Due to the scarcity of annotated real-world data for specific categories, audio-visual generalized zero-shot learning (GZSL) has attracted significant attention. GZSL aims to classify novel classes absent during training while ensuring stable performance on seen classes. However, most existing methods operate implicitly, often neglecting the effective utilization of temporal, spatial, and semantic consistency. To address these challenges, we propose the Temporal Spatial Semantic Fusion network (TSSF). Specifically, we explore both audio and visual modalities using a multi-branch, multi-grained structure comprising a temporal global extraction module, a spatial local refinement module, and a multi-grained fusion module. The temporal global extraction module employs a Transformer-based Spiking Neural Network to extract explicit temporal representations and capture global dependencies. Simultaneously, the spatial local refinement module focuses on spatial information and local details using a window attention mechanism. Furthermore, the temporal and spatial features are hierarchically fused in the multi-grained fusion module, which incorporates both temporal and spatial attention for semantic enrichment. To explore multi-modal interactions, we enhance audio and visual features through cross-modal attention, followed by multi-modal alignment with text embeddings. Experiments on three benchmark audio-visual datasets validate the superiority of our method over state-of-the-art approaches. Notably, TSSF achieves significant improvements of 30.34% and 6.96% in HM and ZSL metrics on the VGG-GZSL dataset.

Graph neural networks (GNNs) are prominent for their effectiveness in processing graph-structured data for semi-supervised node classification tasks. Most existing GNNs perform message passing directly based on the observed graph structure. However, in real-world scenarios, the observed structure is often suboptimal due to multiple factors, significantly degrading the performance of GNNs. To address this challenge, we first conduct an empirical analysis showing that different graph structures significantly impact empirical risk and classification performance. Motivated by our observations, we propose a novel method named Trade-off Graph Structure Learning (TGSL), guided by the multifaceted Graph Information Bottleneck (GIB) principle based on Mutual Information (MI). The key idea behind TGSL is to learn a minimal sufficient graph structure that minimizes empirical risk while maintaining performance. Specifically, we introduce global feature augmentation to capture the structural roles of nodes, and global structure augmentation to uncover global relationships between nodes. The augmented graphs are then processed by structure estimators with different parameters for refinement and redefinition, respectively. Additionally, we innovatively leverage multifaceted GIB as the optimization objective by maximizing the MI between the labels and the representation derived from the final structure, while constraining the MI between this representation and that based on the redefined structures. This trade-off helps avoid capturing irrelevant information from the redefined structures and enhances the final representation for node classification. We conduct extensive experiments across a range of datasets under clean and attacked conditions. The results demonstrate the outstanding performance and robustness of TGSL over state-of-the-art baselines.

The challenge of transforming partial point clouds into complete meshes still persists, with current methods facing issues like data accessibility constraint, shape preservation failure and poor robustness on real-scan data. Drawing inspiration from the structural information of objects to enhance the completion, we introduce an innovative weakly-supervised shape completion method leveraging structural decomposition without the necessity of SDFs during training. By representing objects as abstract structural frameworks and part details, our method initiates by forecasting the structure of the input partial point clouds, and individually restore each component through part decomposition completion and generation. Extracted part details are represented in images, which are porous and incomplete. Hence, we utilize a completion network to complete such details. For multiple results generation, a diffusion-based generation network is employed to generate a variety of details for the missing areas. The predicted structure and details are subsequently converted back into meshes, yielding the complete results. Since the details are depicted in images, our approach eliminates the need for SDFs during the training phase, achieving weakly-supervision. We conduct extensive comparisons on both artificial and real-scan datasets, demonstrating an average improvement of over 38.1% compared to the prior method, and achieving SOTA performance.

Learning with noisy labels (LNL) methods have enabled the deployment of machine learning systems with imperfectly labeled data. However, these methods often struggle to identify noise in the presence of long-tailed (LT) class distributions, where the memorization effect becomes class-dependent. Conversely, LT methods are suboptimal under label noise, as it hinders access to accurate label frequency statistics. This study aims to address the long-tailed noisy data by bridging the methodological gap between LNL and LT approaches. We propose a direct solution, termed Robust Logit Adjustment, which estimates ground-truth labels through label refurbishment, thereby mitigating the impact of label noise. Simultaneously, our method incorporates the distribution of training-time corrected target labels into the LT method logit adjustment, providing class-rebalanced supervision. Extensive experiments on both synthetic and real-world long-tailed noisy datasets demonstrate the superior performance of our method.

Graph Neural Networks (GNNs) have demonstrated remarkable ability in semi-supervised node classification. However, most existing GNNs rely heavily on a large amount of labeled data for training, which is labor-intensive and requires extensive domain knowledge. In this paper, we first analyze the restrictions of GNNs generalization from the perspective of supervision signals in the context of few-shot semi-supervised node classification. To address these challenges, we propose a novel algorithm named NormProp, which utilizes the homophily assumption of unlabeled nodes to generate additional supervision signals, thereby enhancing the generalization against label scarcity. The key idea is to efficiently capture both the class information and the consistency of aggregation during message passing, via decoupling the direction and Euclidean norm of node representations. Moreover, we conduct a theoretical analysis to determine the upper bound of Euclidean norm, and then propose homophilous regularization to constraint the consistency of unlabeled nodes. Extensive experiments demonstrate that NormProp achieve state-of-the-art performance under low-label rate scenarios with low computational complexity.

Current learning-based methods predict NeRF or 3D Gaussians from point clouds to achieve photo-realistic rendering but still depend on categorical priors, dense point clouds, or additional refinements. Hence, we introduce a novel point cloud rendering method by predicting 2D Gaussians from point clouds. Our method incorporates two identical modules with an entire-patch architecture enabling the network to be generalized to multiple datasets. The module normalizes and initializes the Gaussians utilizing the point cloud information including normals, colors and distances. Then, splitting decoders are employed to refine the initial Gaussians by duplicating them and predicting more accurate results, making our methodology effectively accommodate sparse point clouds as well. Once trained, our approach exhibits direct generalization to point clouds across different categories. The predicted Gaussians are employed directly for rendering without additional refinement on the rendered images, retaining the benefits of 2D Gaussians. We conduct extensive experiments on various datasets, and the results demonstrate the superiority and generalization of our method, which achieves SOTA performance. The code is available at https://github.com/murcherful/GauPCRender.

Paintings possess profound cultural and historical backgrounds. Unlike real-life images, they convey complex semantics beyond simple visual features. This diversity and complexity make painting style classification highly challenging, and many popular visual models struggle with it. To address this issue, we propose an art knowledge-driven framework(AKDF) to improve models’ comprehension of art knowledge. AKDF utilizes multimodal models and prompts to extract style-related textual descriptions from images. Text and image features will be fused in enhanced bilinear pooling module, thus integrating art knowledge into the output features. Additionally, we design a contrastive learning auxiliary task based on label embeddings to introduce further art knowledge about style labels. Besides, AKDF incorporates texture features and adds another genre classification auxiliary task to provide more painting information. We constructs two datasets based on WikiArt due to its comprehensive and challenging nature. The extensive experiment results demonstrate the superiority of the AKDF, which effectively develops the performance of various models by over three percentage points across two datasets.

Recent advancements in Large Language Models (LLMs) have catalyzed the exploration of Chain of Thought (CoT) approaches, particularly in extending their application to multimodal tasks to enhance reasoning capabilities. However, current studies often fail to fully leverage the inferential capabilities of these models, as they primarily focus on selecting similar questions or images rather than exploring identical inferences, thereby limiting the potential for analogical learning. In this paper, we propose an inference retrieval-augmented method that incorporates two strategies to identify training set examples with inferential processes similar to those of the target problem. Initially, we fine-tune a vanilla model to generate the pseudo-inference relevant to the question, which is then utilized to retrieve analogous examples. Next, we hypothesize that questions sharing similar contexts may share inferences, thus retrieving examples based on similarity in contextual questions. Ultimately, we construct multiple reasoning pathways from the retrieved examples and employ a voting mechanism to determine the most frequent answer. Our method surpasses all few-shot approaches and most supervised methods on the ScienceQA dataset, achieving an accuracy of 87.37% with ChatGPT and exceeding the human benchmark in several categories.

Experience Replay (ER) allows Deep Reinforcement Learning (RL) agent to reuse past experience, as though recall the same Experience repeatedly. ER enables RL algorithm to be trained by reusing previous states, so that RL agent can obtain more accurate value estimations and action selections. Current policy algorithms either have a rule-based replay policy or uniformly replay past experiences, which may be suboptimal. The agent is updated based on the replay data to maximize the cumulative reward, and the replay policy is updated to provide the more valuable experience for the agent. In this work, we propose a novel experience replay algorithm Regret Minimization Experience Replay (RMER), to improve the immediate reward via sampling and ensure certain exploration capability of agent. Finally, we prove the ascendency of RMER with different off- policy algorithms on the suite of Open AI gym continuous control tasks.

Understanding emotions in art paintings and generating comments about emotions is a highly challenging task due to their rich semantics and complex expressions. However, existing related methods tend to neglect the critical role of artistic emotion features in artworks at the visual level, while general vision language models(VLMs) lack a comprehensive grasp of art domain knowledge. To address this, we propose a knowledge-driven framework based on artistic emotion knowledge, which helps VLMs better comprehend emotions in artworks. The framework primarily comprises the artistic-emotion visual tower and the ANP cross-attention module. Specifically, the artistic-emotion visual tower introduces additional emotion tokens before visual features are fed into transformer layers. It integrate low-level art features into emotion tokens and infuses high-level art features into emotion tokens in the penultimate transformer layer. Moreover, the attention mechanism of visual tower is optimized by emotion bias, which enhances the model’s sensitivity to art features related to emotions. The ANP cross-attention module will extract affective entities directly from paintings as adjective-noun pairs and apply cross-attention with visual tower outputs. We have conducted extensive experiments on Artemis 1.0 and Artemis 2.0 datasets. The results demonstrate that our framework effectively improves the performance of VLMs in the task of understanding artistic emotions on eight metrics, outperforming existing methods.

Pretrained multimodal fusion models often suffer significant performance drops when faced with missing data due to sensor failures, transmission loss, or privacy constraints in real-world scenarios. Their complex architectures and large size render direct adaptation infeasible, positioning input-level imputation as a practical alternative. However, existing methods typically use fixed network structures and a single shared representation for generating different missing modalities, ignoring the dynamic and differentiated inter-modal correlations under varying missing scenarios. To address these deficiencies, we proposes a type-induced missing-aware mixture-of-experts architecture. The model dynamically routes observed modality features to specialized experts based on missing patterns and feature distributions, enabling adaptive inter-modal correlation extraction. It also employs learnable type-specific query vectors and a cross-attention mechanism to generate tailored representations for each modality, preserving crucial modality-specific information for reconstruction. Extensive experiments confirm the superior performance of the proposed approach in imputing missing data across diverse scenarios.

Confidence estimation for generative large language models (LLMs) is critical for assessing the reliability of their step-by-step reasoning. While existing methods aggregate token probabilities uniformly or use semantic weighting, they often overlook two key factors: critical low-probability tokens within individual reasoning steps and inconsistencies in logical transitions between consecutive steps. To address these limitations, we introduce Step-Aware Confidence Estimation (SACE), which separately analyzes each reasoning step, capturing both intra-step token distribution patterns and inter-step logical transitions. These features are then integrated through a lightweight model to predict confidence scores. Our experiments on LLMs with different parameter sizes and various reasoning datasets demonstrate that SACE significantly outperforms the existing methods, achieving improvements of up to 8% AUROC score. Moreover, SACE requires only a single generation and shows strong transferability across datasets and model architectures, highlighting its practicality for real-world deployment.

Real-world machine learning applications seldom provide perfect labeled data, posing a challenge in developing models robust to noisy labels. Recent methods prioritize noise filtering based on the discrepancies between model predictions and the provided noisy labels, assuming samples with minimal classification losses to be clean. In this work, we capitalize on the consistency between the learned model and the complete noisy dataset, employing the data’s rich representational and topological information. We introduce LaplaceConfidence, a method that to obtain label confidence (i.e., clean probabilities) utilizing the Laplacian energy. Specifically, it first constructs graphs based on the feature representations of all noisy samples and minimizes the Laplacian energy to produce a low-energy graph. Clean labels should fit well into the low-energy graph while noisy ones should not, allowing our method to determine data’s clean probabilities. Furthermore, LaplaceConfidence is embedded into a holistic method for robust training, where co-training technique generates unbiased label confidence and label refurbishment technique better utilizes it. We also explore the dimensionality reduction technique to accommodate our method on large-scale noisy datasets. Our experiments demonstrate that LaplaceConfidence outperforms state-of-the-art methods on benchmark datasets under both synthetic and real-world noise. Code available at https://github.com/chenmc1996/LaplaceConfidence .

ABSTRACT The visual dialogue task requires computers to comprehend image content and preceding question‐and‐answer history to accurately answer related questions, with each round of dialogue providing the necessary historical context for subsequent interactions. Existing research typically processes multiple questions related to a single image as independent samples, which results in redundant modelling of the images and their captions and substantially increases computational costs. To address the challenges above, we introduce a fast transformer for visual dialogue, termed FastVDT, which utilises novel attention masks and continuous positional encoding. FastVDT models multiple image‐related questions as an integrated entity, accurately processing prior conversation history in each dialogue round while predicting answers to multiple questions. Our method effectively captures the interrelations among questions and significantly reduces computational overhead. Experimental results demonstrate that our method delivers outstanding performance on the VisDial v0.9 and v1.0 datasets. FastVDT achieves comparable performance to VD‐BERT and VU‐BERT while reducing computational costs by 80% and 56%, respectively.

Structural representation is crucial for reconstructing and generating editable 3D shapes with part semantics. Recent 3D shape generation works employ complicated networks and structure definitions relying on hierarchical annotations and pay less attention to the details inside parts. In this paper, we propose the method that parameterizes the shared structure in the same category using a differentiable template and corresponding fixed-length parameters. Specific parameters are fed into the template to calculate cuboids that indicate a concrete shape. We utilize the boundaries of three-view renderings of each cuboid to further describe the inside details. Shapes are represented with the parameters and three-view details inside cuboids, from which the SDF can be calculated to recover the object. Benefiting from our fixed-length parameters and three-view details, our networks for reconstruction and generation are simple and effective to learn the latent space. Our method can reconstruct or generate diverse shapes with complicated details, and interpolate them smoothly. Extensive evaluations demonstrate the superiority of our method on reconstruction from point cloud, generation, and interpolation.

Detecting out-of-distribution (OOD) inputs is critical for reliable machine learning, but deep neural networks often make overconfident predictions, even for OOD inputs that deviate from the distribution of training data. Prior methods relied on the widely used softmax cross-entropy (CE) loss that is adequate for classifying in-distribution (ID) samples but not optimally designed for OOD detection. To address this issue, we propose CASE, a simple and effective OOD detection method by explicitly improving intra-class Compactness And inter-class Separability of feature Embeddings. To enhance the separation between ID and OOD samples, CASE uses a dual-loss framework, which includes a separability loss that maximizes the inter-class Euclidean distance to promote separability among different class centers, along with a compactness loss that minimizes the intra-class Euclidean distance to encourage samples to be close to their class centers. In particular, the class centers are defined as a free optimization parameter of the model and updated by gradient descent, which is simple and further enhances the OOD detection performance. Extensive experiments demonstrate the superiority of CASE, which reduces the average FPR95 by 37.11% and improves the average AUROC by 15.89% compared to the baseline method using a softmax confidence score on the more challenging CIFAR-100 model.

Federated learning (FL) provides a privacy-preserving approach for collaborative training of machine learning models. Given the potential data heterogeneity, it is crucial to select appropriate collaborators for each FL participant (FL-PT) based on data complementarity. Recent studies have addressed this challenge. Similarly, it is imperative to consider the inter-individual relationships among FL-PTs where some FL-PTs engage in competition. Although FL literature has acknowledged the significance of this scenario, practical methods for establishing FL ecosystems remain largely unexplored. In this paper, we extend a principle from the balance theory, namely “the friend of my enemy is my enemy”, to ensure the absence of conflicting interests within an FL ecosystem. The extended principle and the resulting problem are formulated via graph theory and integer linear programming. A polynomial-time algorithm is proposed to determine the collaborators of each FL-PT. The solution guarantees high scalability, allowing even competing FL-PTs to smoothly join the ecosystem without conflict of interest. The proposed framework jointly considers competition and data heterogeneity. Extensive experiments on real-world and synthetic data demonstrate its efficacy compared to five alternative approaches, and its ability to establish efficient collaboration networks among FL-PTs.

As Large Language Models (LLMs) become increasingly influential in reasoning tasks, ensuring their trustworthiness and introspective self-awareness is critical. This research introduces the Think-Solve-Verify (TSV) framework, an innovative strategy tailored to explore LLMs’ trustworthiness, introspective self-awareness, and collaborative reasoning. This method accentuates a model’s capability to construct introspective reasoning processes from answers and ensure their trustworthiness. The reasoning with TSV consistently performs at or near the top across the majority of datasets with a single interaction with LLM. Moreover, we refine the voting process of self-consistency within the Chain-of-Thought (CoT) approach, leading to notable accuracy enhancements. In our evaluations, this approach improved performance from 67.3% to 72.8% on the AQuA dataset. Furthermore, we delve into the model’s ability to explain the given answers, highlighting the significance of discerning genuine comprehension from mere guesswork.

Graph neural networks (GNNs) have proven highly effective in handling graph-structured data. However, most existing GNNs rely on the homophily assumption, hindering their performance on heterophilic graphs. This limitation is partially due to aggregation containing irrelevant nodes. In this work, we propose a novel GNN model based on node correlation called NoC-GNN, to address the deficiencies of existing techniques. NoC-GNN retains relevant nodes while removing irrelevant nodes, enhancing the effectiveness of nodes in the mixed state. NoC-GNN first constructs a new graph structure based on the k-nearest neighbor (kNN) graph to aggregate relevant nodes, and then constructs a matrix based on the new graph structure to remove possibly irrelevant nodes. Finally, the attention mechanism is used to adaptively integrate relevant, irrelevant, and self-information to model both homophilic and heterophilic graphs. Experimental results demonstrate that NoC-GNN achieves superior performance across a wide range of semi-supervised node classification tasks.

In many real-world applications of Deep Neural Networks (DNNs) in visual recognition, data augmentation stands out as a premier tool for enhancing model robustness. Stemming from the understanding of the common mechanisms of data augmentation methods, we introduce the mask-based "data augmentation boost" (DaBoost) method, a strategic approach that exploits the control of game interaction strength. Our empirical results are telling: DaBoost not only consistently surpasses the state-of-the-art PixMix method but also achieves impressive robustness metrics, with a vanilla WideResNet registering a mere 6.5% mCE and a 2.3% RMS calibration error on CIFAR-10 data. An intriguing observation from our study is the Long-Rope Effect. We discerned that penalizing high-order interactions inadvertently leads to a boost in mid-order interactions, mirroring patterns inherent to human cognitive processes. This interplay hints at the potential avenues for optimizing DNNs’ performance further.

In the realm of visual recognition, data augmentation stands out as a pivotal technique to amplify model robustness. Yet, a considerable number of existing methodologies lean heavily on heuristic foundations, rendering their intrinsic mechanisms ambiguous. This manuscript takes both a theoretical and empirical approach to understanding the phenomenon. Theoretically, we frame the discourse around data augmentation within game theory’s constructs. Venturing deeper, our empirical evaluations dissect the intricate mechanisms of emblematic data augmentation strategies, illuminating that these techniques primarily stimulate mid- and high-order game interactions. Beyond the foundational exploration, our experiments span multiple datasets and diverse augmentation techniques, underscoring the universal applicability of our findings. Recognizing the vast array of robustness metrics with intricate correlations, we unveil a streamlined proxy. This proxy not only simplifies robustness assessment but also offers invaluable insights, shedding light on the inherent dynamics of model game interactions and their relation to overarching system robustness. These insights provide a novel lens through which we can re-evaluate model safety and robustness in visual recognition tasks.

Visual dialog involves answering questions by analyzing both images and dialogue history. While current multimodal research has effectively modeled the interactions among images, dialogue history, and questions, it incurs significant computational overhead and complexity. To address these challenges, this paper introduces a MultiModal Star Transformer (MMST) that effectively models the interactions between visual and textual modalities, as well as within each modality, with linear computational overhead. MMST utilizes a relay token for each modality, allowing each satellite token to interact with its two adjacent tokens, its previous state, and the two relay tokens. The introduction of relay tokens ensures that every two non-adjacent satellite tokens are two-hop neighbors, thus enabling MMST to support both intramodal long-range connections and intermodal interactions efficiently. Experimental results on the Visdial v0.9 and v1.0 datasets demonstrate that MMST performs comparably to full-attention models.

Modern neural networks are known to give overconfident predictions for out-of-distribution inputs when deployed in the open world. It is common practice to leverage a surrogate outlier dataset to regularize the model during training, and recent studies emphasize the role of uncertainty in designing the sampling strategy for outlier datasets. However, the OOD samples selected solely based on predictive uncertainty can be biased towards certain types, which may fail to capture the full outlier distribution. In this work, we empirically show that diversity is critical in sampling outliers for OOD detection performance. Motivated by the observation, we propose a straightforward and novel sampling strategy named DOS (Diverse Outlier Sampling) to select diverse and informative outliers. Specifically, we cluster the normalized features at each iteration, and the most informative outlier from each cluster is selected for model training with absent category loss. With DOS, the sampled outliers efficiently shape a globally compact decision boundary between ID and OOD data. Extensive experiments demonstrate the superiority of DOS, reducing the average FPR95 by up to 25.79% on CIFAR-100 with TI-300K.

Multi-modal multi-label emotion recognition has gained significant attention in the field of affective computing, enabling various signals to distinguish complex emotions accurately. However, previous studies primarily focus on capturing invariant representations, neglecting the importance of incorporating the fluctuation of temporal information which affects the model robustness. In this paper, we propose a novel Spiking Multi-modal Interactive Label-guided Enhancement network (SMILE). It introduces the spiking neural network with dynamic thresholds, allowing flexible processing of temporal information to enhance the model robustness. Furthermore, it employs the scale spiking fusion to enrich semantic information. In addition to modality-specific refinement, SMILE integrates the modality-interactive exploration and label-modality matching modules to capture multimodal interaction and label-modality dependence. Experimental results on benchmark datasets CMU-MOSEI and NEMu demonstrate the superiority of SMILE over state-ofthe-art models. Notably, SMILE achieves a significant 28.5% improvement in accuracy compared to the benchmark method when evaluated on NEMu dataset.

Graph Neural Networks have achieved remarkable accuracy in semi-supervised node classification tasks. However, these results lack reliable uncertainty estimates. Conformal prediction methods provide a theoretical guarantee for node classification tasks, ensuring that the conformal prediction set contains the ground-truth label with a desired probability (e.g., 95\%). In this paper, we empirically show that for each node, aggregating the non-conformity scores of nodes with the same label can improve the efficiency of conformal prediction sets while maintaining valid marginal coverage. This observation motivates us to propose a novel algorithm named $\textit{Similarity-Navigated Adaptive Prediction Sets}$ (SNAPS), which aggregates the non-conformity scores based on feature similarity and structural neighborhood. The key idea behind SNAPS is that nodes with high feature similarity or direct connections tend to have the same label. By incorporating adaptive similar nodes information, SNAPS can generate compact prediction sets and increase the singleton hit ratio (correct prediction sets of size one). Moreover, we theoretically provide a finite-sample coverage guarantee of SNAPS. Extensive experiments demonstrate the superiority of SNAPS, improving the efficiency of prediction sets and singleton hit ratio while maintaining valid coverage.

Co-training, an advanced form of self-training, allows multiple base models to learn collaboratively, leading to superior performance in semi-supervised learning tasks. However, its widespread adoption is hindered by high computational costs and intricate design choices. To address these challenges, we present Multi-Head Co-Training, a streamlined and efficient framework that consolidates individual models into a multi-head structure, adding minimal extra parameters. Each classification head in this unified model collaborates with others via a “Weak and Strong Augmentation” strategy, with diversity organically introduced through robust data augmentation . Consequently, our approach implicitly promotes diversity while incurring only a minor increase in computational overhead, making co-training more accessible. We validate the effectiveness of Multi-Head Co-Training through an empirical study on standard semi-supervised learning benchmarks. For example, our method achieves up to a 3.1% accuracy improvement on the semi-supervised CIFAR dataset compared to recent methods. Recognizing the necessity for more practical performance metrics beyond accuracy, we assess our framework from three additional perspectives: robust generalization, uncertainty, and computational efficiency. To evaluate robust generalization, we expand the conventional SSL experimental setting to a more comprehensive open-set semi-supervised learning scenario. For uncertainty assessment, we conduct experiments on model calibration and selective classification benchmarks. For example, our method achieves up to a 4.3% accuracy improvement on the open-set semi-supervised CIFAR dataset. Our extensive experiments confirm that our proposed framework better captures prediction confidence and uncertainty, rendering it more suitable for SSL deployment in open environments. The code is available at https://github.com/chenmc1996/Multi-Head-Co-Training .

Outliers will inevitably creep into the captured point cloud during 3D scanning, degrading cutting-edge models on various geometric tasks heavily. This paper looks at an intriguing question that whether point cloud completion and segmentation can promote each other to defeat outliers. To answer it, we propose a collaborative completion and segmentation network, termed CS-Net, for partial point clouds with outliers. Unlike most of existing methods, CS-Net does not need any clean (or say outlier-free) point cloud as input or any outlier removal operation. CS-Net is a new learning paradigm that makes completion and segmentation networks work collaboratively. With a cascaded architecture, our method refines the prediction progressively. Specifically, after the segmentation network, a cleaner point cloud is fed into the completion network. We design a novel completion network which harnesses the labels obtained by segmentation together with farthest point sampling to purify the point cloud and leverages KNN-grouping for better generation. Benefited from segmentation, the completion module can utilize the filtered point cloud which is cleaner for completion. Meanwhile, the segmentation module is able to distinguish outliers from target objects more accurately with the help of the clean and complete shape inferred by completion. Besides the designed collaborative mechanism of CS-Net, we establish a benchmark dataset of partial point clouds with outliers. Extensive experiments show clear improvements of our CS-Net over its competitors, in terms of outlier robustness and completion accuracy.

Noisy labels damage the performance of deep networks. For robust learning, a prominent two-stage pipeline alternates between eliminating possible incorrect labels and semi-supervised training. However, discarding part of noisy labels could result in a loss of information, especially when the corruption has a dependency on data, e.g., class-dependent or instance-dependent. Moreover, from the training dynamics of a representative two-stage method DivideMix, we identify the domination of confirmation bias: pseudo-labels fail to correct a considerable amount of noisy labels, and consequently, the errors accumulate. To sufficiently exploit information from noisy labels and mitigate wrong corrections, we propose Robust Label Refurbishment (Robust LR)—a new hybrid method that integrates pseudo-labeling and confidence estimation techniques to refurbish noisy labels. We show that our method successfully alleviates the damage of both label noise and confirmation bias. As a result, it achieves state-of-the-art performance across datasets and noise types, namely CIFAR under different levels of synthetic noise and mini-WebVision and ANIMAL-10N with real-world noise.

Detecting out-of-distribution (OOD) samples is crucial to the safe deployment of a classifier in the real world. However, deep neural networks are known to be overconfident for abnormal data. Existing works directly design score function by mining the inconsistency from classifier for in-distribution (ID) and OOD. In this paper, we further complement this inconsistency with reconstruction error, based on the assumption that an autoencoder trained on ID data cannot reconstruct OOD as well as ID. We propose a novel method, READ (Reconstruction Error Aggregated Detector), to unify inconsistencies from classifier and autoencoder. Specifically, the reconstruction error of raw pixels is transformed to latent space of classifier. We show that the transformed reconstruction error bridges the semantic gap and inherits detection performance from the original. Moreover, we propose an adjustment strategy to alleviate the overconfidence problem of autoencoder according to a fine-grained characterization of OOD data. Under two scenarios of pre-training and retraining, we respectively present two variants of our method, namely READ-MD (Mahalanobis Distance) only based on pre-trained classifier and READ-ED (Euclidean Distance) which retrains the classifier. Our methods do not require access to test time OOD data for fine-tuning hyperparameters. Finally, we demonstrate the effectiveness of the proposed methods through extensive comparisons with state-of-the-art OOD detection algorithms. On a CIFAR-10 pre-trained WideResNet, our method reduces the average FPR@95TPR by up to 9.8% compared with previous state-of-the-art.

Unsupervised completion of real scene objects is of vital importance but still remains extremely challenging in preserving input shapes, predicting accurate results, and adapting to multi-category data. To solve these problems, we propose in this paper an Unsupervised Symmetric Shape-Preserving Autoencoding Network, termed USSPA, to predict complete point clouds of objects from real scenes. One of our main observations is that many natural and manmade objects exhibit significant symmetries. To accommodate this, we devise a symmetry learning module to learn from those objects and to preserve structural symmetries. Starting from an initial coarse predictor, our autoencoder refines the complete shape with a carefully designed upsampling refinement module. Besides the discriminative process on the latent space, the discriminators of our USSPA also take predicted point clouds as direct guidance, enabling more detailed shape prediction. Clearly different from previous methods which train each category separately, our USSPA can be adapted to the training of multi-category data in one pass through a classifier-guided discriminator, with consistent performance on single category. For more accurate evaluation, we contribute to the community a real scene dataset with paired CAD models as ground truth. Extensive experiments and comparisons demonstrate our superiority and generalization and show that our method achieves state-of-the-art performance on unsupervised completion of real scene objects.

Heterogeneous Information Networks (HINs) are prevalent in real-world systems. Recent advances in network embedding provide an effective way of encoding HINs into low-dimensional vectors. However, there is a growing concern that existing HIN embedding algorithms may suffer from the problem of generating biased representations, resulting in discrimination against certain demographic groups. In this paper, we propose a flexible debiasing framework for fair HIN embedding to address this issue. Specifically, we first formalize measurements and the definition of fairness in HIN embedding. Then, we propose a debiasing framework named FairHGNN, including a novel meta-path sampling method that focuses on mitigating the bias in random walks, and a fairness constraint with Wasserstein distance to alleviate the algorithmic bias in Graph Neural Networks (GNNs). Experimental results on real-world datasets validate the efficacy of FairHGNN in promoting fairness and maintaining good utility.

Core-selecting combinatorial auctions (CAs) restrict the auction result in the core such that no coalitions could improve their utilities by engaging in collusion. The minimum-revenue-core (MRC) rule is a widely used core-selecting payment rule to maximize the total utilities of all bidders. However, the MRC rule can suffer from severe unfairness since it ignores individuals' utilities. To address this limitation, we propose to explore the leximin principle to achieve fairness in core-selecting CAs since the leximin principle prefers to maximize the utility of the worst-off; the resulting bidder-leximin-optimal (BLO) payment rule is then theoretically analyzed and an effective algorithm is further provided to compute the BLO outcome. Moreover, we conduct extensive experiments to show that our algorithm returns fairer utility distributions and is faster than existing algorithms of core-selecting payment rules.

Out-of-distribution (OOD) detection is critical for ensuring the safe deployment of machine learning models in the open world. Due to the simplicity and intuitiveness of distance- based methods, i.e., samples are detected as OOD if they are relatively far away from the centroids or prototypes of in-distribution (ID) classes, they have attracted widespread attention from researchers in the field of OOD detection. However, prior OOD detection methods directly take off-the- shelf loss functions, like widely used softmax cross-entropy (CE) loss, that suffices for classifying ID samples, but is not optimally designed for OOD detection. In this work, we propose CESED, an improved CE loss applied to the scalable Squared Euclidean Distance vector, which exploits hyper- spherical evenly-distributed class centroids for OOD detection. CESED can promote strong ID-OOD separability because it explicitly encourages maximization of inter-class distances and minimization of intra-class distances. Extensive experiments demonstrate that CESED achieves superior detection performance on a comprehensive suite of benchmark datasets. For the more challenging case where CIFAR-100 is used as ID, our method achieves a 31.98% reduction in average FPR95 and 6.20% reduction in ID test error compared to the baseline method using a softmax confidence score.

Abstract Short text classification has been a fundamental task in natural language processing, which benefits various applications, such as sentiment analysis, news tagging, and intent recommendation. However, classifying short texts is challenging due to the information sparsity in the text corpus. Besides, the performance of existing machine learning classification models largely relies on sufficient training data, yet labels can be scarce and expensive to obtain in real‐world text classification scenarios. In this article, we propose a novel self‐supervised short text classification method. Specifically, we first model the short text corpus as a heterogeneous graph to address the information sparsity problem. Then, we introduce a self‐attention‐based heterogeneous graph neural network model to learn short text embeddings. In addition, we adopt a self‐supervised learning framework to exploit internal and external similarities among short texts. Experiments on five real‐world short text benchmarks validate the effectiveness of our proposed method compared with the state‐of‐the‐art methods.

Objective. The data processing of medical test report has always been one of the important contents in biological information domain, especially the process of extracting the effective information from the report so as to assist doctors with the correct medical plan. Usual methods neglect the implicit relationship between features. More features are generally not a better choice because more noise is generated between feature combinations. We propose a practical feature selection strategy RMFS, which aims to select the optimal combination of features. Materials and Methods. Based on the above situation, in this paper, 64 features are extracted from a real medical test report dataset for stroke and feature selection is defined as a reinforcement learning problem to optimize the feature combination by minimizing regret. We select three current mainstream feature selection methods and conduct comparative experiments. Results. We processed and completed a dataset derived from real medical test reports of stroke. We redefine the feature selection problem as a reinforcement learning problem and propose an optimization strategy based on regret minimization and train weight parameters in a DQN network. Experimental results demonstrate that our method can identify feature combinations with higher prediction accuracy. Discussion. RMFS shows a strong robustness to the randomness of the environment and has high computational efficiency and accuracy. Compared with the previous feature selection methods, our method yields superior results. Conclusion. The experimental results demonstrate that our method can obtain a more accurate prediction rate under the same feature scale and we can achieve baseline performance with fewer features.

Co-training, extended from self-training, is one of the frameworks for semi-supervised learning. Without natural split of features, single-view co-training works at the cost of training extra classifiers, where the algorithm should be delicately designed to prevent individual classifiers from collapsing into each other. To remove these obstacles which deter the adoption of single-view co-training, we present a simple and efficient algorithm Multi-Head Co-Training. By integrating base learners into a multi-head structure, the model is in a minimal amount of extra parameters. Every classification head in the unified model interacts with its peers through a “Weak and Strong Augmentation” strategy, in which the diversity is naturally brought by the strong data augmentation. Therefore, the proposed method facilitates single-view co-training by 1). promoting diversity implicitly and 2). only requiring a small extra computational overhead. The effectiveness of Multi-Head Co-Training is demonstrated in an empirical study on standard semi-supervised learning benchmarks.

Multi-modal Multi-label Emotion Recognition (MMER) aims to identify various human emotions from heterogeneous visual, audio and text modalities. Previous methods mainly focus on projecting multiple modalities into a common latent space and learning an identical representation for all labels, which neglects the diversity of each modality and fails to capture richer semantic information for each label from different perspectives. Besides, associated relationships of modalities and labels have not been fully exploited. In this paper, we propose versaTile multi-modAl learning for multI-labeL emOtion Recognition (TAILOR), aiming to refine multi-modal representations and enhance discriminative capacity of each label. Specifically, we design an adversarial multi-modal refinement module to sufficiently explore the commonality among different modalities and strengthen the diversity of each modality. To further exploit label-modal dependence, we devise a BERT-like cross-modal encoder to gradually fuse private and common modality representations in a granularity descent way, as well as a label-guided decoder to adaptively generate a tailored representation for each label with the guidance of label semantics. In addition, we conduct experiments on the benchmark MMER dataset CMU-MOSEI in both aligned and unaligned settings, which demonstrate the superiority of TAILOR over the state-of-the-arts.

Elevator maintenance plays an important role in the safety and economic aspects of daily life, and the location of elevator maintenance sites is crucial in this scenario. A reasonable location can help reduce economic expenses and ensure the healthy operation of the system of maintenance sites. However, due to the lack of candidate location sets and unevenly distributed data layers, existing coverage siting methods have difficulties and challenges in solving this problem, and ignore the fairness of load among different sites. To address above problems, We propose a heuristic coverage siting method called MRM, which is based on multi-round MeanShift clustering with coverage radius variation. Experiments on a real elevator dataset with noise show that MRM not only achieves 99.21 % coverage with fewer sites but also achieves fairly good global fairness.

With the rapid development of Artificial Intelligence (AI), a long line of past papers have shown concerns about the data distribution shifts problem in image classification models via deep learning. Moreover, there is also an Out-of-Distribution (OOD) problem in perturbation-based explanation methods for DNNs ineXplainable Artificial Intelligence (XAI), because the generated perturbation samples may be not the same distribution as the original dataset. We explore the limitations of post-hoc Learning to Explain (L2X) explanation methods that use approximators to mimic the behavior of DNNs. We propose a training pipeline called Feature Importance eXplanation(-based) Calibration (FIXC), which efficiently extracts feature importance without using imitations of existing DNNs. We use feature importance as additional information to calibrate data distribution shifts. The evaluation of the corrupted dataset and DNNs benchmarks shows that the FIXC effectively improves the classification accuracy of corrupted images. Experiments on popular vision datasets show that the FIXC outperforms state-of-the-art methods on calibration metrics While the training pipeline provides a calibrated feature importance explanation. We also provide an analysis of our method based on game interaction theory.

Learning robust representations for nodes in graphs is crucial for graph learning tasks. Graph Neural Networks(GNNs) attract much attention recently as the frameworks achieve great success in node representation learning. Existing state-of-the-art GNN methods (like GCN) aggregate messages from neighbor nodes through message passing neural network to update representations for nodes. However, the message passing strategy fails to capture the structural similarity between nodes. Besides, it assumes that neighbor nodes are independent and ignores abundant local neighbor structures around nodes in real networks. This weakness may hurt the performance of GNNs in some classification tasks. To capture the overlooked information, in the experimental investigation, we found that co-occurrence probabilities based on random walks can preserve local neighbor structures among nodes well. Furthermore, we propose a novel but effective method to encode local structure information into node features by co-occurrence probabilities. We call this method Local Structure Enhanced Graph Neural Network, short as LSEGNN. Extensive experiments are conducted in benchmark datasets and the results show the effectiveness of our method.

Most existing point cloud completion methods assume the input partial point cloud is clean, which is not practical in practice, and are Most existing point cloud completion methods assume the input partial point cloud is clean, which is not the case in practice, and are generally based on supervised learning. In this paper, we present an unsupervised generative adversarial autoencoding network, named UGAAN, which completes the partial point cloud contaminated by surroundings from real scenes and cutouts the object simultaneously, only using artificial CAD models as assistance. The generator of UGAAN learns to predict the complete point clouds on real data from both the discriminator and the autoencoding process of artificial data. The latent codes from generator are also fed to discriminator which makes encoder only extract object features rather than noises. We also devise a refiner for generating better complete cloud with a segmentation module to separate the object from background. We train our UGAAN with one real scene dataset and evaluate it with the other two. Extensive experiments and visualization demonstrate our superiority, generalization and robustness. Comparisons against the previous method show that our method achieves the state-of-the-art performance on unsupervised point cloud completion and segmentation on real data.

Although achieving remarkable progress, it is very difficult to induce a supervised classifier without any labeled data. Unsupervised domain adaptation is able to overcome this challenge by transferring knowledge from a labeled source domain to an unlabeled target domain. Transferability and discriminability are two key criteria for characterizing the superiority of feature representations to enable successful domain adaptation. In this paper, a novel method called learning TransFerable and Discriminative Features for unsupervised domain adaptation (TFDF) is proposed to optimize these two objectives simultaneously. On the one hand, distribution alignment is performed to reduce domain discrepancy and learn more transferable representations. Instead of adopting Maximum Mean Discrepancy (MMD) which only captures the first-order statistical information to measure distribution discrepancy, we adopt a recently proposed statistic called Maximum Mean and Covariance Discrepancy (MMCD), which can not only capture the first-order statistical information but also capture the second-order statistical information in the reproducing kernel Hilbert space (RKHS). On the other hand, we propose to explore both local discriminative information via manifold regularization and global discriminative information via minimizing the proposed class confusion objective to learn more discriminative features, respectively. We integrate these two objectives into the Structural Risk Minimization (RSM) framework and learn a domain-invariant classifier. Comprehensive experiments are conducted on five real-world datasets and the results verify the effectiveness of the proposed method.

Graph Neural Networks (GNNs) have achieved remarkable success in graph-related tasks by combining node features and graph topology elegantly. Most GNNs assume that the networks are homophilous, which is not always true in the real world, i.e., structure noise or disassortative graphs. Only a few works focus on generalizing graph neural networks to heterophilous or low homophilous networks, where connected nodes may have different labels. In this paper, we design a simple and effective Graph Structure Learning strategy based on Feature and Label consistency (GSLFL) to increase the homophilous level of networks for generalizing any existing GNNs to heterophilous networks. Specifically, we first introduce a method to learn graph structure based on node features and then modify the graph structure based on label consistency. Further, we combine the GSLFL with three existing GNNs to learn node representations and graph structure together. And we design a self-training method to iteratively train models and modify graph structure with pseudo-labels. Finally, our empirical results on 6 public networks with homophily or heterophily, and structure attacks show that our methods outperform the state-of-the-art methods in most cases.

Exploiting label correlation is crucially important in multi-label learning, where each instance is associated with multiple labels simultaneously. Multi-label learning is more complex than single-label learning for that the labels tend to be correlated. Traditional multi-label learning algorithms learn independent classifiers for each label and employ ranking or threshold on the classification results. Most existing methods take label correlation as prior knowledge, which have worked well, but they failed to make full use of label dependency. As a result, the real relationship among labels may not be correctly characterized and the final prediction is not explicitly correlated. To address these problems, we propose a novel high-order multi-label learning algorithm of Label collAboration based Multi-laBel learning (LAMB). With regard to each label, LAMB utilizes collaboration between its own prediction and the prediction of other labels. Extensive experiments on various datasets demonstrate that our proposed LAMB algorithm achieves superior performance over existing state-of-the-art algorithms. In addition, one real-world dataset of channelrhodopsins chimeras is assessed, which would be of great value as pre-screen for membrane proteins function.

Time series has wide applications in the real world and is known to be difficult to forecast. Since its statistical properties change over time, its distribution also changes temporally, which will cause severe distribution shift problem to existing methods. However, it remains unexplored to model the time series in the distribution perspective. In this paper, we term this as Temporal Covariate Shift (TCS). This paper proposes Adaptive RNNs (AdaRNN) to tackle the TCS problem by building an adaptive model that generalizes well on the unseen test data. AdaRNN is sequentially composed of two novel algorithms. First, we propose Temporal Distribution Characterization to better characterize the distribution information in the TS. Second, we propose Temporal Distribution Matching to reduce the distribution mismatch in TS to learn the adaptive TS model. AdaRNN is a general framework with flexible distribution distances integrated. Experiments on human activity recognition, air quality prediction, and financial analysis show that AdaRNN outperforms the latest methods by a classification accuracy of 2.6% and significantly reduces the RMSE by 9.0%. We also show that the temporal distribution matching algorithm can be extended in Transformer structure to boost its performance.

Graph Neural Networks (GNNs) have achieved significant success in handling graph-structured data, such as knowledge graphs, citation networks, molecular structures, etc. However, most of them are usually shallow structures because of the over-smoothing problem that the representations of nodes are indistinguishable when stacking many layers. Several recent studies have tried to design deep GNNs for powerful expression ability by enlarging the receptive fields to aggregate information from high-order neighbors. But deep models may give rise to overfitting problem. In this paper, we propose a novel insight to aggregate more useful information based on multi-view which does not require deep structures. Specifically, we first design two complementary views to describe global topology and feature similarity of nodes. Then we devise an attention strategy to fuse node representations, named M ulti-V iew G raph C onvolutional N etowrk(MV-GCN). Further, we introduce a self-supervised technique to learn node representations by contrastive learning on different views, which can learn distinctive node embeddings from a large number of unlabeled data, named M ulti-V iew C ontrastive G raph C onvolutional Network(MV-CGC). Finally, we conduct extensive experiments on six public datasets for node classification, which prove the superiority of two proposed models compared with state-of-the-art methods.

Node classification is an important yet challenging task in various network applications, and many effective methods have been developed for a single network. While for cross-network scenarios, neither single network embedding nor traditional domain adaptation can directly solve the task. Existing approaches have been proposed to combine network embedding and domain adaptation for cross-network node classification. However, they only focus on domain-invariant features, ignoring the individual features of each network, and they only utilize 1-hop neighborhood information (local consistency), ignoring the global consistency information. To tackle the above problems, in this paper, we propose a novel model, Adversarial Separation Network(ASN), to learn effective node representations between source and target networks. We explicitly separate domain-private and domain-shared information. Two domain-private encoders are employed to extract the domain-specific features in each network and a shared encoder is employed to extract the domain-invariant shared features across networks. Moreover, in each encoder, we combine local and global consistency to capture network topology information more comprehensively. ASN integrates deep network embedding with adversarial domain adaptation to reduce the distribution discrepancy across domains. Extensive experiments on real-world datasets show that our proposed model achieves state-of-the-art performance in cross-network node classification tasks compared with existing algorithms.

Mobile crowdsensing has been regarded as an efficient paradigm for performing large-scale sensing tasks. In this paper, we consider a specific scenario, where the crowdsensing platform needs to collect the sensing data in a requested time window (RTW), and mobile users would bid for their sensing time windows. This process could be modeled as a reverse auction. In this context, the Vickrey-Clark-Groves (VCG) mechanism becomes a generic auction mechanism that uniquely guarantees both truthfulness and efficiency, but it is vulnerable to false-name bidding and generates high overpayment for the platform. Thus in this paper, we design the core-selecting mechanism to solve VCG’s vulnerability and improve the revenue. We demonstrate that our proposed mechanism achieves the properties of RTW feasibility, efficiency, individual rationality, and false-name-proofness. Besides, to minimize the incentives of users to deviate from truthful-telling, we adopt a VCG-nearest payment rule and propose an efficient algorithm called CCG-TWC. Our extensive simulation results show that the core-selecting mechanism could reduce VCG’s overpayment by about 10%.

In Multi-View Multi-Label (MVML) learning, each instance is represented with several feature vectors and associated with multiple labels simultaneously. Existing models are mainly built on a shared latent space among multiple views and tend to select features that are distinguishable for all labels. However, they ignore the distinctive characteristic of each view and fail to extract the discriminative information related to corresponding labels. To address the above challenges, we proposed a novel approach named Label-specific Alignment with Adversarial Multi-view Representation (LAAMR). For one thing, shared and private information is extracted with adversarial training and orthogonal constraint, which ensures complementarity and consistency among multiple views. For another, label semantic information is well captured with multi-head attention, and then utilized as the guidance to exploit label-view dependence with alignment attention, resulting in label-specific representations. Experiments on various MVML datasets validate the effectiveness of LAAMR.

Wastewater treatment is an important link to solve the pollution of water resources and realize the recycling of water resources. Predicting effluent water quality by influent water quality is an important part of the actual production process. The existing forecasting methods have the problem of low accuracy and efficiency, and cannot be used well in the actual production process. To tackle the above issues, we propose a novel method to use the transformer model to fit the wastewater treatment process which improves the prediction of effluent water quality. Experiments on various datasets validate the effectiveness of our proposed method. In the worst case, the sample prediction accuracy rate of our model achieves as high as 96.8%. At the same time, our model training with thousands of real production data, the training time is not more than two minutes. As far as we know, we are the first to use the transformer model in wastewater treatment tasks and achieve fast and accurate predictions. In the future, we will apply our model in actual production to achieve accurate predictions in dynamic data update scenarios.

Graph Neural Networks (GNNs) are powerful tools for modeling graph-structured data to solve the tasks such as node classification, link prediction along with graph classification. For the graph classification task, properly defining the pooling strategies to vary the size and structure of the input graph, is of vital importance to generate a graph-level representation of the input graph. However, the existing GNN models usually fail to effectively capture the graph substructure information in pooling process. Besides, the importance of nodes(supernodes) within a graph has not been well-reflected. To remedy these limitations, we propose Gated Structure Aware Pooling (GSAPool), a sparse and differentiable pooling method, which focuses on retaining the graph substructure information during the process of pooling in an end-to-end fashion. Specifically, GSAPool utilizes dual gates along with a self-attention network to integrate the local structure to form clusters' embeddings. It also employs a novel formulation to capture the importance of each node/supernode in an input graph. Experiment results show that GSAPool achieves competitive graph classification performance over the state-of-the-art graph representation learning methods.

With the acceleration of urbanization, there is an explosive growth trend of the stroke burden in China. Consequently, it is necessary to adopt a regular screen to reduce brain damage and other complications. For one thing, the risk factors screened can be divided into 6 subsets: medical history, family history, fasting blood, lifestyle, physical characteristics, and demographic information. For another, the screen results can be categorized into 5 groups: normal carotid plaque, thickening, unstable plaque, and arterial stenosis. To address the above problems, Multi-View Multi-Label (MVML) learning provides a fundamental framework for handling such tasks, which treats the risk factors as multi-view data and the screen results as multi-label data. Besides, it is a challenging issue of how to effectively utilize view correlations and label correlations in MVML learning. In this paper, we propose a novel Multi-view Multi-label Learning with Dual-Attention Networks (MML-DAN) for stroke screen, which can strike a balance between label-specific views and label correlations. Meanwhile, the attention mechanism aims to focus on the most pertinent information of the corresponding labels rather than using all available risk factors equally. Experiments on one real-world dataset of stroke screen demonstrate the superiority of the proposed MML-DAN for stroke screen prediction. To further validate the effectiveness and generalization capability of MML-DAN, extensive experiments on 5 benchmark MVML datasets show that MML-DAN significantly outperforms other state-of-the-art MVML algorithms.

Transfer learning has been demonstrated to be successful and essential in diverse applications, which transfers knowledge from related but different source domains to the target domain. Online transfer learning(OTL) is a more challenging problem where the target data arrive in an online manner. Most OTL methods combine source classifier and target classifier directly by assigning a weight to each classifier, and adjust the weights constantly. However, these methods pay little attention to reducing the distribution discrepancy between domains. In this paper, we propose a novel online transfer learning method which seeks to find a new feature representation, so that the marginal distribution and conditional distribution discrepancy can be online reduced simultaneously. We focus on online transfer learning with multiple source domains and use the Hedge strategy to leverage knowledge from source domains. We analyze the theoretical properties of the proposed algorithm and provide an upper mistake bound. Comprehensive experiments on two real-world datasets show that our method outperforms state-of-the-art methods by a large margin.

Personalized recommendation has become increasingly pervasive due to its great commercial value in business. Deep neural networks can automatically exvacate the behavior patterns from the historical interaction records, which has achieved excellent results in related tasks. Among them, the variational auto-encoders have been shown to be superior for learning to rank and recommendation on massive data. However, prior work neglects the association between user behavior and side information, which affects the quality of recommendation services to some extent. In this paper, we propose a feature-aware attentive variational auto-encoder for top-N recommendation. The attention mechanism is utilized to capture the relationship between user's representation and side information through a sub network, balancing the fusion weight of attributes in the main network. In addition, this method tries to construct combination of features in the high-dimensional embedding space, helping mining the promotion of side information at a finer scale. Experiments conducted on real-world datasets demonstrate the effectiveness over the state-of-art methods.

Heterogeneous Information Networks (HINs) are prevalent in our daily life, such as social networks and bibliography networks, which contain multiple types of nodes and links. Heterogeneous information network embedding is an effective HIN analysis method, it aims at projecting network elements into a lower-dimensional vector space for further machine learning related evaluations, such as node classification, node clustering, and so on. However, existing HIN embedding methods mainly focus on extracting the semantic-related information or close neighboring relations, while the high-level proximity of the network is also important but not preserved. To address the problem, in this paper we propose CGAT, a semi-supervised heterogeneous information network embedding method. We optimize the graph attention network by adding additional convolution layers, thereby we can extract multiple types of semantics and preserve high-level information in HIN embedding at the same time. Also, we utilize label information in HINs for semi-supervised training to better obtain the model parameters and HIN embeddings. Experimental results on real-world datasets demonstrate the effectiveness and efficiency of the proposed model.

Multi-modal multi-label learning provides a fundamental framework for complex objects, which can be represented with multiple modalities and annotated with multiple labels simultaneously. Different modalities can usually provide complementary information, which may lead to improved performance. What's more, exploiting label correlations is crucially important to multi-label learning. However, most existing multi-label learning approaches do not sufficiently consider the complementary information among different modalities. In this paper, we propose a novel end-to-end deep learning framework named Rethinking Modal-oriented Label Correlations (RMLC), which sequentially polish the label prediction with each individual modality. In order to explicitly account for the correlated prediction of multiple labels, RMLC leverages an efficient sequential modal-based exploration to rethink label correlations. The final prediction of each label involves the collaboration between modal-specific prediction and the prediction of other labels based on cross-modal interaction. Comprehensive experiments on benchmark datasets validate the effectiveness and competitiveness of the proposed RMLC approach.

In the age of the development of artificial intelligence, we face the challenge on how to obtain high-quality data set for learning systems effectively and efficiently. Crowdsensing is a new powerful tool which will divide tasks between the data contributors to achieve an outcome cumulatively. However, it arouses several new challenges, such as incentivization. Incentive mechanisms are significant to the crowdsensing applications, since a good incentive mechanism will attract more workers to participate. However, existing mechanisms failed to consider situations where the crowdsourcer has to hire capacitated workers or workers from multiregions. We design two objectives for the proposed multiregion scenario, namely, weighted mean and maximin. The proposed mechanisms maximize the utility of services provided by a selected data contributor under both constraints approximately. Also, extensive simulations are conducted to verify the effectiveness of our proposed methods.

Integral membrane proteins (MPs) make up a large proportion of the genomes of many organisms and the amount of sequenced proteins is growing at an unprecedented pace. As a result, MPs function pre-screen is of great necessity. To be functional, MPs must be expressed and localized through a series of elaborate sub-cellular processes. It is worth noting that subtle changes in sequence may lead to drastic changes in expression and localization. In light of the above observation, taking the complementary information of sequence-based proteins into consideration, a novel boosting Dual-Modal Dual-Label (DMDL) algorithm with hypothesis reuse is proposed. On the one hand, DMDL treats sequence features and structure features as dual-modal. On the other hand, DMDL considers eukaryotic expression and plasma membrane localization as dual-label, which would be of great value as a pre-screen for MPs function. Meanwhile, two label sets interact, which can not only make the best use of two modal features, but also could effectively exploit the relationship between two label sets. An assessment of real-world channelrhodopsins (ChRs) chimeras clearly validate the effectiveness of DMDL algorithm compared with state-of-the-art algorithms. In addition, extensive experiments are performed on adapted public datasets, showing effectiveness of hypothesis reuse mechanism in DMDL.

With the emergence of diverse data collection techniques, objects in real applications can be represented as multi-modal features. What's more, objects may have multiple semantic meanings. Multi-modal and Multi-label [1] (MMML) problem becomes a universal phenomenon. The quality of data collected from different channels are inconsistent and some of them may not benefit for prediction. In real life, not all the modalities are needed for prediction. As a result, we propose a novel instance-oriented Multi-modal Classifier Chains (MCC) algorithm for MMML problem, which can make convince prediction with partial modalities. MCC extracts different modalities for different instances in the testing phase. Extensive experiments are performed on one real-world herbs dataset and two public datasets to validate our proposed algorithm, which reveals that it may be better to extract many instead of all of the modalities at hand.

Point-based value iteration methods are a class of practical algorithms for solving the POMDP model. The critical process of these methods is the exploration of the belief point set B. Forward search value iteration(FSVI) can reduce the complexity and improve efficiency significantly by using the optimal strategy of the underlying MDP. However, it does not utilize the observations of the model, making it not so efficient in the large-scale POMDP problems. A probabilistic forwardsearching value iteration algorithm (PFSVI) is presented in the paper to make up the shortage of FSVI. During the exploration, PFSVI uses the alias method to sample the action a* based on weighted QMDPfunction and sample the state based on b and the transition function. Then, PFSVI selects the observation z, which lets the successor point ba*,zfarthest from B. PFSVI can improve the effect by sampling according to the environment and reaching more vast space than FSVI. Experiment results of four benchmarks show that PFSVI can achieve better global optimal solutions than FSVI and PBVI, especially in large-scale problems.

Aspect-based sentiment analysis (ABSA) is a fine-grained sentiment analysis problem, which has received more and more attention in recent years. Convolutional Neural Networks and their variants have shown potentialities for tackling the problem recently. Building upon this line of research, we propose a novel architecture named Multi-Granularity Position-Aware Convolutional Memory Network (MP-CMN) for ABSA in this paper. MP-CMN utilizes multiple convolutional layers to extract features of different granularities to build the convolutional memories, and then incorporates aspect information and position information into convolutional memory network via attention mechanism. To make the mechanism of our model clear, we also make some visualization and case studies. Experiment results on standard SemEval 2014 datasets demonstrate the effectiveness of the proposed model.

Wastewater treatment is a pivotal approach dealing with water pollution problem. During the actual production process of wastewater treatment, predicting input water's output quality acts as a critical step. Existing methods rely mainly on biotechnology and demand a long time cost, which cannot be applied in practice. Therefore, we propose a novel approach to address this problem, which utilizes the recent advance in deep neural networks, and achieves both high accuracy and low running cost simultaneously. Only less than 3% test samples have prediction errors that exceed limits even in the worst case, and only less than 40 seconds are needed when training our model on a real-world dataset of thousands of samples. To our best knowledge, we are the first to apply convolutional neural network to predict output water quality indicators accurately and rapidly. Furthermore, our proposed model supports dynamic update to cope with the variations of wastewater treatment plant configurations.

Aspect-based sentiment analysis (ABSA) is a fine-grained task. Recurrent Neural Network (RNN) model armed with attention mechanism seems a natural fit for this task, and actually it achieves the state-of-the-art performance recently. However, previous attention mechanisms proposed for ABSA may attend irrelevant words and thus downgrade the performance, especially when dealing with long and complex sentences with multiple aspects. In this paper, we propose a novel architecture named Hierarchical Gate Memory Network (HGMN) for ABSA: firstly, we employ the proposed hierarchical gate mechanism to learn to select the related part about the given aspect, which can keep the original sequence structure of sentence at the same time. After that, we apply Convolutional Neural Network (CNN) on the final aspect-specific memory. We conduct extensive experiments on the SemEval 2014 and Twitter dataset, and results demonstrate that our model outperforms attention based state-of-the-art baselines.

Most existing multi-label text classification (MLTC) approaches only exploit label correlations from label pairwises or label chains. However, in the real world, features of instances have much importance for classification. In this paper, we propose a simple but efficient framework for MLTC called Hybrid Latent Dirichlet Allocation Multi-Label (HLDAML). To be specific, the topics of text features (i.e., a concrete description of documents) and the topics of label sets (i.e., a summarization of documents) can be obtained from training data by topic model before building models for multi-label classification. After that, hybrid topics can be used in existing approaches to improve the performance of MLTC. Experiments on several benchmark datasets demonstrate that the proposed framework is general and effective when taking text features and label sets into consideration simultaneously. It is also worth mentioning that we construct a new multi-label dataset called Parkinson about diagnosing parkinson disease by Traditional Chinese Medicine.

Multi-label text classification is generally more difficult for its exponential output label space and more flexible input document considering its corresponding number of labels. We observed that some long documents have only one or two labels while some short documents are related to much more labels. In this paper, we propose a dynamic Gated Neural Network architecture to simultaneously process the document through two parts, one to extract the most informative semantics and filter redundant information, the other to capture context semantics with most of the information in the document kept. And semantics from these two parts are dynamically controlled by a gate to perform subsequent classification. And to better the training we incorporate label dependencies into traditional binary cross-entropy loss by exploiting label co-occurrences. Experimental results on AAPD and RCV1-V2 datasets show that our proposed methods achieve state-of-art performance. Further analysis of experimental results demonstrate that the proposed methods not only capture enough feature information from both long and short documents assigned with various labels, but also exploit label dependency to regularize the proposed model to further improve its performance.

Biomedical event extraction plays an important role in the field of biomedical text mining, and the event trigger detection is the first step in the pipeline process of event extraction. Event trigger can clearly indicates the occurrence of related events. There have been many machine learning based methods applied to this area already. However, most previous work have omitted two crucial points: (1) Class Difference: They simply regard non-trigger as same level class label. (2) Information Isolation: Most methods only utilize token level information. In this paper, we propose a novel model based on gate mechanism, which identifies trigger and non-trigger words in the first stage. At the same time, we also introduce additional fusion layer in order to incorporate sentence level information for event trigger detection. Experimental results on the Multi Level Event Extraction (MLEE) corpus achieve superior performance than other state-of-the-art models. We have also performed ablation study to show the effectiveness of proposed model components.

Crowdsourcing is a new paradigm which divides work between participants to achieve a cumulative result. To achieve good service quality for a crowdsourcing system, incentive mechanisms are necessary to attract more workers to participate. Most of existing mechanisms apply only for the crowdsourcing scenario where the crowdsourcer will employ the workers to perform certain tasks only in a single market. In this paper, we consider that the crowd-sourcer would like to allocate tasks to several different sub-markets with a shared budget. The crowdsourcer wants to maximize her utility through the crowdsourcing campaign with certain budget constraint. We design two optimization objectives for our problem, maximin and weighted average. We propose a cross-market profit extract mechanism based on random sampling. Furthermore, we extend our algorithm to the online condition, and give the performance lowerbound for both offline and online cases for the cross-market crowdsourcing system. We also conduct extensive numerical experiments to demonstrate the effectiveness of our approach.

The volume of short text data increases rapidly these years. Data examples include tweets and online Q&A pairs. It is essential to organize and summarize these data automatically. Topic model is one of the effective approaches, whose application domains include text mining, personalized recommendat ion and so on. Conventional models like pLSA and LDA are designed for long text data. However, these models may suffer from the sparsity problem brought by lacking words in short text scenarios. Recent studies such as BTM show that using word co-occurrent pairs is effective to relieve the sparsity problem. However, both BTM and extended models ignore the quantifiable relationship between words. From our perspectives, two more related words should occur in the same topic. Based on this idea, we introduce a model named RIBS, which makes use of RNN to learn relationship. By using the learned relationship, we introduce a model named RIBS-Bigrams, which can display topics with bigrams. Through experiments on two open-source and real-world datasets, RIBS achieves better coherence in topic discovery, and RIBS-Bigrams achieves better readability in topic display. In the document characterization task, the document representation of RIBS can lead better purity and entropy in clustering, higher accuracy in classification.

In multi-label classification settings, one of the most common problems is the massive label output space. To alleviate this, some methods opt to exploit label correlations to reduce the output space during prediction. However, these methods sacrifice efficiency or ignore global label correlations. In addition, label imbalances are another problem that is prevalent in multi-label classification. Current methods of correcting for imbalance oftentimes use single-label methods, which fail to consider label correlations. In this paper, we introduce general frameworks that incorporate topic modeling to seamlessly address both problems. We show that these frameworks can allow even the most naïve methods, such as Binary Relevance, to perform similarly to state-of-the-art methods. Furthermore, we show that our frameworks can also adapt state-of-the-art methods to perform better than the methods by themselves.

Fine-grained sentiment analysis is a useful tool for producers to understand consumers’ needs as well as complaints about products and related aspects from online platforms. In this article, we define a novel task named “Multi-Entity Aspect-Based Sentiment Analysis (ME-ABSA)”. It investigates the sentiment towards entities and their related aspects. It makes the well-studied aspect-based sentiment analysis a special case of this type, where the number of entities is limited to one. We contribute a new dataset for this task, with multi-entity Chinese posts in it. We propose to model context, entity, and aspect memory to address the task and incorporate dependency information for further improvement. Experiments show that our methods perform significantly better than baseline methods on datasets for both ME-ABSA task and ABSA task. The in-depth analysis further validates the effectiveness of our methods and shows that our methods are capable of generalizing to new (entity, aspect) combinations with little loss of accuracy. This observation indicates that data annotation in real applications can be largely simplified.

Inspired by recent works in Aspect-Based Sentiment Analysis (ABSA) on product reviews and faced with more complex posts on social media platforms mentioning multiple entities as well as multiple aspects, we define a novel task called Multi-Entity Aspect-Based Sentiment Analysis (ME-ABSA). This task aims at fine-grained sentiment analysis of (entity, aspect) combinations, making the well-studied ABSA task a special case of it. To address the task, we propose an innovative method that models Context memory, Entity memory and Aspect memory, called CEA method. Our experimental results show that our CEA method achieves a significant gain over several baselines, including the state-of-the-art method for the ABSA task, and their enhanced versions, on datasets for ME-ABSA and ABSA tasks. The in-depth analysis illustrates the significant advantage of the CEA method over baseline methods for several hard-to-predict post types. Furthermore, we show that the CEA method is capable of generalizing to new (entity, aspect) combinations with little loss of accuracy. This observation indicates that data annotation in real applications can be largely simplified.

Transfer learning is an important topic in machine learning and has been broadly studied for many years. However, most existing transfer learning methods assume the training sets are prepared in advance, which is often not the case in practice. Fortunately, online transfer learning (OTL), which addresses the transfer learning tasks in an online fashion, has been proposed to solve the problem. This paper mainly focuses on the heterogeneous OTL, which is in general very challenging because the feature space of target domain is different from that of the source domain. In order to enhance the learning performance, we designed the algorithm called Heterogeneous Ensembled Online Transfer Learning (HetEOTL) using ensemble learning strategy. Finally, we evaluate our algorithm on some benchmark datasets, and the experimental results show that HetEOTL has better performance than some other existing online learning and transfer learning algorithms, which proves the effectiveness of HetEOTL.

Similarity Search has been studied in many different fields of computer science, including data mining, information retrieval, databases and so on. Document similarity self-join is a crucial part of lots of applications, such as near-duplicate document detection, document clustering and web search. On a collection of documents, document similarity self-join finds out all pairs of documents whose similarity values are no lower than a threshold value. However, similarity search is a computation-intensive procedure and consumes a large amount of time as the dataset size increases. Thus, many serial algorithms focus on speeding up the process by decreasing the possible similarity candidates for each query object on high-dimensional sparse datasets, including documents. However, the efficiency of those serial algorithms degrade badly as the threshold decreases. Parallel implementations based on OpenMP or MapReduce also adopt the pruning policy and do not solve the problem thoroughly. In this context, taking into account features of document datasets, we propose 2Step-SSJ, which solves the document similarity self-join in CUDA environment on GPUs. 2Step-SSJ performs the similarity self-join in two steps, i.e., similarity computing on the inverted list and similarity computing on the forward list, which compromises between the memory visiting and dot-product computation. The experimental results show that 2Step-SSJ could solve the problem much faster than existing methods on three benchmark text corpora, achieving the speedup of 2×-23× against the state-of-the-art parallel algorithm in general, while keep a relatively stable running time with different values of the threshold.

The demand for mining massive short-text data from the Internet has promoted researches on topic models. There exist many schemes trying to solve the sparsity problems brought by short texts, mainly based on data aggregation or model improvement. Among them, Biterm Topic Model changes the way of modeling topics, which is on document-level biterms and has shown creativity and effectiveness. However, this may ignore those semantically similar and rarely co-occurrent word pairs, which are denoted as global biterms in this paper. Inspired by the successful application of word embeddings in GPU-DMM, we exploit word embeddings to extract semantically similar word pairs from the whole corpus to help discover better topics. We call this model as GloSS, which takes advantages of both the approach to model topics and word embeddings. Experimental results on two open-source and real datasets are superior to state-of-the-art topic models for short texts.

Transfer learning has made great achievements in many fields and many excellent algorithms have been proposed. In recent years, many scholars have focused on a new research area called online transfer learning, which is different from general transfer learning. Online transfer learning concentrates on how to build a good classifier on the target domain when the training data arrive in an online/sequential manner. This paper focuses on online transfer learning problem based on a single source domain under homogeneous space. The existing algorithms HomOTL-I and HomOTL-II simply ensemble the classifiers on the source and target domains directly. When the distribution difference between the source domain and the target domain is large, it will not result in a good transfer effect. We are inspired by the idea of the boosting algorithm, that is we could form a strong classification model by a combination of multiple weak classifications. We train multiple classifiers on the source domain in an offline manner using AdaBoost algorithm, combine these classifiers on source domain with the classifier trained in an online manner on the target domain to form multiple weak combination in an ensemble manner. Based on the above ideas, we propose two algorithms AB-HomOTL-I and AB-HomOTLII, which have different ways to adjust the weights. We tested our algorithms on sentiment analysis dataset and 20newsgroup dataset. The results show that our algorithms are superior to other baseline algorithms.

In our daily life, short texts have been everywhere especially since the emergence of social network. There are countless short texts in online media like twitter, online Q&A sites and so on. Discovering topics is quite valuable in various application domains such as content recommendation and text characterization. Traditional topic models like LDA are widely applied for sorts of tasks, but when it comes to short text scenario, these models may get stuck due to the lack of words. Recently, a popular model named BTM uses word co-occurrence relationship to solve the sparsity problem and is proved effectively. However, both BTM and extended models ignore the inside relationship between words. From our perspectives, more related words should appear in the same topic. Based on this idea, we propose a model named RIBS-TM which makes use of RNN for relationship learning and IDF for filtering high-frequency words. Experiments on two real-world short text datasets show great utility of our model.

Parkinson's disease is a debilitating and chronic disease of the nervous system. Traditional Chinese Medicine (TCM) is a new way for diagnosing Parkinson, and the data of Chinese Medicine for diagnosing Parkinson is a multi-label data set. Considering that the symptoms as the labels in Parkinson data set always have correlations with each other, we can facilitate the multi-label learning process by exploiting label correlations. Current multi-label classification methods mainly try to exploit the correlations from label pairwise or label chain. In this paper, we propose a simple and efficient framework for multi-label classification called Latent Dirichlet Allocation Multi-Label (LDAML), which aims at leaning the global correlations by using the topic model on the class labels. Briefly, we try to obtain the abstract “topics” on the label set by topic model, which can exploit the global correlations among the labels. Extensive experiments clearly validate that the proposed approach is a general and effective framework which can improve most of the multi-label algorithms' performance. Based on the framework, we achieve satisfying experimental results on TCM Parkinson data set which can provide a reference and help for the development of this field.

E-cigarettes(vape) are now the most commonly used tobacco product among youth in the United States. Ads are claiming e-cigarettes help smokers quit, but most of them contain nicotine, which can cause addiction and harm the developing adolescent brain. Therefore national, state and local health organizations have proposed anti-vaping campaigns to warn the potential risks of e-cigarettes. Since there is not definitive evidence that e-cigarettes cause long-term harm, these campaigns received pro-vapors' fight back, and collected a high volume of opponent messages in social media. Thus when we analyze the feedback of anti-vaping campaigns, it is crucial to partition audience into different clusters according to their attitude. Motivated by this, in this paper, we propose the “Community Detection on Anti-vaping Campaign Audience (CodeVan)” problem and design the Sorento method to solve it. Sorento computes users' intimacy scores based on their social connections, repost relations and content similarities. Extensive experiments show the effectiveness of the Sorento algorithm.

In multi-label learning, each object is represented by a single instance while associated with a set of class labels, and labels often have correlations with each other. Exploiting label correlations can improve the performances of classifiers. Current multi-label classification methods mainly consider the correlations from label pairwise or label chain. In this paper, we propose a simple and efficient framework for multi-label classification, called Latent Dirichlet Allocation Multi-Label(LDAML), which aims at leaning the global correlations by using the topic model on the class labels. We regard the label set associated with each instance as a document and each class label in the label set as a word, then obtain the topic in the label space by topic model. The topics of label set which are introduced into feature space as the information of the correlations can improve the prediction. Extensive experiments clearly validate the effectiveness of the proposed approach.

User complaint data are quite valuable because they can reflect deficiencies of companies. Analyzing these short texts can help companies discover what topics users are complaining about. It is critical to locate and respond to these complaints timely so that companies can improve users’ satisfaction and loyalty. As the data volume is large, topic model can help discover key complaint topics quickly. The complaint data are in the form of short texts and streams, traditional topic models like LDA and BTM are not suitable in this scenario, for the reason that LDA is designed for long texts and BTM can not handle streams. This paper firstly proposes an improved shorttext topic model called PMITI-BTM to generate topics from user complaint data statically, and then further extends this algorithm into a dynamic one to suit the streaming feature of data. To further analyze the topic evolution, we finally propose a clustering algorithm called TDWAP to acquire the evolution process of these topics in different time slices. For each algorithm, we do several experiments to prove its efficiency. Results show that our methods not only can improve the performance of short texts topic discovery, but also can discover the evolution of topics.

The MapReduce framework proposed by Google to process large data sets is an efficient framework used in many areas, such as social network, scientific research, electronic business, etc. Hence, many MapReduce frameworks are proposed and implemented on different platforms. However, these MapReduce frameworks have limitations, and they cannot handle the collision problem in the map phase, and the unbalanced workload problem in the reduce phase. In this paper, an Adaptive MapReduce Framework (A-MapCG) is proposed based on the MapCG framework, to further improve the MapReduce performance on GPU platforms. Based on the experiments, we observed that for certain MapReduce applications emitting multiple Key/value (K/V) pairs for the same key, the atomic collision problem degrades the map phase performance of the MapReduce framework substantially. In addition, the workload unbalance problem wastes parallel computing resources and limits the overall reduction phase performance of the MapReduce framework on GPU platforms. A-MapCG uses segmentation table and intra-warp combination to reduce the number of collisions during the map phase. A-MapCG also adopts balanced workload assignment to improve the reduce phase performance. The proposed A-MapCG framework is evaluated on the Tesla K40 GPU hosted by Intel Core i7-4790. The case study shows that the map phase of A-MapCG achieves a speedup of 4.63 over MapCG for the test case, Word Count, with a 64MB workload. The average reduce phase speedup of A-MapCG over MapCG with parallel reductions of Word Count is 6.92. The average reduce phase speedup of A-MapCG over MapCG with serial reductions of Word Count is 4.11.

Traditional Chinese Medicine (TCM) is a holistic integrative medical approach. Exploring the relations between the herbal formulae and the symptoms is a crucial problem in researches of TCM. Unlike existing researches, we treat it as a both multi-instance learning and multi-label learning problem. In this paper, we propose a novel approach, which named Weighted Sampling based on Similar Herbs MIML (WSSH-MIML), to predict one formula's primary symptoms based on multi-instance multi-label framework. We compare the performance of our model with other three state-of-the-art multi-label learning algorithms and the experimental results indicate ours is superior. This study suggests that the MIML technique provides a new research paradigm for mining meaningful TCM information.

We study path auction mechanisms for buying path between two given nodes in a social network, where edges are owned by strategic agents. The well known VCG mechanism is the unique solution that guarantees both truthfulness and efficiency. However, in social network environments, the mechanism is vulnerable to false-name manipulations where agents can profit from placing multiple bids under fictitious names. Moreover, the VCG mechanism often leads to high overpayment. In this paper, we present core-selecting path mechanisms that are robust against false-name bids and address the overpayment problem. Specifically, we provide a new formulation for the core, which greatly reduces the number of core constraints. Based on the new formulation, we present a Vickery-nearest pricing rule, which finds the core payment profile that minimizes the L∞distance to the VCG payment profile. We prove that the Vickery-nearest core payments can be computed in polynomial time by solving linear programs. Our experiment results on real network datasets and reported cost dataset show that our Vickery-nearest core-selecting path mechanism can reduce VCG's overpayment by about 20%.

In most of the GPU kernel programs, the synchronization statements are inserted manually by the programmers, which is very labor intensive, and error-prone. In this paper, we propose a synchronization optimization framework to automatically insert synchronization statements into the GPU kernels at compile time, while eliminating the redundant synchronization statements. We have shown that our framework can not only insert the synchronizations correctly, but also eliminate the redundant synchronizations, which outperforms the existing compiler frameworks that introduce redundant synchronizations using the most conservative strategy. Taking the GPU kernels as the input, our framework leverages data dependence analysis to insert synchronizations. We extend CETUS, a source-to-source compiler framework, to implement our synchronization optimization framework. Experimental results show that our proposed framework achieved 100% correctness by combining extensive evaluation and manual comparison. In addition, the number of synchronization statements in GPU kernels is reduced by 32.5%, and the number of synchronization statements executed is reduced by 28.2% on average by our synchronization optimization framework compared to the original GPU kernels.

Web page segmentation aims to break a page into sections that can reveal the information presentation structure and appear coherent to readers. In this paper, we propose a new web page segmentation framework based on the process of analyzing and understanding web page structure. After extracting the segmentation graph structure, we formulate the label assignment task which determines whether each boundary should segment current block or not on a graph as a structured learning problem. Computation of highest scoring label assignment relies on Viterbi algorithm and joint feature function captures the dependency among boundaries. To solve the learning of parameters, we adopt a learning model based on perceptron algorithm. Furthermore, utilizing the previous framework, we propose a web information crawling application framework which integrates web page segmentation and semantic block classification process.

In the past few years, we have witnessed a rapid development of online social media, from which we can access various short texts. Understanding the topic patterns of these short text is significant. Traditional topic models, like LDA, are not suitable when applied to short text topic analysis due to data sparsity. A lot of efforts have been made to solve this problem. However, there is still significant space to improve the effectiveness of these short text specific methods. In this paper, we proposed a novel word co-occurrence network based method, referred to as biterm pseudo document topic model (BPDTM), which extended the previous biterm topic model(BTM) for short text. We utilized the word co-occurrence network to construct biterm pseudo documents. The proposed model is promising since it represents words with their semantic adjacent biterms and is able to model the corpus-level semantic relation between two words. Besides, BPDTM naturally lengthens the documents, which alleviate the influence for performance exerted by data sparsity. Experiments demonstrated that our model outperformed two baselines, i.e. LDA and BTM, which proved its effectiveness on short text topic modeling task.

Parkinson disease is a chronic, degenerative disease of the central nervous system, which commonly occurs in the elderly. Until now, no treatment has shown efficacy. Traditional Chinese Medicine is a new way for Parkinson, and the data of Chinese Medicine for Parkinson is a multi-label dataset. Classifier Chains(CC) is a popular multi-label classification algorithm, this algorithm considers the relativity between labels, and contains the high efficiency of Binary classification algorithm at the same time. But CC algorithm does not indicate how to obtain the predicted order chain actually, while more emphasizes the randomness or artificially specified. In this paper, we try to apply Multi-label classification technology to build a model of Chinese Medicine for Parkinson, which we hope to improve this field. We propose a new algorithm ETCC based on CC model. This algorithm can optimize the order chain on global perspective and have a better result than the algorithm CC.

Parkinson disease is a chronic, degenerative disease of the central nervous system, which commonly occurs in the elderly. Until now, no treatment has shown efficacy. Traditional Chinese Medicine is a new way for Parkinson, and the data of Chinese Medicine for Parkinson is a multi-label dataset. Classifier Chains(CC) is a popular multi-label classified algorithm, this algorithm considers the relativity between labels, and contains the high efficiency of Binary classification algorithm at the same time. But CC algorithm does not indicate how to obtain the predicted order chain actually, while more emphasizes the randomness or artificially specified. In this paper, we try to apply Multi-label classification technology to build a model of Chinese Medicine for Parkinson, which we hope to improve this field. We propose a new algorithm ETCC based on CC model. This algorithm can optimize the order chain on global perspective and have a better result than the algorithm CC.

Sparse matrix vector multiplication (SpMV) is a basic building block of many scientific applications. Several GPU accelerated SpMV algorithms for the CSR format suffer from workload unbalance for irregular matrices. In this paper, we propose a new auxiliary array assisted CSR format called local segmented reduction based CSR (LSRB-CSR), which enables synchronization free preprocessing and efficient SpMV algorithm with the light weight auxiliary arrays. It is efficient for both regular matrices and irregular matrices with tiny preprocessing overhead. We compare our LSRB-CSR based SpMV algorithm with the CSR-based SpMV from cuSPARSE, the SpMV algorithm based on segmented reduction adopted by CUDPP library, and the CSR5-based SpMV algorithm for both regular and irregular sparse matrices. Compared to cuSparse, our LSRB-CSR based SpMV algorithm could improve the performance by 26% on regular matrices and up to 4750% on irregular matrices. Compared to CUDPP, our LSRB-CSR based SpMV algorithm could improve the average SpMV performance by 210% on regular matrices and 250% on irregular matrices. Our LSRB-CSR based SpMV algorithm has comparable performance as the CSR5 based SpMV algorithm for regular matrices, and achieves better performance over the CSR5 based SpMV algorithm for irregular matrices. Experimental results show that the conversion overhead from the CSR to the LSRB-CSR is only 1/10 of the overhead from the CSR to the CSR5 on average.

The exponential growth of the size of the search space has always been an obstacle to POMDP planning. Heuristics are often used to reduce the search space size and improve computational efficiency. As the advantage of the feature of POMDP problems should be taken into deeper consideration, we analyze the clustering feature of reachable space of POMDP problems and apply policy iteration based on this clustering feature. With insights from theoretical analysis, we have developed a practical POMDP algorithm Point Clustering Feature Based Policy Iteration (PCFBPI). Empirically, PCFBPI is competitive with PBPI in terms of solution quality and convergence efficiency on some large-scale problems.

In our daily life, anomalies are everywhere in various application domains. Most anomalies may cause huge losses if we fail to detect them in advance. A lot of researches on this field have been carried out for years so as to detect anomalies as soon as possible. Among them, machine learning is one of the most used techniques. Previous work tries to improve detection by choosing various classifiers, which has achieved some success. But few has considered the different losses each anomaly might cause. As we know, anomalies with higher significance will cause higher losses. In this paper, we aim to minimize the losses by proposing an improved cost-sensitive GBDT algorithm named LF-GBDT. LF-GBDT is designed to optimize a self-defined loss function. Experiments with both traditional classification algorithms such as CART, Adaboost etc. And cost-sensitive algorithms such as MetaCost, CSC show that our method can both improve the detection of important anomalies and reduce the total losses.

Traditional Chinese medicine (TCM) is a holistic medical approach and the formula's composition discipline is still a mystery. Detecting a formula's structure and herb communities/clusters in TCM Formula networks (TCMF) is a mainly existing problem in data mining of the data sets. In this paper, we devise a novel community similarity calculating method in the process of clustering, which is called Random Walk Hierarchical Clustering (RWHC) algorithm, to identify herb communities by using clustering algorithms based on the formula network of atrophic lung disease. And we also use classic NG modularity function to evaluate the experimental results. The studies suggest that the TCM network clustering approach provides a new research paradigm for mining TCM data from an experience-based medicine, will accelerate TCM drug discovery, and also improve current drug discovery strategies.

Data imbalance is a common problem both in single-label classification (SLC) and multi-label classification (MLC). There is no doubt that the predicting result suffers from this problem. Although, a broad range of studies associate with imbalance problem, most of them focus on SLC and for MLC is relatively less. Actually, this problem arising in MLCis more frequent and complex than in SLC. In this paper, we proceed from dealing with imbalance problem for MLC and propose a new approach called DEML. DEML transforms the whole label set of multi-label dataset into some subsets and each subset is treated as a multi-class dataset with balanced class distribution, which not only addressing imbalance problem but also preserving dataset integrity and consistency. Extensive experiments show that DEML possesses highly competitive performance both in computation and effectiveness.

Point-based value iteration methods improve computational efficiency by reducing the search space size. Although global optimization can be obtained by algorithms such as HSVI and GapMin, their exploration of the optimal action is overly optimistic which therefore slows down the efficiency. In this paper, we propose a novel heuristic search method nso-HSVI (not-so-optimistic Heuristic Search Value Iteration) which uses a Monte-Carlo method to estimate the probabilities that actions are optimal according to the distribution of actions' Q-value function and applies the action of the maximum probability. Experimental results show that nso-HSVI outperforms HSVI, and by a large margin when the scale of the POMDP increases.

Model checking, which can be represent as the paradigm "specifying→verifying", is an effective technology for automatic system verification. But model checking is generally used only to verify the correctness of a system, not to modify it. Sometimes this can be a major limitation. We propose a framework for ATL alpha beta model revision in this paper, study the computational complexity of its related problems, and show that the proposed framework consists with the minimal change principle in propositional belief update. By this paper, We extend the paradigm of ATL model checking to "specifying→verifying→revising" and establish the theoretical basis for automatic revising the strategic ability related properties of coordinated open systems.

The booming development of Online Social Networks (OSNs) provides a novel way of expressing emotions. Research on emotion analysis for tweets in OSNs is becoming increasingly popular in recent years. Traditional emotion analysis only classifies one tweet into a single emotion category. However, in reality, a tweet may belong to several different emotion categories. In the paper, our goal is to predict all emotion labels of each tweet. We use graphic emoticons, punctuation expressions together with a tiny but accurate lexicon to label data and provide a Multi-label Emotion Classification algorithm (MEC) for tweets in Weibo (so called Chinese Twitter). Our method has superior performance to the state-of-the-art method under both single-label and multi-label evaluation measures. We also carried out a case study on Weibo dataset of Malaysia Missing Flight. We came to several meaningful conclusions such as "The outbreak of Anger has a delay after breaking point of Sadness".

Text summarization is of great importance to solve information overload. Salience and coverage are two most important issues for summaries. Most existing models extract summaries by selecting the top sentences with highest scores without using the relationships between sentences, and usually represent the sentences simply basing on lexical or statistical features. As a result, those models can not achieve salience or coverage very well. In this paper, we propose a novel summarization model called Sentence Selection with Semantic Representation (SSSR). SSSR ensures both salience and coverage by learning semantic representations for sentences and applying a well-designed selection strategy to select summary sentences. The selection strategy used in SSSR is to select sentences that can reconstruct the original document with least distortion by means of linear combination. Besides, we improve our selection strategy by reducing redundant information. Then we learn two semantic representations for sentences: (1) weighted mean of word embeddings, (2) deep coding. Both of them are semantic and compact, and can capture similarities between sentences. Extensive experiments on datasets DUC2006 and DUC2007 validate our model.

Landmark heuristics are perhaps the most accurate current known admissible heuristics for optimal planning. A disjunctive action landmark can be seen a form of at-least-one constraint on the actions it contains. In many domains, some critical propositions have to be established for a number of times.Propositional landmarks are too weak to express this kind of constraints.In this paper, we propose to generalize landmarks to multi-valued landmarks to represent the more general cardinality constraints. We present a class of local multi-valued landmarks that can be efficiently extracted from propositional landmarks.By encoding multi-valued landmarks into CNF formulas, we can also use SAT solvers to systematically extract multi-valued landmarks.Experiment evaluations show that multi-valued landmark based heuristics are more close to $h^*$ andcompete favorably with the state-of-the-art of admissible landmark heuristics on benchmark domains.

Identifying outlier objects against main community evolution trends is not only meaningful itself for the purpose of finding novel evolution behaviors, but also helpful for better understanding the mainstream of community evolution. With the definition of community belongingness matrix of data objects, we constructed the transition matrix to least square optimize the pattern of evolutionary quantity between two consecutive belongingness snapshots. A set of properties about the transition matrix is discussed, which reveals its close relation to the step by step community membership change. The transition matrix is further optimized using robust regression methods by minimizing the disturbance incurred by the outliers, and the outlier factor of the anomalous object was defined. Being aware that large proportion of trivial but nomadic objects may exist in large datasets. This paper focus only on the influential community evolutionary outliers which both show remarkable difference from the main body of their community and sharp changes of their membership role within the communities. An algorithm on detection such kind of outliers are purposed in the paper. Experimental results on both synthetic and real world datasets show that the proposed approach is highly effective and efficient in discovering reasonable influential evolutionary community outliers.

Mining community structures has become a general problem which exists in many fields including: Computer-Science, Mathematics, Physics, Biology, Sociology and so on. It has developed rapidly and been used widely in many applications: web data mining, social network analysis, criminal network mining, protein interaction network analysis, metabolic network analysis, genetic network analysis, customers relationship mining and user online behavior analysis, etc. Most community detection algorithms try to obtain the global information of the network, but increasing large scale of the current network makes it computationally expensive. In the meanwhile, the different influence and different behavior of nodes in the network are ignored. In fact, if we know the local information of the network or the interested node, we can easily detect the local community. This paper proposes a multi-resolution local community detection algorithm named MRCDA which uses local structural information in the network to optimize the multi-resolution modularity based on the Potts spin-glass model. A local community can be detected through continuous optimization of the function by expanding from an initial influential node computed by a modified PageRank sorting algorithm. The proposed MRCDA has been tested on both synthetic and real world networks and tested against other algorithms. The experiments demonstrate its efficiency and accuracy.

As more and more people join social network, viral marketing on online social network becomes a new trend of advertising. Motivated by this, plenty of research focuseson how to maximize the information propagation, which is called the influence maximization problem. Traditional work has made significant progress on this topic. However all ad companies have marketing budget, the research of influence maximization problem should take account of cost control. Under the condition of cost control, we model each user's cost of helping spread information as a feature of each node in the network. Then we modify several most widely studied algorithms to suit the new model. In this paper, a new algorithm called CPP-SNS is proposed, which selects seeds according to cost performance of nodes. Further improvements, based on strategy of partial node loading and submodular property of spread function, make CPP-SNS more effective in practical scenarios. Extensive experiments show this method has a good performance in different social networks. Based on results of our research, we also provide some advice for the practical marketing.

GP-GPUs have been used as the platform for many applications due to their powerful computation ability and massively parallel features. In this paper, we first investigate the CSR sparse matrix format, the performance of existing optimized SpMV (Sparse matrix-vector multiplication) algorithms, and analyze the memory access patterns of the SpMV algorithms. Based on the analysis of the memory access patterns, we propose a new thread scheduling technique that can take advantage of inter-warp locality and intra-warp locality simultaneously, and also can achieve memory coalescing automatically. This proposed new scheduling technique will change the memory access pattern of SpMVs significantly. The simulation results show that the performance of the SpMV using the new proposed thread scheduling technique achieves much better performance than the implementation of the SpMV optimized by other techniques.

Most community detection algorithms are trying to obtain the global information of the network. But increasingly large scale of the current network makes accessing to global information very difficult. In the meanwhile, the network shows power-law distribution and sparse features. And local community mining algorithms which use these features have more advantages over global mining methods. In this paper, we proposed a local community detection algorithm based on the core members named LLCDA (Leader based Local Community Detecting Algorithm) which uses local structural information in the network to optimize a local objective function. A local community can be detected through continuous optimization of the function by expanding from an initial core member computed by a modified PageRank sorting algorithm. The proposed LLCDA algorithm has been tested on both synthetic and real world networks, and it has been compared with other community detecting algorithms. The experimental results validated our proposed LLCDA and showed that significant improvements have been achieved by this technique.

The issues of coalition formation have been investigated from many aspects and recently more and more attention has been paid to overlapping coalition formation. The (optimal) coalition structure generation(CSG) problem is one of the essential problems in coalition formation which is an important topic of cooperation in multiagent system. In this paper, we strictly define the coalition structure generation problem of overlapping extensions. Based on these definitions, we systematically prove some computational complexity results for overlapping coalition formation(OCF) games and threshold task games(TTGs). Moreover, dynamic programming and greedy approaches are adopted to solve the CSG for TTG.

The study of community detection has received more and more attention in recent years, the problem is very difficult and of great importance in many fields such as sociology, biology and computer science. But most of the algorithms proposed so far could not utilize the weight information within weighted networks, and many of them are so time-consuming that they are not fit for the large-scale networks. We propose a new center-based method, which is especially designed for weighted networks. And the method is also suitable for large-scale network because of its low computational complexity. We demonstrate our method on a synthetic network and two real-world networks. The result shows the high efficiency and precision of our method.

Detecting communities from networks has been given great attention these years. The traditional approaches were always focusing on the node community, while some recent studies have shown great advantage of link community approach which partitions links instead of nodes into communities. We proposed a novel algorithm LBLC (local based link community) to detect link communities in the network based on local information. A local link community can be detected by maximizing a local link fitness function from a seed link, which was ranked by another algorithm previously. The proposed LBLC algorithm has been tested on both synthetic and real world networks, and it has been compared with other link community detecting algorithm. The experimental results showed LBLC achieves significant improvement on link community structure.

Strategic ability updating relates to establishing some required properties, which can be expressed by strategic abilities, in a multicomponent reactive system. We model such a reactive system as a concurrent game structure (CGS), which is the semantic model of Alternating-time Temporal Logic (ATL). Then, we propose coalitional commitment as a tool for achieving the required strategic ability updating. Intuitively, a coalitional commitment can extend the state space of a CGS by a context function and then delete some transitions by a coalitional normative system (CNS). We propose coordinated ATL (co-ATL) for reasoning about strategic abilities in the structures obtained from a CGS by implementing a CNS. The model-checking problem for co-ATL is proved to be PTIME-complete, just like that of ATL, and is thus tractable. Then, we characterize the limitation of coalitional commitment power by identifying the set of co-ATL formulas whose satisfaction cannot be established and the set of co-ATL formulas whose satisfaction cannot be avoided. Afterward, we show that the effectiveness problem, feasibility problem, and synthesis problem for coalitional commitment are PTIME-complete, NP-complete, and FNP-complete, respectively. Finally, we treat the coalitional commitment synthesis problem as an extended planning problem and present an algorithm based on the planning as model checking paradigm. Our work can be seen as an improvement for both social law research and planning via model checking research.

Estimating agent's skill ratings from team competition results has many applications in the real world. Existing models assume skills are the same for all contexts. However, skills are context-sensitive in a variety of cases. In this paper, we present a Factor-Based Context-Sensitive Skill Rating System(FBCS-SRS). Instead of estimating agent skills under every context, which is hard due to data sparisity, we propose a factor model where individual skills are modelled by the inner product of an agent factor vector and a context factor vector. Collapsed Gibbs sampling is used for approximate inference. We formulate the problem of sampling linear constraint factors as a variant of MAX-SAT, and solve it by linear programming algorithms . We validate our model on two real-world datasets. Experiments show that FBCS-SRS achieves significantly higher prediction accuracy than other skill rating systems. The improvement is even more obvious when there are a lot of contexts.

Various attribute and relation information is used in social recommendation systems. However, previous approaches fail to use them in a unified way. In this paper, we propose a unified framework for social recommendation. Entities like users and items are described by their tags. We model each entity using topic models like Latent Dirichlet Allocation(LDA) and then connect these topic models to form a multi-relational network. Various relations between entities in recommender systems such as rating relation or user friend relation can be expressed as edges in the multi-relational network. We evaluate our model on a real-life dataset collected from a commercial recommender website. Experiments validate the generative performance and predictive performance of our model.

Based on the planning via model checking paradigm, we address the problem of coalitional planning in this paper. Informally, coalitional planning is the problem of planning for a subset of agents in a multi-agent system to force the whole multi-agent system to satisfy some goals. We use the language of ATL as the goal language and the semantic structure of ATL, i.e., concurrent game structure, to formalize the planning domain. We separate the concept of goal and planning object and use execution structures to interpret the goals. And then, we define a algorithm for coalitional planning and formally prove its correctness. Distinguished from the previous work, in coalitional planning all the ATL formulas can be considered as goals, thus the expressive power of ATL is sufficiently applied.

Point-based algorithms are a class of approximate methods for Partially Observable Markov Decision Processes (POMDPs). They do backup operators on a belief set only. This paper will propose a preprocessing method for point-based algorithms (PPBA). This method preprocesses each sampled belief point, and before generating alpha-vectors it estimates which action and alpha-vectors to be selected first, in so doing repeated computing is eliminated. Base-alpha is also defined in this paper, which cancels meaningless computing with sparseness of problem.

The imbalance of dataset will directly affect the precision of classifier. PC4.5, an improved C4.5 algorithm is proposed. The experiments in MIT dataset indicate that PC4.5 is effective on imbalanced dataset and the scale of the decision tree could be reduced.

This paper focuses on how to automatically determine the suitable clustering number in image segmentation and designs an algorithm of CANA using spectral clustering. Experiment results indicate that ACNA can provide superior performance. An application sample in image punching is introduced

An illumination invariant face recognition method based on neural network ensemble architecture is proposed. Given a face image with an arbitrary illumination direction, it can complete recognition in a uniform way with high performance without knowing or estimating the illumination direction. Experimental result shows that the recognition ratio of the ensemble architecture is higher than the conventional approach that uses a single neural network to recognize faces of a specific illumination direction.

This work presents a texture based image retrieval method, called RAH, in the frequency domain. In RAH, texture features are represented as power spectral histograms of images. RAH mainly consists of two parts: the extraction of texture features based on the histogram of image power spectral in the frequency domain and the similarity measurement function based on the texture features. The experimental results indicate that RAH outperforms the traditional cooccurrence matrix texture representation.