliable, robust and accurate, it is necessary to devise effective DNNs based models for medical applications. In this paper, we propose an ensemble framework of DNNs for the problem of medical image segmentation with a note that combining multiple models can obtain better results compared to each constituent one. We introduce an effective combining strategy for individual segmentation models based on swarm intelligence, which is a family of optimization algorithms inspired by biological processes. The problem of expensive computational time of the optimizer during the objective function evaluation is relieved by using a surrogate-based method. We train a surrogate on the objective function information of some populations and then use it to predict the objective values of each candidate in the subsequent populations. Experiments run on a number of public datasets indicate that our framework achieves competitive results within reasonable computation time.

Self-driving vehicles, also known as Autonomous Vehicles (AVs), are steadily becoming very popular due to their huge benefits. They can improve safety, convenience and transport interconnectivity as well as reduce congestion, pollution and emissions. The generation of the comfort motion signal for AVs passenger via the calculation of accurate motion cues with lower motion discomforts is important to promote the adoption of Avs in society. Model predictive control (MPC) is currently used in AVs for tracking the motion signal with good accuracy. However, the higher efficiency of MPC is directly related to the right setting of the weights. In addition, the tracking of time-varying longitudinal velocity is not possible without using linear time-varying (LTV) MPC. In this study, an LTV MPC system is designed and developed as a highly efficient motion tracking mechanism for AVs to reduce the motion tracking error and motion discomfort. In addition, biogeography-based optimisation (BBO) is employed to determine the optimal weights of the LTV MPC controller, which further reduces the motion tracking error and increases the motion comfort for users. The empirical study demonstrates that a BBO-tuned LTV MPC controller decreases the mean square error of motion tracking by 4.79% as compared with that of a manually-tuned version. Moreover, the mean square errors of the lateral deviation and relative yaw decrease by 91.22% and 19.14% as compared with those from a manually-tuned LTV MPC counterpart, respectively.

Motion signal should be generated via the AV control system targeting the maximum motion comfort for the users. Nonlinear model predictive control (MPC) is recently used in AVs to achieve this critical task. However, nonlinear MPC has lots of hyperparameters, including weights and MPC horizons, that should be tuned systematically to reach the system’s high efficiency. The energy usage and motion comfort have a direct relationship. The generation of high-fidelity motion cues for AV users leads to higher energy usage. Hence, there is a need for the use of a multi-objective optimisation technique to tune the weights wisely to satisfy the appropriate energy usage and motion comfort for the AV users. In this study, multi-objective NSGA-II is employed, for the first time, to tune the weights of a nonlinear MPC-based controller in AVs. The proposed method is designed and developed using MATLAB/SIMULINK software. The simulation results show minimum energy usage by generation of smooth motion signals, delivering maximum comfort to AV users.

Abstract Genetic algorithms are a practical approach for finding near‐optimal solutions for nondeterministic polynomial‐hard problems. In this work we exploit the parallel processing capability of graphics processing units and Nvidia's CUDA programming platform to accelerate the island model genetic algorithm by modifying the evolutionary operations to fit the hardware architecture and have successfully achieved significant computational speedups.

Parallel and distributed computing has been the basis to many emerging areas, such as smart networks, cloud computing, big data analysis, and blockchain technology.Without the development of parallel and distributed computing technologies and various types of systems, it is not possible to meet the requirement on efficiency, accuracy, scalability, and reliability for various critical applications that support our modern economy and society.While parallel and distributed computing played a vital role in modern science, engineering, biology, medicine, pharmacy, astronomy, geology, and archaeology, its application has also been extended to business, finance, economics, management, government, and defense, covering all aspects of our modern society and life.Furthermore, parallel and distributed computing has emerged in recent advances of many hotspot research directions including artificial intelligence, machine learning, Internet of Things, bioinformatics, digital medicine, cybersecurity, and social computing, resulting in numerous ground-breading discoveries that are changing our society and life.

Many image processing and pattern recognition problems can be formulated as binary quadratic programming (BQP) problems. However, solving a large BQP problem with a good quality solution and low computational time is still a challenging unsolved problem. Current methodologies either adopt an independent random search in a semi-definite space or perform search in a relaxed biconvex space. However, the independent search has great computation cost as many different trials are needed to get a good solution. The biconvex search only searches the solution in a local convex ball, which can be a local optimal solution. In this paper, we propose a BQP solver that alternatingly applies a deterministic search and a stochastic neighborhood search. The deterministic search iteratively improves the solution quality until it satisfies the KKT optimality conditions. The stochastic search performs bootstrapping sampling to the objective function constructed from the potential solution to find a stochastic neighborhood vector. These two steps are repeated until the obtained solution is better than many of its stochastic neighborhood vectors. We compare the proposed solver with several state-of-the-art methods for a range of image processing and pattern recognition problems. Experimental results showed that the proposed solver not only outperformed them in solution quality but also with the lowest computational complexity.

The pervasive deployment of the Internet of Things (IoT) has significantly facilitated manufacturing and living. The diversity and continual updates of IoT systems make their security a crucial challenge, among which the detection of malicious network traffic turns out to be the most common yet destructive threat. Despite the efforts on feature engineering and classification backend designing, established intrusion detection systems sometimes lack robustness and are inflexible against the shift of the traffic distribution. To deal with these disadvantages, we design a fuzzy system for the online defense of IoT. Our framework incorporates a full Bayesian possibilistic clustering module for feature processing and an ensemble module motivated by reinforcement learning and adaptive boosting that dynamically fits the streaming data. The proposed clustering module overcomes the issue of determining the number of clusters and can dynamically identify new patterns. The classifier backend combines a collection of fuzzy decision trees that provide readable decision boundaries. The ensembled classifiers can accommodate the drift of data distribution to optimize the long-time performance. Our proposal is tested on settings including one dataset collected from real IoT systems and is compared to numerous competitors. Experimental results verified the advantage of our system regarding accuracy and stability.

In this paper, we propose a novel method for boundary detection in close-range hyperspectral images. This method can effectively predict the boundaries of objects of similar colour but different materials. To effectively extract the material information in the image, the spatial distribution of the spectral responses of different materials or endmembers is first estimated by hyperspectral unmixing. The resulting abundance map represents the fraction of each endmember spectra at each pixel. The abundance map is used as a supportive feature such that the spectral signature and the abundance vector for each pixel are fused to form a new spectral feature vector. Then different spectral similarity measures are adopted to construct a sparse spectral-spatial affinity matrix that characterizes the similarity between the spectral feature vectors of neighbouring pixels within a local neighborhood. After that, a spectral clustering method is adopted to produce eigenimages. Finally, the boundary map is constructed from the most informative eigenimages. We created a new HSI dataset and use it to compare the proposed method with four alternative methods, one for hyperspectral image and three for RGB image. The results exhibit that our method outperforms the alternatives and can cope with several scenarios that methods based on colour images cannot handle.

Classification of malicious software, especially in a very large dataset, is a challenging task for machine intelligence. Malware can have highly diversified features, each of which has highly heterogeneous distributions. These factors increase the difficulties for traditional data analytic approaches to deal with them. Although deep learning based methods have reported good classification performance, the deep models usually lack interpretability and are fragile under adversarial attacks. To solve these problems, fuzzy systems have become a competitive candidate in malware analysis. In this article, a new fuzzy-based approach is proposed for malware classification. We focused on portable executable files in the Windows platform and analyzed the distributions of static features and content-oriented features. Fuzzification was used to reduce the ubiquitous impact of noise and outliers in a very large dataset. Finally, a novel boosted classifier consisted of fuzzy decision trees and support vector machine is proposed to perform the malware classification. By using fuzzy decision trees, the inner structure of the classifier can be readily interpreted as discriminative rules, whereas the novel boosting strategy provides state-of-the-art classification performance. Extensive experimental results showed that our method significantly outperformed several state-of-the-art classifiers.

Recent research has demonstrated that lip-based speaker authentication systems can not only achieve good authentication performance but also guarantee liveness. However, with modern DeepFake technology, attackers can produce the talking video of a user without leaving any visually noticeable fake traces. This can seriously compromise traditional face-based or lip-based authentication systems. To defend against sophisticated DeepFake attacks, a new visual speaker authentication scheme based on the deep convolutional neural network (DCNN) is proposed in this paper. The proposed network is composed of two functional parts, namely, the Fundamental Feature Extraction network (FFE-Net) and the Representative lip feature extraction and Classification network (RC-Net). The FFE-Net provides the fundamental information for speaker authentication. As the static lip shape and lip appearance is vulnerable to DeepFake attacks, the dynamic lip movement is emphasized in the FFE-Net. The RC-Net extracts high-level lip features that discriminate against human imposters while capturing the client's talking style. A multi-task learning scheme is designed, and the proposed network is trained end-to-end. Experiments on the GRID and MOBIO datasets have demonstrated that the proposed approach is able to achieve an accurate authentication result against human imposters and is much more robust against DeepFake attacks compared to three state-of-the-art visual speaker authentication algorithms. It is also worth noting that the proposed approach does not require any prior knowledge of the DeepFake spoofing method and thus can be applied to defend against different kinds of DeepFake attacks.

The control of epidemics taking place in complex networks has been an increasingly active topic in public health management. In this article, we propose an efficient networked epidemic control system, where a modified susceptible-exposed-infected-vigilant (SEIV) model is first built to simulate epidemic spreading. Then, different from existing continuous resource models which abstractly map resources to parameters of epidemic models, a concrete resource description model is built to simulate real-world goods/services and their allocation. Based on the two models, a cost-constraint subset selection problem in epidemic control is identified. To solve the problem, a swarm-based stochastic optimization policy is proposed, where each particle in the swarm can determine its own solutions according to the guidance of its superior peers and historical searching experience of the whole swarm, without extra problem-relative information. Theoretical proof about system equilibrium is provided, which is consistent with experimental observations. The competitive performance of the proposed optimizer is validated by theoretical analysis and comparison experiments. Finally, an application case is provided to illustrate the practicability.

Deep Neural Networks have achieved many successes when applying to visual, text, and speech information in various domains. The crucial reasons behind these successes are the multi-layer architecture and the in-model feature transformation of deep learning models. These design principles have inspired other sub-fields of machine learning including ensemble learning. In recent years, there are some deep homogenous ensemble models introduced with a large number of classifiers in each layer. These models, thus, require a costly computational classification. Moreover, the existing deep ensemble models use all classifiers including unnecessary ones which can reduce the predictive accuracy of the ensemble. In this study, we propose a multi-layer ensemble learning framework called MUlti-Layer heterogeneous Ensemble System (MULES) to solve the classification problem. The proposed system works with a small number of heterogeneous classifiers to obtain ensemble diversity, therefore being efficiency in resource usage. We also propose an Evolutionary Algorithm-based selection method to select the subset of suitable classifiers and features at each layer to enhance the predictive performance of MULES. The selection method uses NSGA-II algorithm to optimize two objectives concerning classification accuracy and ensemble diversity. Experiments on 33 datasets confirm that MULES is better than a number of well-known benchmark algorithms.

Recent research shows that the lip feature can achieve reliable authentication performance with a good liveness detection ability. However, with the development of sophisticated human face generation methods by the deepfake technology, the talking videos can be forged with high quality and the static lip information is not reliable in such case. Meeting with such challenge, in this paper, we propose a new deep neural network structure to extract robust lip features against human and Computer-Generated (CG) imposters. Two novel network units, i.e. the feature-level Difference block (Diffblock) and the pixel-level Dynamic Response block (DRblock), are proposed to reduce the influence of the static lip information and to represent the dynamic talking habit information. Experiments on the GRID dataset have demonstrated that the proposed network can extract discriminative and robust lip features and outperform two state-of-the-art visual speaker authentication approaches in both human imposter and CG imposter scenarios.

Research has shown that the human lip and its movements are a rich source of information related to speech content and speaker's identity. Lip image segmentation, as a fundamental step in many lip-reading and visual speaker authentication systems, is of vital importance. Because of variations in lip color, lighting conditions and especially the complex appearance of an open mouth, accurate lip region segmentation is still a challenging task. To address this problem, this article proposes a new fuzzy deep neural network having an architecture that integrates fuzzy units and traditional convolutional units. The convolutional units are used to extract discriminative features at different scales to provide comprehensive information for pixel-level lip segmentation. The fuzzy logic modules are employed to handle various kinds of uncertainties and to provide a more robust segmentation result. An end-to-end training scheme is then used to learn the optimal parameters for both the fuzzy and the convolutional units. A dataset containing more than 48 000 images of various speakers, under different lighting conditions, was used to evaluate lip segmentation performance. According to the experimental results, the proposed method achieves state-of-the-art performance when compared with other algorithms.

Historical financial data are frequently used in technical analysis to identify patterns that can be exploited to achieve trading profits. Although technical analysis using a variety of technical indicators has proven to be useful for the prediction of price trends, it is difficult to use them to formulate trading rules that could be used in an automatic trading system due to the vague nature of the rules. Moreover, it is challenging to determine a specified combination of technical indicators that can be used to detect good trading points and trading rules since different stock may be affected by different set of factors. In this paper, we propose a novel trading point forecasting framework that incorporates a bicluster mining technique to discover significant trading patterns, a method to establish the fuzzy rule base, and a fuzzy inference system optimized for trading point prediction. The proposed method (called BM-FM) was tested on several historical stock datasets and the average performance was compared with the conventional buy-and-hold strategy and five previously reported intelligent trading systems. Experimental results demonstrated the superior performance of the proposed trading system.

In ensemble systems, the predictions of base classifiers are aggregated by a combining algorithm (meta-classifier) to achieve better classification accuracy than using a single classifier. Experiments show that the performance of ensembles significantly depends on the choice of meta-classifier. Normally, the classifier selection method applied to an ensemble usually removes all the predictions of a classifier if this classifier is not selected in the final ensemble. Here we present an idea to only remove a subset of each classifier's prediction thereby introducing a simultaneous meta-data and meta-classifier selection method for ensemble systems. Our approach uses Cross Validation on the training set to generate meta-data as the predictions of base classifiers. We then use Ant Colony Optimization to search for the optimal subset of meta-data and meta-classifier for the data. By considering each column of meta-data, we construct the configuration including a subset of these columns and a meta-classifier. Specifically, the columns are selected according to their corresponding pheromones, and the meta-classifier is chosen at random. The classification accuracy of each configuration is computed based on Cross Validation on meta-data. Experiments on UCI datasets show the advantage of proposed method compared to several classifier and feature selection methods for ensemble systems.

Lip image segmentation, as the first step in many lip-related tasks (e.g. automatic lipreading), is of vital significance for the subsequent procedures. Nowadays, with the increasing computational power of the mobile devices, mobile applications become more and more popular. In this paper, a new approach is proposed, which is able to segment the lip region in natural scenes and is of acceptable computational complexity to be implemented in mobile devices. Two networks including a complex teacher network and a compact student network with the same structure are employed. With the proposed remedy loss and the alternative knowledge distillation scheme, the student network can learn useful knowledge from the teacher network effectively and efficiently, and even rectify some of its segmentation errors. A dataset containing 49 people captured under natural scenes by various cellphone cameras is adopted for evaluation and the experiment results have demonstrated that the proposed student network even outperforms the teacher network with much less computational cost.

When a deterministic search approach is too costly, such as for non-deterministic polynomial-hard problems, finding near-optimal solutions with approximation algorithms, such as the genetic algorithm, is the only practical approach to reduce the execution time. In this paper, we exploit the capability of graphics processing units (GPU), specifically Nvidia's CUDA platform, to accelerate the genetic algorithm by modifying the evolutionary operations to fit the hardware architecture. This has allowed us to achieve significant computational speedups compared to the non-GPU counterparts.

With the advancement of storage and processing technology, an enormous amount of data is collected on a daily basis in many applications. Nowadays, advanced data analytics have been used to mine the collected data for useful information and make predictions, contributing to the competitive advantages of companies. The increasing data volume, however, has posed many problems to classical batch learning systems, such as the need to retrain the model completely with the newly arrived samples or the impracticality of storing and accessing a large volume of data. This has prompted interest on incremental learning that operates on data streams. In this study, we develop an incremental online multi-label classification (OMLC) method based on a weighted clustering model. The model is made to adapt to the change of data via the decay mechanism in which each sample's weight dwindles away over time. The clustering model therefore always focuses more on newly arrived samples. In the classification process, only clusters whose weights are greater than a threshold (called mature clusters) are employed to assign labels for the samples. In our method, not only is the clustering model incrementally maintained with the revealed ground truth labels of the arrived samples, the number of predicted labels in a sample are also adjusted based on the Hoeffding inequality and the label cardinality. The experimental results show that our method is competitive compared to several well-known benchmark algorithms on six performance measures in both the stationary and the concept drift settings.

In this paper, we introduced a new approach of combining multiple classifiers in a heterogeneous ensemble system. Instead of using numerical membership values when combining, we constructed interval membership values for each class prediction from the meta-data of observation by using the concept of information granule. In the proposed method, the uncertainty (diversity) of the predictions produced by the base classifiers is quantified by the interval-based information granules. The decision model is then generated by considering both bound and length of the intervals. Extensive experimentation using the UCI datasets has demonstrated the superior performance of our algorithm over other algorithms including six fixed combining methods, one trainable combining method, AdaBoost, bagging, and random subspace.

Image classification is considered to be one of the critical tasks in hyperspectral remote sensing image processing. Recently, a convolutional neural network (CNN) has established itself as a powerful model in classification by demonstrating excellent performances. The use of a graphical model such as a conditional random field (CRF) contributes further in capturing contextual information and thus improving the classification performance. In this paper, we propose a method to classify hyperspectral images by considering both spectral and spatial information via a combined framework consisting of CNN and CRF. We use multiple spectral band groups to learn deep features using CNN, and then formulate deep CRF with CNN-based unary and pairwise potential functions to effectively extract the semantic correlations between patches consisting of 3-D data cubes. Furthermore, we introduce a deep deconvolution network that improves the final classification performance. We also introduced a new data set and experimented our proposed method on it along with several widely adopted benchmark data sets to evaluate the effectiveness of our method. By comparing our results with those from several state-of-the-art models, we show the promising potential of our method.

This paper proposes a novel approach of weakly supervised video object segmentation, which only needs one pixel to guide the segmentation. We use two deep neural networks to get the instance-level semantic segmentation masks and optical flow maps of each frame. An object probability map to the first frame in video is generated by combining the semantic masks, the optical flow maps and the guiding pixel. The object probability map propagates forward and backward and becomes more accurate to each frame. Finally, an energy minimization problem on a function that consists of unary term of object probability and pairwise terms of label smoothness potentials is solved to get the pixel-wise object segmentation mask of each frame. We evaluate our method on a benchmark dataset, and the experimental results show that the proposed approach achieves impressive performance in comparison with state-of-the-art methods.

Motivation: Protein-peptide interactions are one of the most important biological interactions and play crucial role in many diseases including cancer. Therefore, knowledge of these interactions provides invaluable insights into all cellular processes, functional mechanisms, and drug discovery. Protein-peptide interactions can be analyzed by studying the structures of protein-peptide complexes. However, only a small portion has known complex structures and experimental determination of protein-peptide interaction is costly and inefficient. Thus, predicting peptide-binding sites computationally will be useful to improve efficiency and cost effectiveness of experimental studies. Here, we established a machine learning method called SPRINT-Str (Structure-based prediction of protein-Peptide Residue-level Interaction) to use structural information for predicting protein-peptide binding residues. These predicted binding residues are then employed to infer the peptide-binding site by a clustering algorithm. Results: SPRINT-Str achieves robust and consistent results for prediction of protein-peptide binding regions in terms of residues and sites. Matthews' Correlation Coefficient (MCC) for 10-fold cross validation and independent test set are 0.27 and 0.293, respectively, as well as 0.775 and 0.782, respectively for area under the curve. The prediction outperforms other state-of-the-art methods, including our previously developed sequence-based method. A further spatial neighbor clustering of predicted binding residues leads to prediction of binding sites at 20-116% higher coverage than the next best method at all precision levels in the test set. The application of SPRINT-Str to protein binding with DNA, RNA and carbohydrate confirms the method's capability of separating peptide-binding sites from other functional sites. More importantly, similar performance in prediction of binding residues and sites is obtained when experimentally determined structures are replaced by unbound structures or quality model structures built from homologs, indicating its wide applicability. Availability and implementation: http://sparks-lab.org/server/SPRINT-Str. Contact: [email protected]. Supplementary information: Supplementary data are available at Bioinformatics online.

Spectral-spatial feature extraction is an important task in hyperspectral image processing. In this paper we propose a novel method to extract distinctive invariant features from hyperspectral images for registration of hyperspectral images with different spectral conditions. Spectral condition means images are captured with different incident lights, viewing angles, or using different hyperspectral cameras. In addition, spectral condition includes images of objects with the same shape but different materials. This method, which is named spectral-spatial scale invariant feature transform (SS-SIFT), explores both spectral and spatial dimensions simultaneously to extract spectral and geometric transformation invariant features. Similar to the classic SIFT algorithm, SS-SIFT consists of keypoint detection and descriptor construction steps. Keypoints are extracted from spectral-spatial scale space and are detected from extrema after 3D difference of Gaussian is applied to the data cube. Two descriptors are proposed for each keypoint by exploring the distribution of spectral-spatial gradient magnitude in its local 3D neighborhood. The effectiveness of the SS-SIFT approach is validated on images collected in different light conditions, different geometric projections, and using two hyperspectral cameras with different spectral wavelength ranges and resolutions. The experimental results show that our method generates robust invariant features for spectral-spatial image matching.

This paper proposes a method that uses both spectral and spatial information to segment remote sensing hyperspectral images. After a hyperspectral image is over-segmented into superpixels, a deep Convolutional Neural Network (CNN) is used to perform superpixel-level labelling. To further delineate objects from a hyperspectral scene, this paper attempts to combine the properties of CNN and Conditional Random Field (CRF). A mean-field approximation algorithm for CRF inference is used and formulated with Gaussian pairwise potentials as Recurrent Neural Network. This combined network is then plugged into the CNN which leads to a deep network that has robust characteristics of both CNN and CRF. Preliminary results suggest the usefulness of this framework to a promising extent.

Safe and sustainable road systems is a common goal in all countries. Applications to assist with road asset management and crash minimization are sought universally. Road accident is a special case of trauma that constitutes a major cause of disability, untimely death and loss of loved ones as well as family bread winners. Therefore, methods to reduce accident severity are of great interest to traffic agencies and the public at large. To analysis the traffic accident factors effectively we need a noise-free dataset. Road accident fatality rate depends on many factors and it is a very challenging task to investigate the dependencies between the attributes because of the many environmental and road accident factors. Noisy data in the dataset could obscure the discovery of important factors and mislead conclusions. In this paper, we present a novel approach called Neural network and Co-appearance based analysis for Noisy Attributes values Identification (NCNAI). NCNAI separates noisy records from clean records and also identifies the incorrect attributes values. We evaluate our algorithm using two publicly available traffic accident databases of United States, one is the largest open federal database (explore.data.gov) in United States and another one is based on the National Incident Based Reporting System (NIBRS) of city and county of Denver (data.opencolorado.org). We compare our technique with three existing methods using several well-known evaluation criteria, i.e. value based Error Recall (vER), value based Error Precision (vEP), record based Error Recall (rER), record based Error Precision (rEP), and record Removal Ratio (rRR). Our results indicate that the proposed method performs significantly better than the three existing algorithms.

The amount of available content of different types of services is so large nowadays that one cannot realistically have a real time overview of the content. Recommendation engines were developed to solve the problem of information overload, and save time and effort when looking for appealing content. In this paper, we present an enhanced Naïve Bayes model for rating prediction of a program based on content description information. As our prediction model has to deal with categorical data, a probabilistic Bayesian network is used. The model uses a set of features to predict user rating based on past observation. We also simulated recommendation from a program offer. The recommendation system presented in this paper is flexible and robust enough to handle a sparse data set with very few records of feature description. Experiments were performed on a Yahoo movie data set and they indicated the promising performance of our approach over an existing technique.

MOTIVATION: Cellular interactions of kinesin-1, an adenosine triphosphate (ATP)-driven motor protein capable of undergoing multiple steps on a microtubule (MT), affect its mechanical processivity, the number of steps taken per encounter with MT. Even though the processivity of kinesin has been widely studied, a detailed study of the factors that affect the stepping of the motor along MT is still lacking. RESULTS: We model the cellular interactions of kinesin as a probabilistic timed automaton and use the model to simulate the mechanical processivity of the motor. Theoretical analysis suggests: (i) backward stepping tends to be powered by ATP hydrolysis, rather than ATP synthesis, (ii) backward stepping powered by ATP synthesis is more likely to happen with limiting ATP concentration ([ATP]) at high loads and (iii) with increasing load the frequency of backward stepping powered by ATP hydrolysis at high [ATP] is greater than that powered by ATP synthesis at limiting [ATP]. Together, the higher frequency of backward stepping powered by ATP hydrolysis than by ATP synthesis is found to be a reason for the more dramatic falling of kinesin processivity with rising load at high [ATP] compared with that at low [ATP]. Simulation results further show that the processivity of kinesin can be determined by the number of ATP hydrolysis and synthesis kinetic cycles taken by the motor before becoming inactive. It is also found that the duration of a backward stepping cycle at high loads is more likely to be less than that of a forward stepping cycle. CONTACT: [email protected] or [email protected].

In this paper we introduce a novel approach for classifier and feature selection in a multi-classifier system using Genetic Algorithm (GA). Specifically, we propose a 2-part structure for each chromosome in which the first part is encoding for classifier and the second part is encoding for feature. Our structure is simple in the implementation of the crossover as well as the mutation stage of GA. We also study 8 different fitness functions for our GA based algorithm to explore the optimal fitness functions for our model. Experiments are conducted on both 14 UCI Machine Learning Repository and CLEF2009 medical image database to demonstrate the benefit of our model on reducing classification error rate.

In high-throughput applications, accurate segmentation of biomedical images can be considered as an important step for recognizing cells that have the phenotype of interest. In this paper, while conventional fuzzy clustering is not able to implement the local and global spatial information, a novel spatial fuzzy clustering cell image segmentation algorithm is proposed. The segmentation procedure is divided into two stages: the first stage involves processing the local and global spatial information of the given cell image and a final segmentation stage which is based on the idea of conventional fuzzy clustering. Our idea can be considered as a sequential integration of region based methods and fuzzy clustering for cell image segmentation. Experimental results show that the proposed model yields significantly better performance in comparison with several existing methods.

MOTIVATION: Kinesin is a smart motor protein that steps processively forward and backward along microtubules (MTs). The mechanical kinetics of kinesin affecting its stepping behavior is not fully understood. Here, we propose a mathematical model to study the mechanical kinetics of forward and backward stepping of kinesin motor based on the four-state discrete stochastic model of the motor. RESULTS: Results show that the probabilities of forward and backward stepping can be modeled using the mean probabilities of forward and backward kinetic transitions, respectively. We show that the backward stepping of kinesin motor starts when the probability of adenosine diphosphate (ADP) binding to the motor is much higher than that of adenosine triphosphate (ATP) binding. Furthermore, our results indicate that the backward stepping is related to both ATP hydrolysis and synthesis with rate limiting factor being ATP synthesis. Low rate of ATP synthesis under high backward loads above 10 pN is also suggested as a reason for the detachment of kinesin motor from MT in the kinetic state MTċKinesinċADPċPi. AVAILABILITY AND IMPLEMENTATION: The code for this work is written in Visual C# and is available by request from the authors.

This paper presents a new modified fuzzy c-means (FCM) clustering algorithm that exploits bilateral symmetry information in image data. With the assumption of pixels that are located symmetrically tend to have similar intensity values; we compute the degree of symmetry for each pixel with respect to a global symmetry axis of the image. This information is integrated into the objective function of the standard FCM algorithm. Experimental results show the effectiveness of the approach. The method was further improved using neighbourhood information, and was compared with conventional fuzzy c-means algorithms.

The analysis of gene expression data obtained from microarray experiments is important for discovering the biological process of genes. Biclustering algorithms have been proven to be able to group the genes with similar expression patterns under a number of experimental conditions. In this paper, we propose a new biclustering algorithm based on evolutionary learning. By converting the biclustering problem into a common clustering problem, the algorithm can be applied in a search space constructed by the conditions. To further reduce the size of the search space, we randomly separate the full conditions into a number of condition subsets (subspaces), each of which has a smaller number of conditions. The algorithm is applied to each subspace and is able to discover bicluster seeds within a limited computing time. Finally, an expanding and merging procedure is employed to combine the bicluster seeds into larger biclusters according to a homogeneity criterion. We test the performance of the proposed algorithm using synthetic and real microarray data sets. Compared with several previously developed biclustering algorithms, our algorithm demonstrates a significant improvement in discovering additive biclusters.

Microarray gene expression data generally suffers from missing value problem due to a variety of experimental reasons. Since the missing data points can adversely affect downstream analysis, many algorithms have been proposed to impute missing values. In this survey, we provide a comprehensive review of existing missing value imputation algorithms, focusing on their underlying algorithmic techniques and how they utilize local or global information from within the data, or their use of domain knowledge during imputation. In addition, we describe how the imputation results can be validated and the different ways to assess the performance of different imputation algorithms, as well as a discussion on some possible future research directions. It is hoped that this review will give the readers a good understanding of the current development in this field and inspire them to come up with the next generation of imputation algorithms.

BACKGROUND: Gene clustering for annotating gene functions is one of the fundamental issues in bioinformatics. The best clustering solution is often regularized by multiple constraints such as gene expressions, Gene Ontology (GO) annotations and gene network structures. How to integrate multiple pieces of constraints for an optimal clustering solution still remains an unsolved problem. RESULTS: We propose a novel multiconstrained gene clustering (MGC) method within the generalized projection onto convex sets (POCS) framework used widely in image reconstruction. Each constraint is formulated as a corresponding set. The generalized projector iteratively projects the clustering solution onto these sets in order to find a consistent solution included in the intersection set that satisfies all constraints. Compared with previous MGC methods, POCS can integrate multiple constraints from different nature without distorting the original constraints. To evaluate the clustering solution, we also propose a new performance measure referred to as Gene Log Likelihood (GLL) that considers genes having more than one function and hence in more than one cluster. Comparative experimental results show that our POCS-based gene clustering method outperforms current state-of-the-art MGC methods. CONCLUSIONS: The POCS-based MGC method can successfully combine multiple constraints from different nature for gene clustering. Also, the proposed GLL is an effective performance measure for the soft clustering solutions.

Detecting blood vessels in retinal images with the presence of bright and dark lesions is a challenging unsolved problem. In this paper, a novel multiconcavity modeling approach is proposed to handle both healthy and unhealthy retinas simultaneously. The differentiable concavity measure is proposed to handle bright lesions in a perceptive space. The line-shape concavity measure is proposed to remove dark lesions which have an intensity structure different from the line-shaped vessels in a retina. The locally normalized concavity measure is designed to deal with unevenly distributed noise due to the spherical intensity variation in a retinal image. These concavity measures are combined together according to their statistical distributions to detect vessels in general retinal images. Very encouraging experimental results demonstrate that the proposed method consistently yields the best performance over existing state-of-the-art methods on the abnormal retinas and its accuracy outperforms the human observer, which has not been achieved by any of the state-of-the-art benchmark methods. Most importantly, unlike existing methods, the proposed method shows very attractive performances not only on healthy retinas but also on a mixture of healthy and pathological retinas.

Many cellular processes exhibit cyclic behaviors. Hence, one important task in gene expression data analysis is to detect subset of genes that exhibit periodicity in their gene expression time series profiles. Unfortunately, gene expression time series profiles are usually of very short length, with very few periods, unevenly sampled, and are highly contaminated with noise. This makes detection of periodic profiles a very challenging problem. In this paper, we present several effective computational techniques developed recently in our research group for the reliable detection of short periodic gene expression time series profiles.

BACKGROUND: The DNA microarray technology allows the measurement of expression levels of thousands of genes under tens/hundreds of different conditions. In microarray data, genes with similar functions usually co-express under certain conditions only 1. Thus, biclustering which clusters genes and conditions simultaneously is preferred over the traditional clustering technique in discovering these coherent genes. Various biclustering algorithms have been developed using different bicluster formulations. Unfortunately, many useful formulations result in NP-complete problems. In this article, we investigate an efficient method for identifying a popular type of biclusters called additive model. Furthermore, parallel coordinate (PC) plots are used for bicluster visualization and analysis. RESULTS: We develop a novel and efficient biclustering algorithm which can be regarded as a greedy version of an existing algorithm known as pCluster algorithm. By relaxing the constraint in homogeneity, the proposed algorithm has polynomial-time complexity in the worst case instead of exponential-time complexity as in the pCluster algorithm. Experiments on artificial datasets verify that our algorithm can identify both additive-related and multiplicative-related biclusters in the presence of overlap and noise. Biologically significant biclusters have been validated on the yeast cell-cycle expression dataset using Gene Ontology annotations. Comparative study shows that the proposed approach outperforms several existing biclustering algorithms. We also provide an interactive exploratory tool based on PC plot visualization for determining the parameters of our biclustering algorithm. CONCLUSION: We have proposed a novel biclustering algorithm which works with PC plots for an interactive exploratory analysis of gene expression data. Experiments show that the biclustering algorithm is efficient and is capable of detecting co-regulated genes. The interactive analysis enables an optimum parameter determination in the biclustering algorithm so as to achieve the best result. In future, we will modify the proposed algorithm for other bicluster models such as the coherent evolution model.

BACKGROUND: In DNA microarray experiments, discovering groups of genes that share similar transcriptional characteristics is instrumental in functional annotation, tissue classification and motif identification. However, in many situations a subset of genes only exhibits consistent pattern over a subset of conditions. Conventional clustering algorithms that deal with the entire row or column in an expression matrix would therefore fail to detect these useful patterns in the data. Recently, biclustering has been proposed to detect a subset of genes exhibiting consistent pattern over a subset of conditions. However, most existing biclustering algorithms are based on searching for sub-matrices within a data matrix by optimizing certain heuristically defined merit functions. Moreover, most of these algorithms can only detect a restricted set of bicluster patterns. RESULTS: In this paper, we present a novel geometric perspective for the biclustering problem. The biclustering process is interpreted as the detection of linear geometries in a high dimensional data space. Such a new perspective views biclusters with different patterns as hyperplanes in a high dimensional space, and allows us to handle different types of linear patterns simultaneously by matching a specific set of linear geometries. This geometric viewpoint also inspires us to propose a generic bicluster pattern, i.e. the linear coherent model that unifies the seemingly incompatible additive and multiplicative bicluster models. As a particular realization of our framework, we have implemented a Hough transform-based hyperplane detection algorithm. The experimental results on human lymphoma gene expression dataset show that our algorithm can find biologically significant subsets of genes. CONCLUSION: We have proposed a novel geometric interpretation of the biclustering problem. We have shown that many common types of bicluster are just different spatial arrangements of hyperplanes in a high dimensional data space. An implementation of the geometric framework using the Fast Hough transform for hyperplane detection can be used to discover biologically significant subsets of genes under subsets of conditions for microarray data analysis.

It is well known that visual cues of lip movement contain important speech relevant information. This paper presents an automatic lipreading system for small vocabulary speech recognition tasks. Using the lip segmentation and modeling techniques we developed earlier, we obtain a visual feature vector composed of outer and inner mouth features from the lip image sequence for recognition. A spline representation is employed to transform the discrete-time sampled features from the video frames into the continuous domain. The spline coefficients in the same word class are constrained to have similar expression and are estimated from the training data by the EM algorithm. For the multiple-speaker/speaker-independent recognition task, an adaptive multimodel approach is proposed to handle the variations caused by various talking styles. After building the appropriate word models from the spline coefficients, a maximum likelihood classification approach is taken for the recognition. Lip image sequences of English digits from 0 to 9 have been collected for the recognition test. Two widely used classification methods, HMM and RDA, have been adopted for comparison and the results demonstrate that the proposed algorithm deliver the best performance among these methods.

For the image classification task, the color histogram is widely used as an important color feature indicating the content of the image. However, the high-resolution color histograms are usually of high dimension and contain much redundant information which does not relate to the image content, while the low-resolution histograms cannot provide adequate discriminative information for image classification. In this paper, a new color feature representation is proposed which not only takes the correlation among neighbouring components of the conventional color histogram into account but removes the redundant information as well. A high-resolution, uniform quantized color histogram is first obtained from the image. Then the redundant bins are removed and some neighbouring bins are combined together to generate a new feature component to maximize the discriminative ability. The mutual information is adopted to evaluate the discriminative power of a specific feature set and an iterative algorithm is performed to derive the histogram quantization and their corresponding feature generation. To illustrate the effectiveness of the proposed feature representation, an application of detecting adult images, i.e., image classification between erotic and benign images, is carried out. Two widely used classification techniques, SVM and Adaboost, are employed as the classifier. Experimental results show the superior performance of our color representation compared with the conventional color histogram in image classification.

BACKGROUND: Periodogram analysis of time-series is widespread in biology. A new challenge for analyzing the microarray time series data is to identify genes that are periodically expressed. Such challenge occurs due to the fact that the observed time series usually exhibit non-idealities, such as noise, short length, and unevenly sampled time points. Most methods used in the literature operate on evenly sampled time series and are not suitable for unevenly sampled time series. RESULTS: For evenly sampled data, methods based on the classical Fourier periodogram are often used to detect periodically expressed gene. Recently, the Lomb-Scargle algorithm has been applied to unevenly sampled gene expression data for spectral estimation. However, since the Lomb-Scargle method assumes that there is a single stationary sinusoid wave with infinite support, it introduces spurious periodic components in the periodogram for data with a finite length. In this paper, we propose a new spectral estimation algorithm for unevenly sampled gene expression data. The new method is based on signal reconstruction in a shift-invariant signal space, where a direct spectral estimation procedure is developed using the B-spline basis. Experiments on simulated noisy gene expression profiles show that our algorithm is superior to the Lomb-Scargle algorithm and the classical Fourier periodogram based method in detecting periodically expressed genes. We have applied our algorithm to the Plasmodium falciparum and Yeast gene expression data and the results show that the algorithm is able to detect biologically meaningful periodically expressed genes. CONCLUSION: We have proposed an effective method for identifying periodic genes in unevenly sampled space of microarray time series gene expression data. The method can also be used as an effective tool for gene expression time series interpolation or resampling.

It is well known that low bit rate block-based discrete cosine transform coded image exhibits visually annoying coding artifacts. In this paper, we proposed a projection onto convex sets (PCOS)-based deblocking algorithm using a novel region smoothness constraint set for graphic images containing objects with smooth regions. The smoothness constraint set is obtained by an explicit modeling of smooth regions in the image using a spatially adaptive thin-plate spline. In contrast to most deblocking algorithms which enforce smoothness just around the 8/spl times/8 block boundaries, our algorithm enforces smoothness in regions which could possibly span several blocks. We showed that convergence of our algorithm could be reached within one iteration. The performance of the proposed algorithm is evaluated visually and quantitatively in term of peak signal-to-noise ratios and the mean squared difference of slope metric, which measures the impact of the blocking effects, for several graphic images. The results show that our algorithm can effectively suppress blockiness in smooth regions while still preserving the sharpness of object edges.

An image segmentation algorithm based on adaptive fuzzy c-means (FCM) clustering is presented in this paper. In the conventional FCM clustering algorithm, cluster assignment is based solely on the distribution of pixel attributes in the feature space, and does not take into consideration the spatial distribution of pixels in an image. By introducing a novel dissimilarity index in the modified FCM objective function, the new adaptive fuzzy clustering algorithm is capable of utilizing local contextual information to impose local spatial continuity, thus exploiting the high inter-pixel correlation inherent in most real-world images. The incorporation of local spatial continuity allows the suppression of noise and helps to resolve classification ambiguity. To account for smooth intensity variation within each homogenous region in an image, a multiplicative field is introduced to each of the fixed FCM cluster prototype. The multiplicative field effectively makes the fixed cluster prototype adaptive to slow smooth within-cluster intensity variation, and allows homogenous regions with slow smooth intensity variation to be segmented as a whole. Experimental results with synthetic and real color images have shown the effectiveness of the proposed algorithm.

In a constrained total least squares algorithm (CTLS), the selection of a minimal algebraic set of linearly independent random variables to express the noise matrix /spl Delta/C is an important task. A fast algorithm is provided using the possibly dependent random variables set. We showed that it can be viewed as a combination of the CTLS method of V.Z. Mesarovic et al. (see IEEE Trans. Image Process., vol.4, p.1096-107, 1995) and the RLS method when the noise is Gaussian. Our experimental study indicates that our algorithm has better visual and objective quality, while having a much lower computation cost. Moreover, our algorithm can also handle a more general noise model.

Lip image analysis has attracted much interest in recent years because some important speech information is contained in the shape and movement of the lip. To extract such information from the images, accurate and robust lip region segmentation is of vital importance. However, most of the current lip segmentation methods fail to provide accurate results if the person has a beard. We propose a "one object, multiple background" clustering method to solve the problem. Since the non-lip region becomes inhomogeneous in the presence of a beard, multiple background clusters can produce better fitting to a rather complex background region than a single cluster. Spatial information in terms of the physical distance towards the lip center is incorporated to enhance the differentiation between the lip and background region. Experimental results demonstrate that our algorithm provides accurate lip segmentation results for images with beards.

Human facial animation is an interesting and difficult problem in computer graphics. In this paper, a novel B-spline (NURBS) muscle system is proposed to simulate a 3D facial expression and talking animation. The system gets the lip shape parameters from the video, which captures a real person's lip movement, to control the proper muscles to form different phonemes. The muscles are constructed by the non-uniform rational B-spline curves, which are based on anatomical knowledge. By using different number of control points on the muscles, more detailed facial expression and mouth shapes can be simulated. We demonstrate the flexibility of our model by simulating different emotions and lip-sync to a video with a talking head using the automatically extracted lip parameters.

MOTIVATION: Microarray time-series data provides us a possible means for identification of transcriptional regulation relationships among genes. Currently, the most commonly used method in determining whether or not two genes have a potential regulatory relationship is to measure their expressional similarity using Pearson's correlation coefficient. Although this traditional correlation method has been successfully applied to find functionally correlated genes, it does have many limitations. In the hope of overcoming such circumstances and getting more insights into the transcriptional regulatory issue, we propose an autoregressive (AR)-based technique for detection of potential regulated gene pairs from time-series microarray measurements. RESULTS: We use the well-known AR modeling technique to characterize temporal gene expression data from the Spellman's alpha-synchronized yeast cell-cycle experiment. In this method, time-series expression profiles are decomposed into spectral components and correlations between profiles are then computed in a component-wise sense. We show how these component-wise correlations reveal possible regulatory relationships. Our technique is applied on known transcriptional regulations and is able to identify many of those missed by the traditional correlation method.

It is well known that at low-bit-rate block discrete cosine transform compressed image exhibits visually annoying blocking and ringing artifacts. In this paper, we propose a noniterative, wavelet-based deblocking algorithm to reduce both types of artifacts. The algorithm exploits the fact that block discontinuities are constrained by the dc quantization interval of the quantization table, as well as the behavior of wavelet modulus maxima evolution across wavelet scales to derive appropriate threshold maps at different wavelet scales. Since ringing artifacts occur near strong edges, which can be located either along block boundaries or within blocks, suppression of block discontinuities does not always reduce ringing artifacts. By exploiting the behavior of ringing artifacts in the wavelet domain, we propose a simple yet effective method for the suppression of such artifacts. The proposed algorithm can suppress both block discontinuities and ringing artifacts effectively while preserving true edges and textural information. Simulation results and extensive comparative study with both iterative and noniterative methods reported in the literature have shown the effectiveness of our algorithm.

Cluster analysis of gene expression data from a cDNA microarray is useful for identifying biologically relevant groups of genes. However, finding the natural clusters in the data and estimating the correct number of clusters are still two largely unsolved problems. In this paper, we propose a new clustering framework that is able to address both these problems. By using the one-prototype-take-one-cluster (OPTOC) competitive learning paradigm, the proposed algorithm can find natural clusters in the input data, and the clustering solution is not sensitive to initialization. In order to estimate the number of distinct clusters in the data, we propose a cluster splitting and merging strategy. We have applied the new algorithm to simulated gene expression data for which the correct distribution of genes over clusters is known a priori. The results show that the proposed algorithm can find natural clusters and give the correct number of clusters. The algorithm has also been tested on real gene expression changes during yeast cell cycle, for which the fundamental patterns of gene expression and assignment of genes to clusters are well understood from numerous previous studies. Comparative studies with several clustering algorithms illustrate the effectiveness of our method.

A fuzzy c-means based adaptive clustering algorithm is proposed for the fuzzy segmentation of 3D MR brain images, which are typically corrupted by noise and intensity non-uniformity (INU) artifact. The proposed algorithm enforces the spatial continuity constraint to account for the spatial correlations between image voxels, resulting in the suppression of noise and classification ambiguity. The INU artifact is compensated for by the introduction of a pseudo-3D bias field, which is modeled as a stack of smooth B-spline surfaces with continuity enforced across slices. The efficacy of the proposed algorithm is demonstrated experimentally using both simulated and real MR images.

In this paper, we describe the application of a novel spatial fuzzy clustering algorithm to the lip segmentation problem. The proposed spatial fuzzy clustering algorithm is able to take into account both the distributions of data in feature space and the spatial interactions between neighboring pixels during clustering. By appropriate pre- and postprocessing utilizing the color and shape properties of the lip region, successful segmentation of most lip images is possible. Comparative study with some existing lip segmentation algorithms such as the hue filtering algorithm and the fuzzy entropy histogram thresholding algorithm has demonstrated the superior performance of our method.

An adaptive spatial fuzzy c-means clustering algorithm is presented in this paper for the segmentation of three-dimensional (3-D) magnetic resonance (MR) images. The input images may be corrupted by noise and intensity nonuniformity (INU) artifact. The proposed algorithm takes into account the spatial continuity constraints by using a dissimilarity index that allows spatial interactions between image voxels. The local spatial continuity constraint reduces the noise effect and the classification ambiguity. The INU artifact is formulated as a multiplicative bias field affecting the true MR imaging signal. By modeling the log bias field as a stack of smoothing B-spline surfaces, with continuity enforced across slices, the computation of the 3-D bias field reduces to that of finding the B-spline coefficients, which can be obtained using a computationally efficient two-stage algorithm. The efficacy of the proposed algorithm is demonstrated by extensive segmentation experiments using both simulated and real MR images and by comparison with other published algorithms.

A novel protection onto convex sets (POCS) method is presented for the suppression of blocking and ringing artifacts in a compressed image that contains homogeneous regions. A new family of convex smoothness constraint sets is introduced, using the uniformity property of image regions. This set of constraints allows different degrees of smoothing in different regions of the image, while preserving the image edges. The regions are segmented using the fuzzy c-means algorithm, which allows ambiguous pixels to be left unclassified. Experimental results on JPEG compressed images demonstrate that the proposed algorithm yields visually superior images compared to several of the previously reported POCS deblocking algorithms for the class of images considered.

An adaptive, spatially constrained fuzzy clustering algorithm for image segmentation is presented. By using a novel dissimilarity index in the cost function, our fuzzy clustering algorithm is capable of utilising local contextual information in a 3/spl times/3 neighborhood to impose local spatial continuity, thus exploiting the high inter-pixel correlation inherent in most realworld images. This has the effects of smoothing out random noise and resolving classification ambiguities. By introducing a multiplicative bias field to the fixed cluster prototypes, the cluster prototypes effectively become adaptive to nonstationarity in the image intensity. This allows non-planar regions or objects to be segmented meaningfully. Experimental results have shown the effectiveness of the proposed fuzzy clustering algorithm.

This paper presents a new optimization procedure for extracting the point-based lip contour using the active shape model (ASM). A 14-point ASM lip model is used to describe the lip contour. With the aid of fuzzy clustering analysis, a probability map of the color lip image is obtained and a region-based cost function is established. The new optimization procedure operates on the spatial domain (actual contour points) and all the points are pulled towards their desirable locations in each iteration. Hence, the lip contour evolution becomes better controlled and consequently fast convergence is achieved. The new procedure can also achieve real-time performance on lip contour extraction and tracking from lip image sequence.

The lifting scheme for the real field wavelet transform has provided a new insight into its practical implementation. This paper shows that a similar scheme can be developed for the binary field wavelet transform. In particular, by using the Euclidean algorithm, the binary filters can be decomposed into a finite sequence of simple lifting steps over the binary field. This provides an alternative method for the implementations of the binary field wavelet transform for image processing applications. It is found that the new implementation can reduce the number of arithmetic operations involved in the transform and allow an efficient in-place implementation structure.

The use of visual information from lip movements can improve the accuracy and robustness of a speech recognition system. Accurate extraction of visual features associated with the lips is thus very important. In this paper, we present a deformable model-based technique for lip contour extraction from color lip images. The geometric lip model we propose is able to fit different lip shapes. Our method starts by generating a probability map of the lip image using spatial fuzzy clustering. Then the optimum set of model parameters that partitions a lip probability map into lip region and non-lip region is found, such that the joint probability of the two regions is maximised. Experimental results indicate that accurate and robust lip contour extraction is possible using this approach.

A clustering-based lip segmentation algorithm is described here. The fuzzy clustering algorithm presented here is able to take into account the local smoothness property of image data. The objective functional of our algorithm utilizes a new distance metric that takes into account the influence of the neighboring pixels on the centre pixel in a 3 by 3 window. Computational steps involved in the segmentation of color lip image and segmentation result for a set of color lip images are given.