Concurrency bugs occur due to the uncertainty of thread scheduling within concurrent programs. Most of the existing concurrency bug fixing approaches fix concurrency bugs by serializing the execution of all threads involved in concurrency bugs. However, they face the threats of introducing new deadlocks and lead to high runtime overhead while fixing the concurrency bugs. On the other hand, most of the approaches are biased toward fixing only one type of concurrency bug, e.g. data races, deadlocks, or atomicity violations. In this paper, we propose TemCon, a template-based concurrency bug fixing approach that constructs fine-grained semantic change graphs on patch files based on Abstract Syntax Trees (ASTs) to mine accurate fixing templates. First, it constructs fine-grained semantic change graphs on patch files and splits the graphs into three subgraphs (i.e. attribute graphs, operation graphs and text graphs). Then, the fixing templates are extracted by clustering the same graph pairs. Finally, the buggy programs are matched with the fixing templates to generate the patches for fixing concurrency bugs. We constructed a new concurrency bug dataset with 1830 concurrency bug fixing patches and compared our approach with the state-of-the-arts in our experiments. The experimental results show that our approach can correctly fix 887 concurrency bugs without introducing new deadlocks, which is 684 and 93 more than Grail and PFix, respectively. TemCon can achieve a fixing accuracy of 48.47%, which is 5.08–37.38% higher than that of the state-of-the-art approaches.

Multimodal automatic depression detection (ADD) has garnered significant research interest due to its potential for providing fast, objective, and reliable assessments. Despite advancements in the field, several challenges persist. First, many existing methods employ fixed-size feature modeling, which fails to adapt to temporal variations in depressive emotional expression, resulting in the loss of critical depressive behavior information. Second, while attention-based detection methods effectively extract features, they incur substantial computational overhead. Finally, multimodal ensemble processes often encounter modality inertia and modality forgetting issues, which compromise model stability and performance. We propose an innovative multimodal ADD network called the fully convolutional adaptive ensemble network (FC-AEN) to address these challenges. This approach comprises three key modules. The first module, deformable patch embedding (DPE), dynamically adjusts the time range of feature segments during training to ensure comprehensive capture of depressive behaviors. The second module, fully convolutional spatiotemporal detector (FCSTD), enhances computational efficiency to manage larger data scales effectively. The third module, adaptive dynamic ensemble (ADE), addresses the multimodal unbalanced ensemble problem through alternating training and gradient adjustment. We conducted experiments on four datasets (AVEC2013, AVEC2014, AVEC2019, and CMDep) and achieved new state-of-the-art (SOTA) performance across three benchmarks. Specifically, on the AVEC2019 dataset, we achieved a mean absolute error (MAE) of 4.91, representing a relative reduction of 14.9%; on the AVEC2013 dataset, we achieved an MAE of 5.11 and a relative reduction of 5.0%; and on the CMDep dataset, we achieved an MAE of 6.04 and a relative reduction of 3.4%.

Multimodal recommendation is a crucial technology on social media platforms. It is widely applied in scenarios such as product recommendation and advertising delivery. However, existing multimodal recommendation approaches often overlook invariant semantic features that persist across modalities, leading to decreased robustness and generalization. To address this limitation, we proposeGILMRec, a novelgraphinvariantlearning-basedmultimodal social mediarecommendation framework. The GILMRec introduces an invariant feature learning strategy to extract invariant features separately from visual and textual modalities and employs an attention-based fusion mechanism to integrate them into a unified embedding. Specifically, we construct modality-specific similarity graphs and apply top-$t$neighbor aggregation, enhancing the consistency of invariant features while effectively suppressing modality-specific noise. Extensive experiments on three Amazon benchmark datasets and a large-scale dataset [baby, sports, clothing, and compact discs (CDs)] demonstrate that GILMRec consistently outperforms twelve state-of-the-art baselines. The results confirm the efficiency of invariant features in capturing robust multimodal representations and improving recommendation performance, particularly in sparse data scenarios.

Unsupervised Domain Adaptation (UDA) has emerged as a pivotal technique for enhancing machine learning models' performance in unlabeled target domain with domain shifts. This technique is fundamentally achieved by aligning the domain distributions of source and target domains within a latent feature space, thereby enhancing model robustness across heterogeneous data distributions. However, the inherent discrepancy between source and target domain distributions poses significant challenges in identifying the optimal latent space. Furthermore, projecting both domains into suboptimal latent spaces may induce substantial semantic information loss, particularly compromising discriminative feature representations critical for final tasks. In this article, our systematic analysis reveals that natural language representations inherently possess stronger semantic abstraction capabilities than visual features in natural images. As a result, natural language tends to have smaller domain shifts. Motivated by this discovery, we proposed a novel model that systematically transforms visual patterns into structured linguistic representations. This cross-modal translation mechanism leverages the invariant semantic properties of natural language to mitigate domain shifts while preserving task-critical semantic hierarchies. Our model leverages the inherent abstraction capacity of linguistic structures to enhance cross-domain generalization, effectively bridging the visual-semantic gap in unsupervised adaptation scenarios. Our model comprises three core components: 1) text classification branch translating images to text for prediction; 2) image adaptation branch supplementing visual details; and 3) ensemble mechanism reconciling text abstraction with visual granularity through mismatch detection. Extensive experiments on three benchmark datasets validate the effectiveness of our model, achieving state-of-the-art performance.

Alzheimer's disease (AD) is the most prevalent form of dementia, and its early diagnosis is essential for slowing disease progression. Recent studies on multimodal neuroimaging fusion using MRI and PET have achieved promising results by integrating multi-scale complementary features. However, most existing approaches primarily emphasize cross-modal complementarity while overlooking the diagnostic importance of modality-specific features. In addition, the inherent distributional differences between modalities often lead to biased and noisy representations, degrading classification performance. To address these challenges, we propose a Collaborative Attention and Consistent-Guided Fusion framework for MRI and PET based AD diagnosis. The proposed model introduces a learnable parameter representation (LPR) block to compensate for missing modality information, followed by a shared encoder and modality-independent encoders to preserve both shared and specific representations. Furthermore, a consistency-guided mechanism is employed to explicitly align the latent distributions across modalities. Experimental results on the ADNI dataset demonstrate that our method achieves superior diagnostic performance compared with existing fusion strategies.

Wearable intelligence technologies have advanced rapidly in recent years, achieving considerable results in Parkinson's disease (PD) severity classification. Recently, end-to-end deep time-series models have become more popular than traditional machine-learning classifiers, as they do not rely on handcrafted feature extraction. However, for PD severity recognition, most existing studies overlook the time-frequency complementarity and axis correlations present in multivariate signals. To address these gaps, we propose a multivariate time-series fusion framework that converts raw accelerometer and gyroscope signals into continuous wavelet transform (CWT) and recurrence plot (RP) images to capture complementary frequency and recurrence domain information, and jointly integrates these with the original time-domain sequences in end-to-end deep classifiers, as well as handcrafted features extracted at the patient level. We conduct preliminary experiments on wrist IMU data from 95 PD patients using five-fold, patient-level cross-validation focusing on walking activity. Our proposed fusion framework achieves a precision of 72.6 %, outperforming the latest end-toend multivariate time-series baseline model. Moreover, the fusion results are substantially higher than those obtained using CWT images, RP images, or handcrafted features alone, highlighting the effectiveness of jointly leveraging heterogeneous representations. Overall, this study highlights the lack of cross-axis and time-frequency complementary modeling in current end-to-end time-series methods and provides new insights into interpretable deep learning for PD severity classification, thereby offering additional priors and support for advancing disease recognition technologies.

Alzheimer's Disease (AD) is the most prevalent neurodegenerative disorder in aging populations, posing a significant and escalating burden on global healthcare systems. While Multi-Tusk Learning (MTL) has emerged as a powerful computational paradigm for modeling longitudinal AD data, existing frameworks do not account for the time-varying nature of feature correlations. To address this limitation, we propose a novel MTL framework, named Feature Similarity Laplacian graph Multi-Task Learning (MTL-FSL). Our framework introduces a novel Feature Similarity Laplacian (FSL) penalty that explicitly models the time-varying relationships between features. By simultaneously considering temporal smoothness among tasks and the dynamic correlations among features, our model enhances both predictive accuracy and biological interpretability. To solve the non-smooth optimization problem arising from our proposed penalty terms, we adopt the Alternating Direction Method of Multipliers (ADMM) algorithm. Experiments conducted on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset demonstrate that our proposed MTLFSL framework achieves state-of-the-art performance, outperforming various baseline methods. The implementation source can be found at https://github.com/huatxxx/MTL-FSL.

Severity assessment of Parkinson's disease (PD) using wearable sensors offers an effective, objective basis for clinical management. However, general-purpose time series models often lack pathological specificity in feature extraction, making it difficult to capture subtle signals highly correlated with PD. Furthermore, the temporal sparsity of PD symptoms causes key diagnostic features to be easily “diluted” by traditional aggregation methods, further complicating assessment. To address these issues, we propose the Multi-scale Frequency-Aware Adversarial Multi-Instance Network (MFAM). This model enhances feature specificity through a frequency decomposition module guided by medical prior knowledge. Furthermore, by introducing an attention-based multi-instance learning (MIL) framework, the model can adaptively focus on the most diagnostically valuable sparse segments. We comprehensively validated MFAM on both the public PADS dataset for PD versus differential diagnosis (DD) binary classification and a private dataset for four-class severity assessment. Experimental results demonstrate that MFAM outperforms general-purpose time series models in handling complex clinical time series with specificity, providing a promising solution for automated assessment of PD severity.

Automatic Depression Detection (ADD) garners widespread attention due to its convenience and objectivity. While existing research makes significant progress, challenges remain. First, most current ADD methods struggle to balance computational overhead and prediction accuracy. Second, these methods primarily rely on facial images and audio, which are susceptible to external factors, affecting the model’s generalizability. In this study, we propose the Spatiotemporal Ensemble Mamba (STE-Mamba), a framework based entirely on the Mamba architecture for detecting and ensembling spatiotemporal information. This approach reduces computational overhead while effectively capturing long-range spatiotemporal information. Additionally, we extract remote Photoplethysmography (rPPG) and emotion trends (ET) from facial videos, providing two more generalizable physiological modalities for ADD. Experimental results indicate that the inclusion of the ET modality, which only adds two dimensions, improves diagnostic accuracy by approximately 6%. We conduct extensive experiments on five datasets (AVEC2013, AVEC2014, AVEC2017, AVEC2019, CMDep), and the results demonstrate that STE-Mamba is highly competitive in terms of both effectiveness and generalizability. The self-built CMDep can be requested via the following link.

In the domain of target tracking, leveraging auxiliary modalities such as depth, thermal, and event data to enhance tracking robustness has garnered substantial attention. Due to the scarcity of visual multimodal datasets, state-of-the-art approaches primarily rely on parameter-efficient fine-tuning to adapt models. However, existing studies often neglect the adaptation of fine-tuning to specific datasets, frequently focusing only on non-RGB modalities or employing them as prompts. Additionally, current methods overly emphasize modal balance while disregarding modality adaptation to environmental conditions. To address these limitations, we propose a unified visual multimodal detection framework named IMTrack. This model incorporates LoRA (Low-Rank Adaptation) and Adapter for fine-tuning RGB and auxiliary modalities, respectively, and employs confidence-based cross-attention to strengthen feature interaction and adaptability. Furthermore, we introduce a complementary masking multi-scene optimization strategy to enhance robustness in complex environments. Extensive experiments on five datasets validate the efficacy of our model. The results demonstrate that IMTrack surpasses existing methods in many indicators, achieving state-of-the-art performance. Notably, the robustness of our model to auxiliary modes is improved by more than 2%. Our source code is available at: https://github.com/cxy-lzk/IMTrack.

Parkinson’s disease (PD) is a progressive neurodegenerative disorder that significantly diminishes patients’ quality of life. Early and accurate diagnosis is critical for reducing both individual and societal burdens. Although current diagnostic methods can effectively differentiate between PD patients and healthy individuals, they tend to ignore the diversity of PD symptoms and the differences with other similar diseases, such as essential tremor or multiple system atrophy, leading to a higher risk of misdiagnosis. Additionally, existing supervised learning methods rely on subjective labeling by physicians, which is both time-consuming and subjective. To overcome these limitations, we collect multi-sensor activity data from 102 participants in free-living environments, and propose a novel self-supervised learning framework that redefines PD diagnosis as an anomaly detection problem. Additionally, we utilize two large public PD datasets as external cohorts to verify their validity. Extensive experiments demonstrate that our framework not only learns more discriminative features but also significantly enhances the model’s generalization ability, providing a promising solution to reduce misdiagnosis in PD.

The prediction of fertilization rates is a critical area of research in the agricultural field and is essential for ensuring global food security. With the ongoing expansion of the global population and the escalating repercussions of climate change, precise crop fertilization rate predictions have become paramount. This is because accurate predictions can optimize resource allocation and improve agricultural productivity. Moreover, they can provide scientific support for policy-making and agricultural input management, thereby promoting sustainable agricultural development. Despite its importance, the complexity of agricultural systems, which is influenced by multiple factors including climate, geography, soil conditions, and management practices, poses significant challenges to prediction accuracy. In this paper, we propose a deep learning framework based on Graph Neural Networks (GNNs) that effectively incorporates geographical knowledge and multi-dimensional feature information. By modeling spatial relationships through graph structures (nodes and edges), our framework enhances fertilization rate prediction accuracy. We validate the model using two datasets of different scales. The results demonstrate excellent predictive performance across all datasets and strong scalability, highlighting its potential for agricultural fertilization rate prediction.

Pest recognition is of great significance for achieving sustainable development in agriculture. Nevertheless, due to the wide variety of pest species, subtle interspecies differences, and significant intraspecies variations, existing artificial intelligence and Internet of Things (IoT) technologies can only recognize a small number of known pests effectively. In this article, we propose a zero-shot learning pest recognition framework based on ensemble hierarchical attribute prompting, termed EHAPZero. EHAPZero can identify pest images collected by IoT devices, and then transmit the recognition results to the IoT platform for terminal display. Specifically, the image recognition function is implemented by an attribute generation module (AGM), a hierarchical prompting module (HPM), and a semantic-visual interaction module (SVIM). AGM utilizes large language models to construct a knowledge graph of pests. It employs both node importance evaluation algorithms and manual methods to perform dual filtering on attribute nodes within the graph. Inspired by human knowledge reasoning, HPM dynamically predicts different hierarchical attributes of input images within the Transformer intermediate blocks. These predicted attributes are subsequently injected into the intermediate layer features of the Transformer as prompts. To achieve semantic disambiguation and knowledge transfer, SVIM employs a visual-guided semantic representation method and a semantic-guided visual representation method to strengthen cross-domain interaction between semantics and vision. Finally, the final prediction score is derived through ensemble of prediction results across different levels. Extensive experiments show that EHAPZero achieves the new state-of-the-art results on the real-word pest recognition benchmark. The codes are available at:https://github.com/jinqiwen/EHAPZero.

Automatic Depression Detection (ADD) advances rapidly, though challenges persist. First, medical studies show distinct emotional changes between individuals with depression and those without, yet few studies effectively leverage emotions for ADD. Many existing models use a single processing module across various emotions, but the data characteristics differ under different emotions. A single module may not be able to grasp these data characteristics simultaneously, leading to the loss of key depressive clues. Second, fixed scales and perspectives are commonly applied in feature modeling, though depressive states and physiological signals change dynamically. Fixed modeling approaches may truncate or misconnect depressive signals. This study proposes a Multimodal Large Model-driven Ensemble of Expert Networks (MLM-EOE) for ADD, consisting of three components: (1) Multimodal Large Model Sentiment Annotation (MLM-SA) to annotate emotions in raw video data; (2) Ensemble of Experts (EOE) to capture data features under various emotional states; and (3) Modeling Inter and Intra Multiscale Patches (MIIMP) to apply multiscale, multi-perspective modeling to expert-processed data. Extensive experiments on four datasets (AVEC2013, AVEC2014, AVEC2019, and CMDep) validate the effectiveness of MLM-EOE. The self-constructed CMDep dataset is available upon request via the provided link. The code is publicly available at https://github.com/ZulongLin/MLM-EOE.

Depression, driven by growing societal pressures, significantly disrupts individuals’ physical and mental health. Automatic Depression Recognition (ADR) via facial videos has gained attention to enhance diagnostic accuracy and efficiency. However, extant methods often segment videos, losing long-term behavioral cues and introducing noise, while also exhibiting performance drops across diverse cultural and racial datasets. This study proposes a multimodal ADR approach encompassing three key components: (1) Long-term Depression Behavior Module (LDBM) employing a Transformer to capture extended depression cues, (2) Noisy Information Elimination (NIE) strategy leveraging LDBM attention scores to reduce noise and boost diagnostic precision, and (3) Multimodal Spatio-temporal Routing Feature Ensemble (MSRE) that fuses texture, Facial Action Primitives (FAPs), and Remote Photoplethysmography (rPPG) data for improved cross-dataset generalizability. Experiments on AVEC 2013, AVEC 2014, and a newly constructed CMDep dataset of 123 clinically diagnosed participants validate our method, achieving MAE/RMSE scores of 5.38/6.74, 5.09/6.83, and 5.59/8.03, respectively. The CMDep dataset includes facial expression and voice signals, with labels derived from BDI-II scores. Additionally, our method has been integrated into a user-friendly mobile application, providing a tool for real-time self-assessment of depression. This integration broadens the scope of depression detection, making it accessible to diverse populations worldwide.

Research indicates significant differences in the voice and facial expressions of individuals with depression compared to healthy individuals. Consequently, many studies have begun using audio-visual data for automatic depression detection (ADD). Despite significant progress, numerous challenges remain. During feature extraction, many studies fail to capture crucial spatiotemporal information for depression diagnosis, leading to decreased feature quality. In the modality fusion stage, many methods assume that the semantic information of different modalities is temporally aligned, which is not the case. To address the first challenge, this study proposes a linear spatiotemporal detector (LSTD) to model spatiotemporal information through detection and ensemble, enhancing feature quality. To tackle the second challenge, a cross-modal temporal aligner (CMTA) is proposed, using cross-modal attention for alignment before modality fusion to promote thorough fusion. LSTD and CMTA form the proposed spatiotemporal information modeling and modal alignment (SIMMA) framework. Extensive experiments on five depression datasets (AVEC2013, AVEC2014, AVEC2017, AVEC2019, and CMDep) demonstrate that our method outperforms previous studies in effectiveness.

Semi-supervised Object Detection (SSOD) is a method that uses a small amount of labeled data and a large amount of unlabeled data to improve the performance of object detection. However, existing SSOD methods face the challenges of scale imbalance and class inconsistency, resulting in large differences in detection results across different scales and classes. To overcome these challenges, we propose a Scale-Rebalanced Global Proposal Contrast Consistency (SGPC) approach, which has the following three advantages: 1) we design a Scale-Rebalanced Input (SRI) structure, which adjusts the distribution of objects of different scales by resampling the input images at low magnification, thereby enhancing the ability of small object detection; 2) we design a Global Proposal Contrast Consistency Loss (GPCC), which can enhance the intra-class compactness and inter-class diversity of Region of Interest (RoI) features, thereby reducing the class inconsistency in pseudo-labels; and 3) we adopt a loss blending optimization strategy, which optimizes the localization accuracy of pseudo-labels by combining supervised loss and unsupervised loss. We conduct extensive experiments on multiple datasets, and the results show that SGPC significantly outperforms the latest other methods on the SSOD task. On the PASCAL VOC dataset, SGPC achieves 55.90 mAP, on the MS-COCO dataset, SGPC exceeds the supervised methods by more than 10 mAP at different scales, and we also verify the significant improvement and robustness of SGPC on the small object detection datasets VisDrone-2019 and EDD.

The advancement of large models has initiated a transformation in the field of time-series forecasting. Both the repurposing of existing large models and the development of large models tailored for time-series analysis have exhibited impressive performance. In industrial applications, challenges, such as limited data availability and constrained computational resources, render the first approach viable. However, it is important to note that this approach is still in its infancy and lacks both a thorough technical analysis and a unified effective framework. Meanwhile, as large models become a mainstream artificial intelligence paradigm, it is urgent to discuss typical industrial scenarios, such as how automated systems can transition from intelligent to collaborative operation and maintenance. In light of this premise, this article endeavors to advance a generalized technical framework for large model-driven time-series forecasting, under which existing methods can be subsumed. Then, within this overarching technical paradigm, the technical advancements facilitated by diverse methods will be systematically elucidated and analyzed, along with a comparative evaluation conducted across seven benchmark datasets. Concluding this analysis, the implementation pathway for the industrial automation system is delineated that integrates operator action commands to forecast post-action trends to assess action correctness in advance. Finally, the challenges and future directions of large model-based time-series forecasting are outlined.

In environments where intelligent video surveillance systems (IVSSs) are deployed, particularly in review room, the detection of electronic devices constitutes a crucial task. Nevertheless, this task presents significant challenges attributed to the high rates of false positives and false negatives in electronic device detection (EDD), compounded by the low resolution of objects when viewed from multiple angles.To address these challenges, we propose a deep learning-based cascaded detection framework. Specifically, we design a mixed region sampling (MRS) method to enhance the foreground perception with background information and image details. We design a nonstrided downsampling method (ASDP) to map the attention spatial features to depth and improve the detection of low-resolution objects with fine-grained features. We enhance the model’s robustness to different viewing angles by feature perturbation during training. Moreover, we use a cascaded strategy to reduce false positives. To evaluate our method, we construct a real review room dataset (EDD) with 28,000 images from multiple angles. Our method improves the multiview generalization performance by 4.48% mAP and 5.62% mAR. On the public datasets Pascal VOC-2007 and visDrone-2019, our method is also superior to other suboptimal methods. We propose a framework for review environment detection, which is accurate, fast, and generalizable to other scenarios.

As the population rapidly ages, Alzheimer’s disease (AD), the most common form of dementia, urgently requires the identification of reliable structural brain biomarkers and the development of effective therapeutic strategies. Multiple multi-task learning (MTL) paradigms have been developed to enhance model generalization by sharing information between tasks to predict AD progression and accurately identify MRI-associated biomarkers. Unlike previous MTL approaches that consider only a single kind of cognitive score to predict the complicated AD progression over time, we have developed an innovative MTL method to deal with various cognitive scores simultaneously, with each focusing on different aspects of patient cognition. To effectively capture the intricate associations among different cognitive scores at multiple time points, we first propose an Adaptive Multiple Cognitive Objective Temporal (AMCOT) task-relationship binding penalty mechanism. This mechanism adaptively reveals temporal correlations between various cognitive scores at different time points and uses these relationships to predict cumulative disease progression accurately. To select the most informative MRI features in AD progression, we consider integrating the sparse group Lasso into our model. Our algorithms are designed to handle large datasets efficiently. Empirical evaluation on the Alzheimer’s disease dataset shows that our approach significantly outperforms existing state-of-the-art algorithms in both overall and individual task performance. Additionally, we applied stability selection techniques to identify stable MRI biomarkers and analyzed their temporal patterns to gain insights into AD progression. The implementation source can be found at https://github.com/XuanhanFan/MTL-AMCOT-BB.

Automated Parkinson’s diagnosis(PD) through wearable intelligence technologies has achieved considerable results, but its application in the wild environment presents challenges due to small on-state samples and sparse distribution across time, activity patterns, and subjects. To tackle these challenges, this study proposes a novel data augmentation model to improve PD recognition results in the wild. Our model utilizes a three-level augmentation strategy across different times, patterns, and subjects. Specifically, we apply temporal-level augmentation and aggregation to learn distinct representations, while using pattern/subject-level combinations and augmentations to generate additional samples. As a result, this augmentation approach not only facilitates the acquisition of diverse representations for symptoms but also addresses challenges such as missing data and small sample sizes, which are common for medical data in a free-living environments. This proposed model has applied to a real Parkinson’s Disease (PD) dataset collected in low-resource settings, where it achieves impressive accuracy in the fine-grained classification of PD severity (mild, moderate, severe). In conclusion, this study contributes to more accurate PD self-diagnosis in real-world environments, thereby enabling remote drug intervention guidance from doctors.

Parkinson’s disease (PD) is a progressive neurode-generative disorder characterized by motor symptoms such as tremors, rigidity, and bradykinesia. Accurate and early diagnosis is crucial for effective management and treatment. Some quantitative studies have combined wearable technology with machine learning methods, demonstrating a high potential for practical application. However, these studies mostly use single-location, single-sensor data collected from PD patients in clinical settings, neglecting the diversity of PD symptoms and the real-world application scenarios in free-living environments. This paper proposes an auxiliary diagnosis framework for PD based on multi-location, multi-sensor fusion, and unsupervised domain adaptation. The multi-location, multi-sensor fusion can mitigate the asymmetry of Parkinson’s symptoms, while unsupervised domain adaptation helps transfer in-hospital data to free-living environments without the need for manual labeling of the free-living data. Additionally, this paper designs a multi-head attention mechanism that focuses the disease classifier on sensors with strong feature discrimination and good distribution alignment. This experiment relies on wearable sensor data from 60 PD patients and 12 healthy controls, achieving an impressive accuracy of 90.46%, a precision of 88.28%, a recall of 88.09%, and an F1-score of 88.14%.

Wearable sensors combined with machine learning provide an effective solution for assessing Parkinson’s Disease (PD) severity. However, time-series data from wearable sensors often lack window-level labels for PD severity, resulting in weak supervision, which introduces the challenge of label noise. Additionally, patient variability causes distributional discrepancies, further complicating the learning process. To address these issues, we propose Adaptive Domain-Adversarial Multi-Instance Learning (ADAMIL), which combines and refines Multiple-Instance Learning (MIL) with domain-adversarial techniques. We improve traditional MIL by incorporating self-attention mechanisms and learnable positional encoding, enabling ADAMIL to capture temporal dependencies more effectively, thus making it better suited for mitigating label noise in weakly supervised time-series data. Furthermore, ADAMIL refines domain-adversarial learning to autonomously align latent distributions, ensuring robust domain-invariant feature learning without relying on predefined labels. Experimental results show that ADAMIL achieves 85.29% accuracy and 80.57% F1-score in fine-grained PD severity classification, outperforming existing methods. Notably, this performance is achieved using only a single wrist-worn sensor, underscoring its potential for practical use in clinical and home settings. The code is available at https://github.com/xzxzy12345XZY/ADAMIL.

Adversarial learning has achieved great success for unsupervised domain adaptation (UDA). Existing adversarial UDA methods leverage the predicted discriminative information with Nuclear-norm Wasserstein discrepancy for feature alignment. However, the limited memory space makes it very difficult to accurately calculate the Nuclear-norm, which hinders domain adaptation. To address this challenge, we propose a multi-batch Nuclear-norm adversarial network, termed as MBAN. Specifically, we build a dynamic queue to cache features, which encourages to generate a large and consistent output matrix, enabling accurate calculation of the Nuclear-norm. Then, the multi-batch Nuclear-norm discrepancy is proposed, which can effectively improve the transferability and discriminability of the learned features. Experimental results show that MBAN could achieve significant performance improvement, especially when the number of categories is quite large. Code is available at https://github.com/peiwang0518/Multi-BAN.

Alzheimer’s disease (AD) is the most common dementia in today’s aging society. Accurately predicting its progress remains a major challenge. Multi-task learning methods are widely used in AD research to help understand the progression of AD by predicting cognitive performance and identifying key imaging biomarkers. Previous work has selected representative feature subsets from magnetic resonance imaging (MRI) features. The design of these models is based on the assumption that correlations are consistent across different tasks. Specifically, the model only focuses on a single cognitive score in each prediction and ignores the correlation between different cognitive scores. However, clinicians often use a combination of assessment scores and other tests to more comprehensively assess cognitive status and make a diagnosis. Combining scores from multiple cognitive assessments helps improve accurate predictions of disease progression. Previous research models have primarily focused on predicting a single cognitive score longitudinally. In this paper, we propose a multi-target, multi-task learning method that comprehensively considers the correlation between different cognitive scores and the relationship between longitudinal tasks to simultaneously predict multiple cognitive scores to more comprehensively capture the disease characteristics of development, thereby effectively predicting disease progression. We also adopt a structure matrix to explicitly represent the correlation between tasks, further improving the accuracy and interpretability of the model. Results from extensive experiments using the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset show that our method exhibits balanced multi-target performance when dealing with three cognitive scores. Compared to models focusing on a single cognitive target score, our method performs better in the early prediction of cognitive scores.

Multimodal learning is widely used in automated early diagnosis of Alzheimer's disease. However, the current studies are based on an assumption that different modalities can provide more complementary information to help classify the samples from the public dataset Alzheimer's Disease Neuroimaging Initiative (ADNI). In addition, the combination of modalities and different tasks are external factors that affect the performance of multimodal learning. Above all, we summrise three main problems in the early diagnosis of Alzheimer's disease: (i) unimodal vs multimodal; (ii) different combinations of modalities; (iii) classification of different tasks. In this paper, to experimentally verify these three problems, a novel and reproducible multi-classification framework for Alzheimer's disease early automatic diagnosis is proposed to evaluate and verify the above issues. The multi-classification framework contains four layers, two types of feature representation methods, and two types of models to verify these three issues. At the same time, our framework is extensible, that is, it is compatible with new modalities generated by new technologies. Following that, a series of experiments based on the ADNI-1 dataset are conducted and some possible explanations for the early diagnosis of Alzheimer's disease are obtained through multimodal learning. Experimental results show that SNP has the highest accuracy rate of 57.09% in the early diagnosis of Alzheimer's disease. In the modality combination, the addition of Single Nucleotide Polymorphism modality improves the multi-modal machine learning performance by 3% to 7%. Furthermore, we analyse and discuss the most related Region of Interest and Single Nucleotide Polymorphism features of different modalities.

Alzheimer's disease (AD), the most prevalent dementia, gradually reduces the cognitive abilities of patients while also posing a significant financial burden on the healthcare system. A variety of multi-task learning methods have recently been proposed in order to identify potential MRI-related biomarkers and accurately predict the progression of AD. These methods, however, all use a predefined task relation structure that is rigid and insufficient to adequately capture the intricate temporal relations among tasks. Instead, we propose a novel mechanism for directly and automatically learning the temporal relation and constructing it as an Automatic Temporal relation Graph (AutoTG). We use the sparse group Lasso to select a universal MRI feature set for all tasks and particular sets for various tasks in order to find biomarkers that are useful for predicting the progression of AD. To solve the biconvex and non-smooth objective function, we adopt the alternating optimization and show that the two related sub-optimization problems are amenable to closed-form solution of the proximal operator. To solve the two problems efficiently, the accelerated proximal gradient method is used, which has the fastest convergence rate of any first-order method. We have preprocessed three latest AD datasets, and the experimental results verify our proposed novel multi-task approach outperforms several baseline methods. To demonstrate the high interpretability of our approach, we visualise the automatically learned temporal relation graph and investigate the temporal patterns of the important MRI features. The implementation source can be found athttps://github.com/menghui-zhou/MAGPP.

User counterparts, such as user attributes in social networks or user interests, are the keys to more natural Human–Computer Interaction (HCI) . In addition, users’ attributes and social structures help us understand the complex interactions in HCI. Most previous studies have been based on supervised learning to improve the performance of HCI. However, in the real world, owing to signal malfunctions in user devices, large amounts of abnormal information, unlabeled data, and unsupervised approaches (e.g., the clustering method) based on mining user attributes are particularly crucial. This paper focuses on improving the clustering performance of users’ attributes in HCI and proposes a deep graph embedding network with feature and structure similarity (called DGENFS ) to cluster users’ attributes in HCI applications based on feature and structure similarity. The DGENFS model consists of a Feature Graph Autoencoder (FGA) module, a Structure Graph Attention Network (SGAT) module, and a Dual Self-supervision (DSS) module. First, we design an attributed graph clustering method to divide users into clusters by making full use of their attributes. To take full advantage of the information of human feature space, a k-neighbor graph is generated as a feature graph based on the similarity between human features. Then, the FGA and SGAT modules are utilized to extract the representations of human features and topological space, respectively. Next, an attention mechanism is further developed to learn the importance weights of different representations to effectively integrate human features and social structures. Finally, to learn cluster-friendly features, the DSS module unifies and integrates the features learned from the FGA and SGAT modules. DSS explores the high-confidence cluster assignment as a soft label to guide the optimization of the entire network. Extensive experiments are conducted on five real-world data sets on user attribute clustering. The experimental results demonstrate that the proposed DGENFS model achieves the most advanced performance compared with nine competitive baselines.

Multi-task learning models based on temporal smoothness assumption, in which each time point of a sequence of time points concerns a task of prediction, assume the adjacent tasks are similar to each other. However, the effect of outliers is not taken into account. In this paper, we show that even only one outlier task will destroy the performance of the entire model. To solve this problem, we propose two Robust Temporal Smoothness (RoTS) frameworks. Compared with the existing models based on temporal relation, our methods not only chase the temporal smoothness information but identify outlier tasks, however, without increasing the computational complexity. Detailed theoretical analyses are presented to evaluate the performance of our methods. Experimental results on synthetic and real-life datasets demonstrate the effectiveness of our frameworks. We also discuss several potential specific applications and extensions of our RoTS frameworks.

The diagnosis of depression mainly depends on clinicians’ experience and questionnaire results, making it a time-consuming and subjective process that demands significant allocation of human resources. Numerous automatic depression estimation (ADE) systems, based on facial cues, have been introduced to estimate the severity and assist clinicians in diagnosis. However, traditional methods adopt a single sampling frame rate, which makes it leads to a tradeoff between the loss of critical vision information and calculation redundancy. In this paper, we propose a Multi-Frame Rate Attention Convolutional Neural Network (MFRA) to effectively mine the facial cues of patients for estimating the severity of depression. Specifically, we adpot a two-branch network structure, in which one branch uses high frame sampling rate to capture more nuances of facial changes, and the other uses low frame sampling rate to focus more on the spatial information of the video. Furthermore, considering the manually selected local facial features will introduce noise, we introduce attention modules to make MFRA concentrate on the facial regions related to depression. Finally, the feature vectors extracted from the two branches are aggregated to output the severity of depression. The experimental results on two datasets, AVEC 2013 and AVEC 2014, show that this method can effectively capture spatiotemporal features, and the prediction results are superior to most video-based depression prediction methods.

Internet of Things (IoT) enabled technology has rapidly and efficiently facilitate healthcare diagnose and treatment with low-cost and lightweight devices. Big data generated from IoT offers valuable and crucial information to guide decision-making, improve patient outcomes, and decrease healthcare costs, etc. The workshop is aiming to provide an opportunity for researchers and practitioners from both academia and industry to present the state-of-the-art research and applications in utilizing IoT and big data technology for healthcare by presenting efficient scientific and engineering solutions, addressing the needs and challenges for integration with new technologies, and providing visions for future research and development.

In this paper, our goal is to design a simple learning paradigm for long-tail visual recognition, which not only improves the robustness of the feature extractor but also alleviates the bias of the classifier towards head classes while reducing the training skills and overhead. We propose an efficient one-stage training strategy for long-tailed visual recognition called Global and Local Mixture Consistency cumulative learning (GLMC). Our core ideas are twofold: (1) a global and local mixture consistency loss improves the robustness of the feature extractor. Specifically, we generate two augmented batches by the global MixUp and local CutMix from the same batch data, respectively, and then use cosine similarity to minimize the difference. (2) A cumulative head-tail soft label reweighted loss mitigates the head class bias problem. We use empirical class frequencies to reweight the mixed label of the head-tail class for long-tailed data and then balance the conventional loss and the rebalanced loss with a coefficient accumulated by epochs. Our approach achieves state-of-the-art accuracy on CIFAR10-LT, CIFAR100-LT, and ImageNet-LT datasets. Additional experiments on balanced ImageNet and CIFAR demonstrate that GLMC can significantly improve the gen-eralization of backbones. Code is made publicly available at https://github.com/ynu-yangpeng/GLMC.

Federated learning (FL) is a rising distributed machine learning area, which aims to train a high-performing global model with data collected from a number of local clients. Many FL applications receive data over time in the form of data streams. Streaming data are likely to suffer concept drift. It can significantly harm a model’s predictive ability. However, no study has characterized concept drift in FL or investigated how it can affect the global and local models’ performance. This paper aims to provide such understanding by 1) categorizing concept drift in temporal and spatial dimensions with ten features and 2) investigating the impact of the features in depth. We find that: the temporal features degrade FL models to a different extend and do not affect model convergence after the new data concept becomes stable; the spatial features cause data heterogeneity and affect both accuracy and convergence speed.

Internet of Things (IoT) Wearable technology plays a crucial role in assisting the diagnosis of Parkinson’s disease (PD), and an efficient model for auxiliary diagnosis of the severity of PD can help reduce the workload for doctors. However, due to the influence of data collection environments and annotators, noisy label data is inevitable, which may have a negative impact on modeling the severity of PD. To address the above challenges, on the one hand, we collected a large number of activity signal data of Parkinson’s patients in free-living environments, and on the other hand, we proposed an efficient PD stage assessment framework, which includes a noisy label processing method to alleviate the noisy label negative impact. Specifically, we collected signal data from 15 healthy controls and 68 PD patients through 12 activities, and then we proposed a framework for noisy label detection and correction. The experimental results on real PD data sets demonstrated that the proposed framework achieve 75.9% accuracy in PD stage assessment and significantly improve the classification performance of different types of basic classifiers, which is better than other noisy label detection algorithms and other PD stage assessment frameworks. Overall, in this work, we focus on modeling PD severity in free-living environments using a single wearable sensor and reducing the negative impact of noisy label data to better help PD patients manage the disease.

An effective auxiliary diagnostic model for the severity of Parkinson’s disease (PD) could help hospitals reduce their workload, particularly in nations or regions where medical resources are limited. However, a critical challenge persists that hampers the progress of such endeavors. Previous studies have employed label propagation techniques that assign uniform labels to all activity signal segments of a patient, neglecting the complex expression of PD symptoms, thereby introducing label noise. To confront this challenge, we have collected an extensive set of PD activity signals from a clinical setting and have proposed an efficient and robust framework for assessing PD severity. Specifically, we gathered wearable device data on 14 daily activities from 70 PD patients, based on the Unified Parkinson’s Disease Rating Scale Part III. Our data analysis indicates that many segments within the activities were incorrectly labeled, significantly impairing the classification performance of the model. We introduced a novel framework based on Multi-Instance Learning with a Re-weighted Discriminative Instance Mapping (RDIM) to model PD auxiliary diagnosis, aiming to eliminate the impact of label noise present in the data. The results demonstrate that our framework achieves an accuracy of 80.88% in classifying the severity of PD, effectively addressing the label noise caused by coarse-grained label propagation.

Wearable technology has played a crucial role in computer-aided diagnosis and long-term monitoring of Parkinson’s disease (PD). How to efficiently and accurately assess the severity of Parkinson’s disease using wearable devices remains the essential problem. However, in the real free-living environment, we have encountered two issues: weak annotation and class imbalance, which could potentially impede the automatic assessment of Parkinson’s disease. To overcome these challenges, we propose a novel Parkinson’s disease assessment framework in free-living environment. Specifically, clustering methods are used to learn latent categories from the same activities, and use Latent Dirichlet allocation (LDA) topic models to capture latent features of multiple activities. Then, to mitigate the impact of data imbalance, we augment bag-level data while retaining key instance prototypes. The new framework is applied to a PD dataset collected by wearable sensors in the wild. It achieves an impressive 73.49% accuracy in the fine-grained (normal, mild, moderate, severe) classification of PD severity based on hand movements. Overall, this study contributes to more accurate PD self-diagnosis in the wild, enabling remote guidance for drug intervention from doctors.

The prediction of mobility change is a crucial aspect of infectious disease prevention and control, as individual movement can lead to the spread of infectious diseases. After an epidemic outbreak, countries usually implement a series of epidemic prevention and control policies to restrict personnel travel. This paper aims to measure the impact of national epidemic prevention and control policies on mobility change. To achieve this goal, we propose a mobility change prediction model based on LSTM and "local" fully connected layers. The results show that the model’s mobility change predictions in the six mobility categories are consistent with reality, and the model is optimal overall.

In order to control the first wave of COVID-19 pandemic in 2020, many models have shown effectiveness in predicting the spread of new coronary pneumonia and the different interventions. However, few models can collect large amounts of high-quality real-time data faster under the premise of protecting privacy, considering the impact of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) variant and the mass vaccination program as a new intervention. Therefore, we developed a mobile intelligent application that can collect a large amount of real-time data while protecting privacy and conducted a feasibility study by defining a new COVID-19 mathematical model SEMCVRD. By simulating different intervention measures, the prediction model of the mobile intelligent application used in this article simulates the epidemic situation in the U.K. as an example. The findings are as below: the optimal intervention strategy is to suppress the intervention at$P=3$(intervention intensity: the average number of contacts per person per day) before the end of March 2021, then gradually release the intervention intensity at a rate of$P+2$, and finally release the intensity to$P=9$in June 2021. The COVID-19 pandemic will end at the end of June 2021, when the total number of deaths will reach 128772. This strategy will be able to balance the tradeoff between loss of life and economic loss. Compared with the official statistics released by the U.K. government on May 31, 2021, our model can accurately predict the relative error rate of the total number of cases is less than 6.9%, and the relative error rate of the total number of deaths is less than 1%. Furthermore, the model is also suitable for collecting data from countries/regions around the world.

In this paper, we propose a novel efficient multi-task learning formulation for the class of progression problems in which its state will continuously change over time. To use the shared knowledge information between multiple tasks to improve performance, existing multi-task learning methods mainly focus on feature selection or optimizing the task relation structure. The feature selection methods usually fail to explore the complex relationship between tasks and thus have limited performance. The methods centring on optimizing the relation structure of tasks are not capable of selecting meaningful features and have a bi-convex objective function which results in high computation complexity of the associated optimization algorithm. Unlike these multi-task learning methods, motivated by a simple and direct idea that the state of a system at the current time point should be related to all previous time points, we first propose a novel relation structure, termed adaptive global temporal relation structure (AGTS). Then we integrate the widely used sparse group Lasso, fused Lasso with AGTS to propose a novel convex multi-task learning formulation that not only performs feature selection but also adaptively captures the global temporal task relatedness. Since the existence of three non-smooth penalties, the objective function is challenging to solve. We first design an optimization algorithm based on the alternating direction method of multipliers (ADMM). Considering that the worst-case convergence rate of ADMM is only sub-linear, we then devise an efficient algorithm based on the accelerated gradient method which has the optimal convergence rate among first-order methods. We show the proximal operator of several non-smooth penalties can be solved efficiently due to the special structure of our formulation. Experimental results on four real-world datasets demonstrate that our approach not only outperforms multiple baseline MTL methods in terms of effectiveness but also has high efficiency.

Unsupervised domain adaptation, which transfers knowledge from the source domain to the target domain, has still been a challenging problem. However, previous domain adaptation methods typically minimize the domain discrepancy by using the pseudo target labels. Since the pseudo labels can be noisy, which may cause misalignment and unsatisfying adaptation performance. To address the above challenges, we propose an information maximization adaptation network with label distribution priors. We revisit feature alignment in unsupervised domain adaptation from the perspective of distribution alignment, and find that learning discriminant feature representation requires to minimizing distribution discrepancy and maximizing source mutual information between the outputs of the classifier and feature representations. Due to domain shift, maximizing target mutual information may align features to incorrect class directly. We propose a weighted target mutual information by re-weighting the estimated mutual information via the mean prediction confidence in mini-batch, which can eliminate the negative impact of inaccurate estimation. In addition, we introduce a regularization term of label priors distribution to encourage the similarity to the real label distribution. Extensive experimental results on three benchmark datasets show that our proposed method can achieve remarkable results compared with previous methods.

Histopathological images, as the gold standard for cancer diagnosis, record abundant information about microscopic structures and morphological characteristics through staining and scanning. However, the stain variation caused by the deviations in the production process not only confuse the pathological diagnosis, but also depress the generalization of computer-aided diagnosis. Stain normalization provides an efficient preprocessing method, which generates images with the same stain style while preserving the texture information. In this paper, we extend the advantages of adversarial transfer learning and auto-encoder structure to propose an adversarial stain normalization network (ASNN). For the stain style alignment, ASNN liberalizes the selection for the specific template image and adopts a domain discriminator after encoding to eliminate the distribution discrepancies of latent features caused by stain variation. To ensure precisely stain style transfer, we reconstruct the target stain style image using a reconstruction loss-constrained autoencoder and by the loss of reconstruction so that the stain decoder learns the rules for texture-to-stain conversions. We simulate the experiments on two publicly available datasets and utilize multiple indicators to evaluate the qualities of generated images. By comparing classification performance on downstream tasks, ASNN exhibits its superiority than traditional stain normalization methods. Furthermore, ASNN adopts a lighter network structure, which is faster than traditional stain normalization methods, and can stain 500 images within 1 second.

In this paper, we propose a multi-task learning approach for predicting the progression of Alzheimer's disease (AD), known as the most common form of dementia. The vital challenge is to identify how the tasks are related and build learning models to capture such task relatedness. Unlike previous methods that assume low-rank structure, chase the predefined local temporal relatedness or utilize local approximation, we propose a novel penalty termed L ongitudinal S tability A djustment (LSA) to adaptively capture the intrinsic global temporal correlation among multiple time points and thus utilize the accumulated disease progression information. We combine LSA with sparse group Lasso to present a novel multi-task learning formulation to identify biomarkers closely related to cognitive measurement and predict AD progression. Two efficient algorithms are designed for large-scale dataset. Experimental results conducted on two AD data sets demonstrate our framework outperforms competing methods in terms of overall and each task performances. We also perform stability selection to identify stable biomarkers from the MRI feature set and analyze their temporal patterns in disease progression.

The study of microstructure of materials is of great significance in the field of materials science. The interdisciplinary cooperation of materials science and computer science makes it more accurate and efficient to explore the relationship between material microstructure and material properties. Machine learning has potential in exploring the relationship between microstructure and properties of materials. This paper proposes adding the morphology features of grains into the construction of grain knowledge graph to enrich the grain information in the graph. First, an autoencoder extracts the grain morphology features and adds them to the grain knowledge graph. Then, the graph convolutional network is used to extract the features of the graph, and the fully connected network is used to predict the properties of the material. Experiments are performed on actual EBSD scanning data. The experimental results show that the proposed method has noticeable improvement over the competing methods.

Alzheimer's Disease (AD) is the most common reason of dementia that causes serious problems in patients' congnitive functions. Multi-task learning (MTL) has performed well in studies of longitudinal processes in Alzheimer's disease for revealing the progression of AD. Combined with prior knowl-edges in disease progression or medical science, regularization MTL framework could introduce empirical constraints more flexibly. Meanwhile, it brings higher cost during optimization. While it shown that most of formulations could not define the disease progression precisely. Existing regression methods with temporal smoothness method eliminated abnormal fluctuation of cognitive scores, and neglected the sophisticated progression in disease. In this article, we proposed an analytic method to define the progression of AD, and a flexible bandwidth method to encourage the points of disease time sequence temporal smoothness in an appropriate way. To solve three non-smooth penalties in our method, we proposed an optimization method combined accelerated gradient descent (AGD) and alternating direction method of multipliers (ADMM).

The clinical progression of Alzheimer's disease( AD ) can't be accurately evaluated by single modality data alone. Multi-modal data have a good effect on the diagnosis of AD. Clarifying the complementarity between modalities is crucial for the assessment of each stage of AD. Few studies have specifically explored the complementarity between different modalities due to the lack of completely aligned and paired multi-modal data and the limitation of sample size. However, collecting the full set of aligned and paired data is expensive or even impractical. In addition, the limited number of samples poses a great challenge to the robustness of the model. In this paper, different machine learning( ML ) methods were used to explore data complementarity between T1-weighted magnetic resonance imaging ( MRI ), cerebrospinal fluid ( CSF ), and fluorodeoxyglucose-positron emission tomography ( FDG-PET ) modalities. The different modal data of Alzheimer's Neuroimaging Initiative ( ADNI ) and the self-extracted neuroimaging data were experimentally explored. Experiments show that there is obvious complementarity between MRI and CSF. By fusing MRI and CSF data, three binary classification tasks using multi-modal fusion data have achieved varying degrees of improvement. At the same time, we also explored the important features of multi-modal fusion data through SHapley Additive exPlanations ( SHAP ), and found that most important features are supported by relevant literature.

Medical image classification is an important task in computer-aided diagnosis systems. Its performance is critically determined by the descriptiveness and discriminative power of features extracted from images. With rapid development of deep learning, deep convolutional neural networks (CNNs) have been widely used to learn the optimal high-level features from the raw pixels of images for a given classification task. However, due to the limited amount of labeled medical images with certain quality distortions, such techniques crucially suffer from the training difficulties, including overfitting, local optimums, and vanishing gradients. To solve these problems, in this article, we propose a two-stage selective ensemble of CNN branches via a novel training strategy called deep tree training (DTT). In our approach, DTT is adopted to jointly train a series of networks constructed from the hidden layers of CNN in a hierarchical manner, leading to the advantage that vanishing gradients can be mitigated by supplementing gradients for hidden layers of CNN, and intrinsically obtain the base classifiers on the middle-level features with minimum computation burden for an ensemble solution. Moreover, the CNN branches as base learners are combined into the optimal classifier via the proposed two-stage selective ensemble approach based on both accuracy and diversity criteria. Extensive experiments on CIFAR-10 benchmark and two specific medical image datasets illustrate that our approach achieves better performance in terms of accuracy, sensitivity, specificity, and F1 score measurement.

Zero-Shot Learning (ZSL) is an emerging research that aims to solve the classification problems with very few training data. The present works on ZSL mainly focus on the mapping of learning semantic space to visual space. It encounters many challenges that obstruct the progress of ZSL research. First, the representation of the semantic feature is inadequate to represent all features of the categories. Second, the domain shift problem still exists during the transfer from semantic space to visual space. In this paper, we introduce knowledge sharing (KS) to enrich the representation of semantic features. Based on KS, we apply a generative adversarial network to generate pseudo visual features from semantic features that are very close to the real visual features. Abundant experimental results from two benchmark datasets of ZSL show that the proposed approach has a consistent improvement.

Suffering from the semantic insufficiency and domain-shift problems, most of existing state-of-the-art methods fail to achieve satisfactory results for Zero-Shot Learning (ZSL). In order to alleviate these problems, we propose a novel generative ZSL method to learn more generalized features from multi-knowledge in semantic-to-visual embedding. In our approach, the proposed Multi-Knowledge Fusion Net-work (MKFNet) alleviates the semantic insufficiency problem by fusing the domain information of different knowledge, which enables more relevant semantic features to be trained for semantic-to-visual feature embedding. The pro-posed knowledge regularization LKRgreatly improves the intersection between the synthesized visual features generated by MKFNet and the unseen visual features, which can alleviate the domain-shift problem. Empirically, we show that our approach consistently outperforms these state-of-the-art methods on a large number of available benchmarks on the generalized ZSL (GZSL).

Recent advances in deep neural network have achieved remarkable successes in various computer vision tasks. However, deep neural network with millions of parameters may suffer from poor generalization due to overfitting. To improve the generalization performance, many methods have been proposed such as data augmentation, label smoothing and knowledge distillation. In this paper, we extend self-knowledge distillation to enhance the model generalization performance without incurring extra computation cost, called EWR-KD, which is a simple yet effective method to progressively distill knowledge from the model itself. Concretely, it consists of two components: 1) the self-distillation scheme that progressively softens the learning targets by using the past model prediction; 2) the sample-reweighting scheme that dynamically decides the trust degree to transfer more informative knowledge by introducing uncertainty estimation. With the two components, EWR-KD is robust to both corrupt noises and adversarial noises, and can be easily combined with current advanced regularization techniques. We theoretically show that EWR-KD minimizes cross-entropy by adding an epoch-wise regularization, which measures the difference between the past model prediction at ($t-1$)-th epoch and the current prediction at$t$-th epoch. Finally, Extensive experimental results on clean datasets and noisy datasets empirically demonstrate that EWR-KD not only improves the performance of the state-of-the-art baseline but also yields well calibration.

With the aggravation of the population aging problem, Parkinson’s disease (PD) and other neurodegenerative diseases of the elderly are not only a medical problem but also an important social problem. Therefore, early detection of PD is particularly important for reducing complications. Currently, the diagnosis of PD is assessed by specialized physicians through the Uniform PD Rating Scale (UPDRS). This limits the detection rate of PD and the timely assessment of disease progression to a certain extent. Moreover, with the development of artificial intelligence, machine learning has been widely and effectively applied to the assessment and monitoring of PD. Therefore, we use machine learning to distinguish between healthy people and PD patients based on UPDRS. In this paper, we collaborated with the First People’s Hospital of Yunnan Province to collect exercise data from 15 healthy individuals and 15 PD patients using wearable motion sensors. The analysis found that not all activities collected according to the UPDRS were useful. According to our proposed Indicators for distinguishing activities (I-DA) method as defined in this article, the most differentiated activities are found. Retain the activities that contain the most discriminative information, and use these activities to distinguish between healthy people and PD patients. We verify the effectiveness of this method through experiments. We use k-Nearest Neighbor (KNN), eXtreme Gradient Boosting (XGB), and Support Vector Machine (SVM) to execute the classification method. When the selected activities were taken as the whole data set rather than all activities according to our proposed Indicators for distinguishing activities (I-DA) method, the classification accuracy of KNN and XGB were improved by 5.10% and 2.4% respectively. The classification accuracy of SVM was improved by 12.07%. The experimental results show that the accuracy is significantly improved.

Alzheimer’s Disease (AD) is the most common dementia characterized by loss of brain function. Multi-tasking learning methods have been widely used to predict cognitive performance and select important imaging biomarkers in AD research. The temporal smoothness assumption, prevalent for modeling AD progression, means the difference between cognitive scores at two consecutive time points is relatively small. However, it’s not appropriate due to the presence of sample disturbance and the effectiveness of drug therapy. In addition, many multi-task learning methods select discriminative feature subset from MRI features, assuming that correlations between tasks are consistent, which ignores the complex intrinsic correlation structure of tasks. In this paper, we present a multi-task learning framework which utilizes generalized fused Lasso and generalized group Lasso (GFGGL for abbreviation) to model the disease progression with the complex intrinsic nonlinear structures of disease. The proposed framework is more flexible to utilize the inherent nonlinear relation of AD than existing methods for the reason of we represent the intrinsic structure as three correlation matrices which are functions of super parameters. The framework involves (1) two nonlinear structures of disease progression and (2) one nonlinear structure among tasks. An efficient optimization method is designed for the difficult optimization problem due to the presence of three nonsmooth penalties. Extensive experimental results using dataset from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) demonstrate the effectiveness of the proposed method.

Internet of Things (IoT) provides ubiquitous intelligence and pervasive interconnections to diverse physical objects. A key technology to seamlessly integrate different IoT devices into an IoT system is IoT middleware, a software system layer designed to be the intermediary between IoT devices and applications. However, two issues,communication gapandheterogeneous access, prevent the existing IoT middleware from effective application in an IoT system with heterogeneous standards and interfaces. To address this problem, inspired by a graph-based knowledge system for eliminating heterogeneity in business systems, we, in this article, propose a knowledge graph-based multilayer IoT middleware. The proposed multilayer IoT middleware introduces a new layer to bridge the gap between IoT devices with different communication protocols. It is able to uniformly manage all IoT devices by using an IoT knowledge graph. We evaluate the applicability of the proposed approach by a real-life IoT project, a remote monitoring project of rural sewage treatment stations located in Yunnan Province, China. We find that the proposed approach effectively resolves the communication gap and heterogeneous access problems that occurred in the system.

As the size of available networks is continuously increasing (even with millions of nodes), large-scale complex networks are receiving significant attention. While existing overlapping-community detection algorithms are quite effective in analyzing complex networks, most of these algorithms suffer from scalability issues when applied to large-scale complex networks, which can have more than 1,000,000 nodes. To address this problem, we propose an efficient local-expansion-based overlapping-community detection algorithm using local-neighborhood information (OCLN). During the iterative expansion process, only neighbors of nodes added in the last iteration (rather than all neighbors) are considered to determine whether they can join the community. This significantly reduces the computational cost and enhances the scalability for community detection in large-scale networks. A belonging coefficient is also proposed in OCLN to filter out incorrectly identified nodes. Theoretical analysis demonstrates that the computational complexity of the proposed OCLN is linear with respect to the size of the network to be detected. Experiments on large-scale LFR benchmark and real-world networks indicate the effectiveness of OCLN for overlapping-community detection in large-scale networks, in terms of both computational efficiency and detected-community quality.

Electroencephalogram (EEG) signal is often used to assess sleep quality and treat sleep disorders. Many existing methods usually obtain high accuracy through a large number of feature preprocessing and feature extraction of EEG signals, which need a lot of prior knowledge as the basis. In this paper, a novel sleep stage classification framework, named FRL&S2SL, is proposed. The framework combines fast representation learning (FRL) and semantic-to-signal learning (S2SL) and uses single-channel EEG without any preprocessing of EEG signals. In the proposed framework, we utilize convolutional neural networks (CNN) to extract time-invariant features and bidirectional long short-term memory (BiLSTM) models to extract temporal features. Furthermore, auxiliary classifier generative adversarial network (ACGAN) is used to embed semantic features into signal features and to extract knowledge domain features of EEG signals for the first time. According to the American Academy of Sleep Medicine (AASM), sleep is divided into five stages: awake, rapid eye movement (REM) and three non-rapid eye movement (N1/N2/N3). We evaluated our framework using single-channel EEG (Fpz-Oz) from Sleep-EDF dataset, which is subject to the standards specified by AASM. The results show that our framework has achieved state-of-the-art in many evaluation metrics.

Proper scoring of sleep stages may offer more intuitive clinical information for assessing the sleep health and improving the diagnosis of sleep disorders in the smart home healthcare. It usually depends on an accurate analysis of the collected physiological signals, especially for the raw sleep Electroencephalogram (EEG). Most of the methods currently available just rely on the pre-processing or handcrafted features that need prior knowledge and preliminary analysis from the sleep experts and only a few of them take full advantage of the temporal information such as the inter-epoch dependency or transition rules among stages, which are more effective for identifying the differences among the sleep stages. In such cases, we proposed a novel hybrid neural network named HNSleepNet. It utilizes a two-branch CNN with multi-scale convolution kernels to capture the time-invariant features from the adjacent sleep EEG epochs both in time and frequency domains automatically, and attention-based residual encoder-decoder LSTM layers to learn the inter-epoch dependency and transition rules at the Sequence-wise level. After the two-step training, HNSleepNet can perform sequence-to-sequence automatic sleep staging with a raw single channel EEG in an end-to-end way. As the experimental results demonstrated, its performance achieved a better overall accuracy and macro F1-score (MASS: 88%, 0.85, Sleep-EDF: 87%-80%, 0.79-0.74) compared with the state-of-the-art approaches on various single-channels (F4-EOG (Left), Fpz-Cz and Pz-Oz) in two public datasets with different scoring standards (AASM and R&K), We hope this progress can make clinically practical value in promoting home sleep studies on various home health-care devices.

Facial expressions contain rich emotional information, which is an important method in human communication. At present, most of the researches on facial expression recognition is conducted on the frontal faces. However, in real life, the captu re device may capture facial expression data from various poses. Different from the existing techniques, in this paper, We propose a multitask deep learning method that uses the links between var ious poses and expressions to improve the accuracy of expression recognition. We use the adversarial network to supplement the e xpression information of the face for the head poses with a sever e lack of expression information. There are some advantages: Firstly, we use vgg16 to train the expression images of each deflecti on angle separately and find that the expression recognition acc uracy rate for small offset angle (-30 °, -15 °, 15°, 30 °) is larger t han 0 ° angle when the face angle is greater than 45 °, the accur acy of recognition decreases sharply. So we use multitask learnin g to jointly train these small offsets angle images and frontal ima ges, the multitask learning can learn the emotion-preserving repr esentations at various poses to predict the expression class label f rom the input face, and bring again to its recognition accuracy r ate. Secondly, for the poses of a severe lack of expression inform ation, we use TP-GAN to convert a large deflection pose image i nto a frontal face and supplement its expression information. Th e experimental results show that our proposed algorithm has a g ood recognition effect on facial expressions for all poses. Compa red with the most advanced expression recognition methods, this paper has also achieved the state-of-the-art recognition results.

Human activity recognition (HAR) has been widely used in some areas such as smart home, health care and so on. However, there are still some low recognition accuracy cases in actual scenarios. In order to improve the accuracy of recognition, we propose a multi-feature weighted ensemble classification method on triaxial accelerometer sensor data. We perform weighted integration on five base classifiers to obtain the final prediction classification label. Among these five base classifiers, three are K-nearest neighbor (KNN) classifiers with different features respectively using three traditional feature extraction methods from original data. Another two are currently popular deep learning models-Attention Mechanisms on Long Short-Term Memory Network (Attention-LSTM) and Convolutional Neural Network (CNN), which can automatically extract features and classify. We demonstrated the feasibility of this ensemble method on a dataset containing eight human daily activities. Comparing experimental results, our method achieved the best recognition effect, with an accuracy of 95.58%.

In the context of the continuous development of Internet of Things (IoT) technology and Machine learning (ML) technology, its application in the medical field is becoming more and more extensive. However, with a dramatic increase in medical data obtained from the IoT-based medical auxiliary diagnosis system, the impact of label noise problems is also increasing. When training a machine learning algorithm for a supervised-learning task in some clinical applications, uncertainty in the labels of some patients may adversely affect the performance of the algorithm. For example, due to ambiguous patient conditions or poor reliability of diagnostic criteria, even clinical experts may lack confidence in making medical diagnoses for some patients. As a result, some samples used in algorithm training may be mislabeled, which adversely affects the performance of the algorithm. In this paper, we study a classification problem of sample labels with random damage. We propose a new hybrid label noise correction model that generalizes many learning problems, including supervised, unsupervised and semi-supervised learning. This hybrid model can withstand the negative effects of random noise and various non-random label noise. Extensive experimental results using real-world datasets from UCI machine learning repository are provided, the empirical study shows that our approach successfully improves data quality in many cases, in terms of classification accuracy, over existing label noise correction methods.

Traffic sign classification is the core of intelligent transportation and fundamental for constructing an automatic driving system. While supervised classification tasks demonstrate promising classification performance, a particular challenge is how to ensure the confidence for collecting labelled data. This paper presents a semi-supervised approach via label confidence for traffic sign classification to avoid the interference of uncertain labelled data. The idea of the proposed approach is to compare unsupervised information of the data, the supervised information carried by the learning data, and the supervised information which is given by the classification model in order to detect inconsistencies. The approach is able to build a robust classification model. Experimental results on benchmark and real-world dataset demonstrate that our approach significantly outperforms the existing approaches when uncertain labelled data exists.

Recent emergence of “microblogging” services has been driving cyber-physical social system (CPSS) as a hot topic in real-world applications. How to efficiently detect and recognise spam and fake accounts becomes an important task where it requires analysis of microblog user behavior and prediction of their activity. This article attempts to investigate this challenge by proposing a new strategy to effectively model microblogging user activity and dynamically predicting their activities for the CPSS applications. We first analysis and define a set of benchmarks for measuring microblogging user activeness in considering serval key dynamic attributes including change rate of microblogging numbers, user attentions, etc. Then, we build up a new dynamic microblogging user activity prediction model (DUAPM) based on three important characteristics: personal information, social relationship, and user interaction. Finally, an improved logical regression algorithm is proposed for training the model and predicting user activity. Under the evaluation of a sample dataset containing Sina Weibo 3621 users over 20 weeks, it shows that our model deliver average up to 3% higher prediction accuracy than other social media user activity prediction models using traditional logical regression and random forest algorithms. We also take out a CPSS case study of evaluating DUAPM models for analysis and prediction of Twitter users' activity over 16 countries. The results show that our model effectively reflects the distribution and trends of Twitter users' activity with different background and cultures.

In recent years, many cutting-edge research results have emerged in the field of computer vision, especially in the field of image classification. But, researchers are still trying to explore more sophisticated models to further improve the accuracy of image classification. However, many models have failed to achieve satisfactory results due to the complexity of the image and the problems of image noise. Therefore, in order to solve these problems, this paper proposes an unsupervised feature extraction method called multi-granularity convolution denoising autoencoder (MGCDAE). Based on convolutional neural network, the method proposed the concept of multi-granularity convolution kernel to solve the problem of complex image feature extraction. In addition, we introduce denoising autoencoder (DAE) for image noise, which enables our approach to extract more robust features from noise images. The high-level features extracted by the above method are sent to the softmax classifier for classification to evaluate the validity of the feature extraction. Our method has been evaluated on three benchmark data sets, which results show that our approach can extract more discriminative high-level features compared with other existing algorithms.

In medical text auxiliary diagnosis systems, there exists some problems including few labeled samples, imbalanced classes and domains are related but different. Taking advantages of transfer learning, we propose the multi-source transfer learning approach based on ensemble learning to address the above problems. Source data sampling method is designed to ensure the transfer ability of source samples. Then, three classifiers are ensembled to guarantee the robustness. Finally, classifiers from multiple domains are reasonably combined using mutual information to further improve performance. Our approach has been evaluated on the benchmark medical text datasets, and the results show that our approach is superior to the existing algorithms and can meet the requirement of an auxiliary diagnosis in certain extent.

Translation is a data augmentation method widely used in the image classification tasks. We analyze the mechanism of translation and discover that the central area of the images is more likely to be selected as convolutional neural network's input. Inspired by the structure of human retina, we propose the hypothesis that the central area of the image contains more significant information than the marginal one. Comprehensive experiments are presented to prove that hypothesis and reach the conclusion that the receptive field is nonuniform. The central part of the image that is always selected by translation is called the focused area. Motivated by the demand to take use of different focused area and thus take use of different receptive fields, we propose a novel training mechanism that integrate different focused areas in one training process. Our method consists of several stages, each has its own focused area and learning rate, and achieves considerable result in the experiments. We call the integration of focused areas the ensemble of receptive fields.

Modeling and predicting progression of chronic diseases like Alzheimer's disease (AD) has recently received much attention. Traditional approaches in this field mostly rely on harnessing statistical methods into processing medical data like genes, MRI images, demographics, etc. Latest advances of machine learning techniques grant another chance of training disease progression models for AD. This trend leads on exploring and designing new machine learning techniques towards multi-modality medical and health dataset for predicting occurrences and modeling progression of AD. This paper aims at giving a systemic survey on summarizing and comparing several mainstream techniques for AD progression modeling, and discuss the potential and limitations of these techniques in practical applications. We summarize three key techniques for modeling AD progression: multi-task model, time series model and deep learning. In particular, we discuss the basic structural elements of most representative multi-task learning algorithms, and analyze a multi-task disease prediction model based on longitudinal time. Lastly, some potential future research direction is given.

Temporal data clustering can provide underpinning techniques for the discovery of intrinsic structures, which proved important in condensing or summarizing information demanded in various fields of information sciences, ranging from time series analysis to sequential data understanding. In this paper, we propose a novel hidden Markov model (HMM)-based hybrid meta-clustering ensemble with bi-weighting scheme to solve the problems of initialization and model selection associated with temporal data clustering. To improve the performance of the ensemble techniques, the proposed bi-weighting scheme adaptively examines the partition process and hence optimizes the fusion of consensus functions. Specifically, three consensus functions are used to combine the input partitions, generated by HMM-based K -models under different initializations, into a robust consensus partition. An optimal consensus partition is then selected from the three candidates by a normalized mutual information-based objective function. Finally, the optimal consensus partition is further refined by the HMM-based agglomerative clustering algorithm in association with dendrogram-based similarity partitioning algorithm, leading to the advantage that the number of clusters can be automatically and adaptively determined. Extensive experiments on synthetic data, time series, and real-world motion trajectory datasets illustrate that our proposed approach outperforms all the selected benchmarks and hence providing promising potentials for developing improved clustering tools for information analysis and management.

As the rapid growth of social media technologies continues, Cyber-Physical-Social System (CPSS) has been a hot topic in many industrial applications. The use of “microblogging” services, such as Twitter, has rapidly become an influential way to share information. While recent studies have revealed that understanding and modelling microblog user behaviour with massive users’ data in social media are keen to success of many practical applications in CPSS, a key challenge in literatures is that diversity of geography and cultures in social media technologies strongly affect user behaviour and activity. The motivation of this article is to understand differences and similarities between microblogging users from different countries using social media technologies, and to attempt to design a Country-Level Micro-Blog User (CLMB) behaviour and activity model for supporting CPSS applications. We proposed a CLMB model for analysing microblogging user behaviour and their activity across different countries in the CPSS applications. The model has considered three important characteristics of user behaviour in microblogging data, including content of microblogging messages, user emotion index, and user relationship network. We evaluated CLBM model under the collected microblog dataset from 16 countries with the largest number of representative and active users in the world. Experimental results show that (1) for some countries with small population and strong cohesiveness, users pay more attention to social functionalities of microblogging service; (2) for some countries containing mostly large loose social groups, users use microblogging services as a news dissemination platform; (3) users in countries whose social network structure exhibits reciprocity rather than hierarchy will use more linguistic elements to express happiness in microblogging services.

The semi-supervised multi-tasking using least-squares support vector machine can further improve performance by using related information of related tasks, and it inherits the advantages of high training speed and high efficiency of the least square support vector machine. Standard support vector machine is based on supervised learning, and it is necessary to manually mark large amounts of data for obtaining sufficient training data, which is costly and inefficient. In this paper, we apply least squares support vector machine based on semi-supervised learning to the multi-tasks and propose a semi-supervised multi-tasking approach using least-squares support vector machine. Based on related tasks learning simultaneously, multi-task least-squares support vector machine is used to train both labeled and unlabeled samples, overcoming the limitation of slow training, and using the useful information among related tasks to improve the efficiency of all tasks. In the training process, the regional tagging and the tag reset methods are used to reduce the number of iterations to achieve convergence and increases the fault tolerance rate. The experiment on the actual dataset shows the effectiveness of the approach.

The TF-IDF algorithm is commonly used for text information retrieval and data mining. The traditional TF-IDF algorithm does not consider the domain characteristics of the article, and does not consider the distribution ratio. Currently, the solution proposed by many scholars only solves the problems of distribution ratio and the like, and does not solve the problem that the domain keywords have unreasonable weights. The problem has led to the use of domain-specific applications where relevant keywords in some areas have not been given appropriate weights. This paper proposes an improved method based on domain knowledge graph. This method will mainly consider the application of the legal field, and use the legal knowledge graph to make improvements to the TF-IDF algorithm, so as to achieve the reasonable weight assigned to the domain-related keywords in text feature extraction. Experiments show that this method can effectively improving the accuracy of the extraction.

Supervised learning algorithms are to discover the hidden patterns of the statistics, assuming that the training data and the test data are from the same distribution. There are two challenges in the traditional supervised machine learning. One is that the test data distribution always differs largely from the training data distribution in the real world, while another is that there is usually very few labeled data to train a machine learning model. In such cases, transfer learning, which emphasizes the transfer of the previous knowledge from different but related domains and tasks, is recommended to deal with these problems. Traditional transfer learning methods care more about the data itself rather than the task. In fact, there is no one universal feature representation can perfectly benefit the model training work. But different feature representations can discover some independent latent knowledge from the original data. In this paper, we propose an instance-based transfer learning method, which is a weighted ensemble transfer learning framework with multiple feature representations. In our work, mutual information is applied as the smart weighting schema to measure the weight of each feature representation. Extensive experiments have been conducted on three facial expression recognition data sets: JAFFE, KDEF and FERG-DB. The experimental results demonstrate that our approach achieves better performance than the traditional transfer learning method and the non-transfer learning method.

Driven by the automation technologies and health informatics of Industry 4.0, hospitals in China have deployed a complete automation system/platform for healthcare services accessing. Without much more Internet knowledge, elderlies usually seek the third-party to assist them to get healthcare services from Web or APPs, it consequently results in an unexpected situation that scalpers could grab all healthcare services booking by unrighteous means in order to resell to elderlies for a much higher price. Moreover, it is hard for physicians to identify the scalpers due to the complexity, ad-hoc, and multiscenario nature of healthcare processes. In this paper, a novel method is proposed for the identification and creation of user groups of scalpers in mobile healthcare services. The approach utilizes and extends state of the art data analysis approaches in the event-logs of the mobile system to identify user groups. Based on the user groups, user profiles are extracted by identifying representative eventcases from hierarchical user-event clusters. A comprehensive evaluation is conducted in a selected test-set from the event-logs of a mobile healthcare APP. The result shows its accuracy and effectiveness in scalper detection in mobile healthcare APP. Further, a complete case study is deployed in a real word hospital to ensure its utility, efficacy, and reliability.

As a fundamental task in automated video surveillance, person re-identification, which has received increasing attention in recent years, aims to match people across non-overlapping camera views in a multi-camera surveillance system. It has been reported that KISS metric learning has been followed by most of the previous supervised work because of its state of the art performance for person re-identification on VIPeR dataset. However, given only a small number of labeled image pairs available for training, the matching model certainly suffers from unstable learning process and poor matching result. To address this serious practical issue, we proposed a novel semi-supervised KISS metric learning (SS-KISS) approach which makes use of unlabeled data to improve the re-identification performance by 1) combining both global and local information to select the most confident image pairs from the unlabeled data; 2) using an ensemble approach, which explores advantages of supervised and unsupervised learning by reconciling two matching models on which labeled and un-labeled data to an optimal one via smart weighting schema. Extensive experiments have been conducted on three datasets: VIPeR, ETHZ, and i-LiDS, experimental results demonstrate that our approach achieves a sound performance in the case of small amount of labeled data.

In practice, unlabeled data can be cheaply and easily collected from target domain, but it is quite difficult and expensive to obtain a large amount of labeled data. Therefore how to use both of labeled and unlabeled data to improve the learning performance becomes critical issue for many real-world applications. Active Learning and Semi-supervised Learning are right solutions to such problem, and have been intensively studied from different perspectives. The former one advocates that learner is able to control the entire dataset and actively query the labels from the target dataset, the latter one tries to improve the learner's performance by using both of labeled and unlabeled instances at the same time. In this paper, we propose an Active Learning based SVM approach, KA-SVM. According to a cluster hypothesis, we use k-means to construct a pre-selection scheme, which obtains a subset of important instances as training set, then SVM can be optimally trained on such subset rather than entire one. Our approach has been generally evaluated on several benchmark datasets with comparison with other similar approaches, the experiment results demonstrate that our approach has the outstanding performance on both of classification accuracy and computation efficiency.

Ontology learning has become a popular research field recently. However, the typical ontology may not be sufficient to represent uncertainty information. Fuzzy ontology is proposed to solve the uncertainty reasoning problems. But the construction of fuzzy ontology is still a tedious and painstaking task. The cognitive model of fuzzy ontology learning is an automatic model of fuzzy ontology construction that simulates the process of human being recognizing the world. Induction is an important step of the model. In this paper, we present a strategy for the fuzzy ontology induction in the cognitive model of ontology learning which includes some generalization principles and the corresponding induction operators. It generates induction hypothesis through a series of operations. As a result, induction hypothesis are generated from the present ontology and it can generalize the existing ontology.

Among a number of ensemble learning techniques, boosting and bagging are the most popular sampling-based ensemble approaches for classification problems. Boosting is considered stronger than bagging on noise-free data set with complex class structures, whereas bagging is more robust than boosting in cases where noise data are present. In this paper, we extend both ensemble approaches to clustering tasks, and propose a novel hybrid sampling-based clustering ensemble by combining the strengths of boosting and bagging. In our approach, the input partitions are iteratively generated via a hybrid process inspired by both boosting and bagging. Then, a novel consensus function is proposed to encode the local and global cluster structure of input partitions into a single representation, and applies a single clustering algorithm to such representation to obtain the consolidated consensus partition. Our approach has been evaluated on 2-D-synthetic data, collection of benchmarks, and real-world facial recognition data sets, which show that the proposed technique outperforms the existing benchmarks for a variety of clustering tasks.

Temporal data clustering provides underpinning techniques for discovering the intrinsic structure and condensing information over temporal data. In this paper, we present a temporal data clustering framework via a weighted clustering ensemble of multiple partitions produced by initial clustering analysis on different temporal data representations. In our approach, we propose a novel weighted consensus function guided by clustering validation criteria to reconcile initial partitions to candidate consensus partitions from different perspectives, and then, introduce an agreement function to further reconcile those candidate consensus partitions to a final partition. As a result, the proposed weighted clustering ensemble algorithm provides an effective enabling technique for the joint use of different representations, which cuts the information loss in a single representation and exploits various information sources underlying temporal data. In addition, our approach tends to capture the intrinsic structure of a data set, e.g., the number of clusters. Our approach has been evaluated with benchmark time series, motion trajectory, and time-series data stream clustering tasks. Simulation results demonstrate that our approach yields favorite results for a variety of temporal data clustering tasks. As our weighted cluster ensemble algorithm can combine any input partitions to generate a clustering ensemble, we also investigate its limitation by formal analysis and empirical studies.

Time series clustering provides underpinning techniques for discovering the intrinsic structure and condensing/summarizing information conveyed in time series, which is demanded in various fields ranging from bioinformatics to video content understanding. In this paper, we present an unsupervised ensemble learning approach to time series clustering by combining rival-penalized competitive learning (RPCL) networks with different representations of time series. In our approach, the RPCL network ensemble is employed for clustering analyses based on different representations of time series whenever available, and an optimal selection function is applied to find out a final consensus partition from multiple partition candidates yielded by applying various consensus functions for the combination of competitive learning results. As a result, our approach first exploits its capability of the RPCL rule in clustering analysis of automatic model selection on individual representations and subsequently applies ensemble learning for the synergy of reconciling diverse partitions resulted from the use of different representations and augmenting RPCL networks in automatic model selection and overcoming its inherent limitation. Our approach has been evaluated on 16 benchmark time series data mining tasks with comparison to state-of-the-art time series clustering techniques. Simulation results demonstrate that our approach yields favorite results in clustering analysis of automatic model selection.

Unsupervised classification or clustering is an important data analysis technique demanded in various fields including machine learning, data mining, pattern recognition, image analysis and bioinformatics. Recently a large number of studies have attempted to improve clustering by combing multiple clustering solutions into a single consolidated clustering ensemble that has the best performance among given clustering solutions. However, the different clustering ensembles have their own behaviors on data of various characteristics. In this paper, we propose a novel approach to data clustering by constructing a clustering ensemble iteratively based on partitions generated on training subsets sampled from the original dataset. To yield a robust clustering ensemble our approach employs a hybrid sampling scheme inspired by both boosting and bagging techniques originally proposed for supervised learning. Our approach has been evaluated on synthetic data and real-world motion trajectory data sets, and experimental results demonstrate that it yields satisfactory performance for a variety of clustering tasks.

Temporal data clustering provides useful techniques for condensing and summarizing information conveyed in temporal data, which is demanded in various fields ranging from time series analysis to sequential data understanding. In this paper, we propose a novel approach to temporal data clustering by an ensemble of competitive learning networks incorporated by different representations of temporal data. In our approach, competitive learning networks of the rival-penalized learning mechanism are employed for clustering analyses based on different temporal data representations while an optimal selection function is applied to find out a final consensus partition from multiple partition candidates yielded by applying alternative consensus functions to results of competitive learning on different representations. Thanks to its capability of the rival penalized learning rules in automatic model selection and the synergy of fusing diverse partitions on different representations, our ensemble approach yields favorite results, which has been demonstrated in time series and motion trajectory clustering tasks.