Search

This paper aims to predict a set of apps a user will open on her mobile device in the next time slot. Such an information is essential for many smartphone operations, e.g., app pre-loading and content pre-caching, to improve user experience. However, it is hard to build an explicit model that accurately captures the complex environment context and predicts a set of apps at one time. This paper presents a deep reinforcement learning framework, named as DeepAPP, which learns a model-free predictive neural network from historical app usage data. Meanwhile, an online updating strategy is designed to adapt the predictive network to the time-varying app usage behavior. To transform DeepAPP into a practical deep reinforcement learning system, several challenges are addressed by developing a context representation method for complex contextual environment, a general agent for overcoming data sparsity and a lightweight personalized agent for minimizing the prediction time. Extensive experiments on a large-scale anonymized app usage dataset reveal that DeepAPP provides high accuracy (precision 70.6% and recall of 62.4%) and reduces the prediction time of the state-of-the-art by 6.58 times. A field experiment of 29 participants demonstrates DeepAPP can effectively reduce launch time of apps.

Projection techniques are often used to visualize high-dimensional data, allowing users to better understand the overall structure of multi-dimensional spaces on a 2D screen. Although many such methods exist, comparably little work has been done on generalizable methods of inverse-projection -- the process of mapping the projected points, or more generally, the projection space back to the original high-dimensional space. In this paper we present NNInv, a deep learning technique with the ability to approximate the inverse of any projection or mapping. NNInv learns to reconstruct high-dimensional data from any arbitrary point on a 2D projection space, giving users the ability to interact with the learned high-dimensional representation in a visual analytics system. We provide an analysis of the parameter space of NNInv, and offer guidance in selecting these parameters. We extend validation of the effectiveness of NNInv through a series of quantitative and qualitative analyses. We then demonstrate the method's utility by applying it to three visualization tasks: interactive instance interpolation, classifier agreement, and gradient visualization.

Physiological studies have shown that there are some differences in speech and facial activities between depressive and healthy individuals. Based on this fact, we propose a novel Spatio-Temporal Attention (STA) network and a Multimodal Attention Feature Fusion (MAFF) strategy to obtain the multimodal representation of depression cues for predicting the individual depression level. Specifically, we firstly divide the speech amplitude spectrum/video into fixed-length segments and input these segments into the STA network, which not only integrates the spatial and temporal information through attention mechanism, but also emphasizes the audio/video frames related to depression detection. The audio/video segment-level feature is obtained from the output of the last full connection layer of the STA network. Secondly, this paper employs the eigen evolution pooling method to summarize the changes of each dimension of the audio/video segment-level features to aggregate them into the audio/video level feature. Thirdly, the multimodal representation with modal complementary information is generated using the MAFF and inputs into the support vector regression predictor for estimating depression severity. Experimental results on the AVEC2013 and AVEC2014 depression databases illustrate the effectiveness of our method.

As a class of context-aware systems, context-aware service recommendation aims to bind high-quality services to users while taking into account their context requirements, including invocation time, location, social profiles, connectivity, and so on. However, current CASR approaches are not scalable with the huge amount of service data (QoS and context information, users reviews and feedbacks). In addition, they lack a rich representation of contextual information as they adopt a simple matrix view. Moreover, current CASR approaches adopt the traditional user-service relation and they do not allow for multi-relational interactions between users and services in different contexts. To offer a scalable and context-sensitive service recommendation with great analysis and learning capabilities, we provide a rich and multi-relational representation of the CASR knowledge, based on the concept of knowledge graph. The constructed context-aware service knowledge graph (C-SKG) is, then, transformed into a low-dimentional vector space to facilitate its processing. For this purpose, we adopt Dilated Recurrent Neural Networks to propose a context-aware knowledge graph embedding, based on the principles of first-order and subgraph-aware proximity. Finally, a recommendation algorithm is defined to deliver the top-rated services according to the target user's context. Experiments have proved the accuracy and scalability of our CASR approach.

Multi-view clustering has attracted increasing attention in multimedia, machine learning and data mining communities. As one kind of the essential multi-view clustering algorithm, multi-view subspace clustering (MVSC) becomes more and more popular due to its strong ability to reveal the intrinsic low dimensional clustering structure hidden across views. Despite superior clustering performance in various applications, we observe that existing MVSC methods directly fuse multi-view information in the similarity level by merging noisy affinity matrices; and isolate the processes of affinity learning, multi-view information fusion and clustering. Both factors may cause insufficient utilization of multi-view information, leading to unsatisfying clustering performance. This paper proposes a novel consensus one-step multi-view subspace clustering (COMVSC) method to address these issues. Instead of directly fusing multiple affinity matrices, COMVSC optimally integrates discriminative partition-level information, which is helpful to eliminate noise among data. Moreover, the affinity matrices, consensus representation and final clustering labels matrix are learned simultaneously in a unified framework. By doing so, the three steps can negotiate with each other to best serve the clustering task, leading to improved performance. Accordingly, we propose an iterative algorithm to solve the resulting optimization problem. Extensive experiment results on benchmark datasets demonstrate the superiority of our method against other state-of-the-art approaches.

In this article, a frequency-domain approach is developed to deal with the global consensus problem for a class of general second-order multiagent systems (MASs) subject to actuator saturations. By employing the describing function and the generalized Nyquist criterion, the global consensus problem is thoroughly investigated for both undirected and directed topologies. First, the describing function is introduced to characterize the actuator saturations in the s -plane, and the inherent representation error is quantitatively analyzed from a frequency-domain perspective. Then, by means of the Kronecker product, the addressed consensus problem of the MAS is transformed into a corresponding stability analysis problem for a certain multi-input-multi-output (MIMO) system and, consequently, the generalized Nyquist criterion for MIMO systems is exploited to derive the condition for the global consensus of the MAS where the impact from the actuator saturation is explicitly reflected. Finally, numerical simulations are provided to illustrate the validity of the proposed theoretical result.

In most of the existing representation learning frameworks, the noise contaminating the data points is often assumed to be independent and identically distributed (i.i.d.), where the Gaussian distribution is often imposed. This assumption, though greatly simplifies the resulting representation problems, may not hold in many practical scenarios. For example, the noise in face representation is usually attributable to local variation, random occlusion, and unconstrained illumination, which is essentially structural, and hence, does not satisfy the i.i.d. property or the Gaussianity. In this article, we devise a generic noise model, referred to as independent and piecewise identically distributed (i.p.i.d.) model for robust presentation learning, where the statistical behavior of the underlying noise is characterized using a union of distributions. We demonstrate that our proposed i.p.i.d. model can better describe the complex noise encountered in practical scenarios and accommodate the traditional i.i.d. one as a special case. Assisted by the proposed noise model, we then develop a new information-theoretic learning framework for robust subspace representation through a novel minimum weighted error entropy criterion. Thanks to the superior modeling capability of the i.p.i.d. model, our proposed learning method achieves superior robustness against various types of noise. When applying our scheme to the subspace clustering and image recognition problems, we observe significant performance gains over the existing approaches.

As important side information, attributes have been widely exploited in the existing recommender system for better performance. Prior studies usually use a default value (i.e., ``other") to represent the missing attribute, resulting in sub-optimal performance. To address this problem, in this paper, we present an attribute-aware attentive graph convolution network. In particular, we first construct a graph, where users, items, and attributes are three types of nodes and their associations are edges. Thereafter, we leverage the graph convolution network to characterize the complicated interactions among . Furthermore, to learn the node representation, we adopt the message-passing strategy to aggregate the messages passed from the other directly linked types of nodes (e.g., a user or an attribute). Towards this end, we are capable of incorporating associate attributes to strengthen the user and item representation learning, and thus naturally solve the attribute missing problem. Given that for different users, the attributes of an item have different influence on their preference to this item, we design a novel attention mechanism to filter the message passed from an item to a target user by considering the attribute information. Extensive experiments on several publicly accessible datasets demonstrate the superiority of our model over several state-of-the-art methods.

Lately, variational autoencoder (VAE) has arisen as one of the most prevalent methods for unsupervised learning of intricate distributions. Despite being successful in deep feature extraction and uncertain data modelling, it still suffers from the instability and reconstruction error due to random sampling in the latent subspace representation of original input space. In this paper, to deal with those limitations, constrained VAE (CVAE) is proposed by utilizing input sample information. Enthused by parallel interaction mechanism between the ventral and dorsal stream of human brain in object recognition, parallel interaction spatial-temporal CVAE (PIST-CVAE) is proposed to extract spatial and temporal features from input samples. Lower dimensional nonlinear features extracted from PIST-CVAE are used to build the soft sensor. The effectiveness of CVAE and PIST-CVAE is demonstrated in an industrial case study, a polyester polymerization process.

The meaningful patterns embedded in high-dimensional multi-view data sets typically tend to have a much more compact representation that often lies close to a low-dimensional manifold. Identification of hidden structures in such data mainly depends on the proper modeling of the geometry of low-dimensional manifolds. In this regard, this article presents a manifold optimization-based integrative clustering algorithm for multi-view data. To identify consensus clusters, the algorithm constructs a joint graph Laplacian that contains denoised cluster information of the individual views. It optimizes a joint clustering objective while reducing the disagreement between the cluster structures conveyed by the joint and individual views. The optimization is performed alternatively over k -means and Stiefel manifolds. The Stiefel manifold helps to model the nonlinearities and differential clusters within the individual views, whereas k -means manifold tries to elucidate the best-fit joint cluster structure of the data. A gradient-based movement is performed separately on the manifold of each view so that individual nonlinearity is preserved while looking for shared cluster information. The convergence of the proposed algorithm is established over the manifold and asymptotic convergence bound is obtained to quantify theoretically how fast the sequence of iterates generated by the algorithm converges to an optimal solution. The integrative clustering on benchmark and multi-omics cancer data sets demonstrates that the proposed algorithm outperforms state-of-the-art multi-view clustering approaches.

In this paper, we address Emotion Recognition in Conversation (ERC) where conversational data are presented in a multimodal setting. Psychological evidence shows that self and inter-speaker influence are two central factors to emotion dynamics in conversation. State-of-the-art models do not effectively synthesise these two factors. Therefore, we propose an Adapted Dynamic Memory Network (A-DMN) where self and inter-speaker influences are modelled individually and further synthesised oriented towards the current utterance. Specifically, we model the dependency of the constituent utterances in a dialogue video using a global RNN to capture inter-speaker influence. Likewise, each speaker is assigned an RNN to capture their self influence. Afterwards, an Episodic Memory Module is devised to extract contexts for self and inter-speaker influence and synthesise them to update the memory. This process repeats itself for multiple passes until a refined representation is obtained and used for final prediction. Additionally, we explore cross-modal fusion in the context of multimodal ERC, and propose a convolution-based method which proves effective in extracting local interactions and computationally efficient. Extensive experiments demonstrate that A-DMN outperforms the state-of-the-art models on benchmark datasets.

The problem of representative selection amounts to sampling few informative exemplars from large datasets. Existing approaches to data selection often fall short of simultaneously handling non-linear data structures, sampling concise and non-redundant subsets, rejecting outliers, and yielding interpretable outcomes. This paper presents a novel representative selection approach, dubbed MOSAIC, for drawing descriptive sketches of arbitrary manifold structures. Resting upon a novel quadratic formulation, MOSAIC advances a multi-criteria selection approach that maximizes the global representation power of the sampled subset, ensures novelty of the samples by minimizing redundancy, and rejects disruptive information by effectively detecting outliers. Theoretical analyses shed light on geometrical characterization of the obtained sketch and reveal that the sampled representatives maximize a well-defined notion of data coverage in a transformed space. In addition, we present a highly scalable randomized implementation of the proposed algorithm shown to bring about substantial speedups. MOSAIC's superiority in achieving the desired characteristics of a representative subset all at once while exhibiting remarkable robustness to various outlier types is demonstrated via extensive experiments conducted on both real and synthetic data with comparisons to state-of-the-art algorithms.

This work presents a hybrid and hierarchical deep learning model for midterm load forecasting. The model combines exponential smoothing (ETS), advanced long short-term memory (LSTM), and ensembling. ETS extracts dynamically the main components of each individual time series and enables the model to learn their representation. Multilayer LSTM is equipped with dilated recurrent skip connections and a spatial shortcut path from lower layers to allow the model to better capture long-term seasonal relationships and ensure more efficient training. A common learning procedure for LSTM and ETS, with a penalized pinball loss, leads to simultaneous optimization of data representation and forecasting performance. In addition, ensembling at three levels ensures a powerful regularization. A simulation study performed on the monthly electricity demand time series for 35 European countries confirmed the high performance of the proposed model and its competitiveness with classical models such as ARIMA and ETS as well as state-of-the-art models based on machine learning.

Attributed networks are ubiquitous in the real world, such as social networks. Therefore, many researchers take the node attributes into consideration in the network representation learning to improve the downstream task performance. In this article, we mainly focus on an untouched ``oversmoothing'' problem in the research of the attributed network representation learning. Although the Laplacian smoothing has been applied by the state-of-the-art works to learn a more robust node representation, these works cannot adapt to the topological characteristics of different networks, thereby causing the new oversmoothing problem and reducing the performance on some networks. In contrast, we adopt a smoothing parameter that is evaluated from the topological characteristics of a specified network, such as small worldness or node convergency and, thus, can smooth the nodes' attribute and structure information adaptively and derive both robust and distinguishable node features for different networks. Moreover, we develop an integrated autoencoder to learn the node representation by reconstructing the combination of the smoothed structure and attribute information. By observation of extensive experiments, our approach can preserve the intrinsical information of networks more effectively than the state-of-the-art works on a number of benchmark datasets with very different topological characteristics.

In order to facilitate the representation of fuzzy relation matrix (FRM) models, a new system identification technique is proposed in this work to recognize the architecture and parameters of FRM models based on the semi-tensor product (STP) operation. Firstly, a fuzzy STP algorithm is defined for fuzzy inference. Secondly, a novel FRM framework is proposed for system parameter identification. Thirdly, the recognized FRM parameters are optimized to improve fuzzy system performance by the use of a hybrid training method based on the least squares estimator and the recursive Levenberg-Marquaedt algorithm. The effectiveness of the proposed structure and parameter identification techniques is verified by simulation of a multi-steps-ahead prediction modeling. Simulation results show that the proposed fuzzy STP technology is efficient for system identification, and the proposed matrix expression can be used to design multi-input multi-output (MIMO) systems with fuzzy FRM models.

Network embedding aims to learn the low-dimensional node representations for networks, which has attracted an increasing amount of attention in recent years. Most existing efforts in this field attempt to embed the network based on node similarity, which generally relies on edge existence statistics of the network. Instead of relying on the global edge existence statistics for every node pair, in this article, we utilize the information between a pair of nodes in a local way and propose a model, called node pair information preserving network embedding (NINE), based on adversarial networks. The main idea lies in preserving the node pair information (NI) by means of adversarial networks. The architecture of the proposed NINE model consists of three main components, namely: 1) NI embedder; 2) NI generator; and 3) NI discriminator. In the NI embedder, to avoid the complicated similarity calculation for a pair of nodes, the original NI vector calculated from the direct neighbor information of the two nodes is adopted as features, and the edge existence information is taken as labels to learn the embedded NI vector in a supervised learning manner. The second component is the NI generator, which takes the original node representation vectors of a node pair as input and outputs the generated NI vector. In order to make the generated NI vector follow the same distribution of the corresponding embedded NI vector, the generative adversarial network (GAN) is adopted, resulting in the third component, called the NI discriminator. Extensive experiments are conducted on seven real-world datasets in three downstream tasks, namely: 1) network reconstruction; 2) link prediction; and 3) node classification. Comparison results with seven state-of-the-art models demonstrate the effectiveness, efficiency, and rationality of our model.

Granular representations of crisp and fuzzy sets play an important role in rule induction algorithms based on rough set theory. In particular, arbitrary fuzzy sets can be approximated using unions of simple fuzzy sets called granules. These granules, in turn, have a straightforward interpretation in terms of human-readable fuzzy “if..., then...” rules. In this paper, we are considering a fuzzy rough set model based on ordered weighted average (OWA) aggregation over considered values. We show that this robust extension of the classical fuzzy rough set model, which has been applied successfully in various machine learning tasks, also allows for a granular representation. In particular, we prove that when approximations are defined using a directionally convex t-norm and its residual implicator, the OWA-based lower and upper approximations are definable as unions of fuzzy granules. This result has practical implications for rule induction from such fuzzy rough approximations.

Existing face anti-spoofing models using deep learning for multi-modality data suffer from low generalization in the case of using variety of presentation attacks such as 2D printing and high-precision 3D face masks. One of the main reasons is that the non-linearity of multi-spectral information used to preserve the intrinsic attributes between a real and a fake face are not well extracted. To address this issue, we propose a multi-modility data based two-stage cascade framework for face anti-spoofing. The proposed framework has two advantages. Firstly, we design a two-stage cascade architecture that can selectively fuse low-level and high-level features from different modalities to improve feature representation. Secondly, we use multi-modality data to construct a distance-free spectral on RGB and infrared (IR) to augment the non-linearity of data. The presented data fusion strategy is different from popular fusion approaches, since it can strengthen discrimination ability of network models on physical attribute features than identity structure features under certain constraints. In addition, a multi-scale patch based weighted fine-tuning strategy is designed to learn each specific local face region. Experimental results show that the proposed framework achieves better performance than other state-of-the-art methods on both benchmark datasets and self-established datasets, especially on multi-material masks spoofing.

Knowledge graphs are large graph-structured knowledge bases with incomplete or partial information. Numerous studies have focused on knowledge graph embedding to identify the embedded representation of entities and relations, thereby predicting missing relations between entities. Previous embedding models primarily regard (subject entity, relation, and object entity) triplet as translational distance or semantic matching in vector space. However, these models only learn a few expressive features and hard to handle complex relations, i.e., 1-to-N, N-to-1, and N-to-N, in knowledge graphs. To overcome these issues, we introduce a multi-scale dynamic convolutional network (M-DCN) model for knowledge graph embedding. This model features topnotch performance and an ability to generate richer and more expressive feature embeddings than its counterparts. The subject entity and relation embeddings in M-DCN are composed in an alternating pattern in the input layer, which helps extract additional feature interactions and increase the expressiveness. Multi-scale lters are generated in the convolution layer to learn different characteristics among input embeddings. Specically, the weights of these lters are dynamically related to each relation to model complex relations. The performance of M-DCN on the ve benchmark datasets is tested via experiments. Results show that the model can effectively handle complex relations and achieve state-of-the-art link prediction results on most evaluation metrics.

OWA operators were introduced by Yager and their Choquet integral-based representation was shown by Grabisch. Based on bi-capacities and related Choquet integral introduced by Grabisch and Labreuche, a new generalization of OWA operators, namely BIOWA operators are introduced. Our approach is exemplified by several examples. Bi-capacities leading to the standard Yager's OWA operators on real line are completely characterized. We introduce and exemplify quantifiers generating BIOWA operators with an arbitrary arity, and their orness. Finally, orness of BIOWA operators is also introduced and studied.

Aspect-based sentiment analysis (ABSA) is crucial for exploring user feedbacks and preferences on produces or services. Although numerous classical deep learning-based methods have been proposed in previous literature, several useful cues (e.g., contextual, lexical, and syntactic) are still not fully considered and utilized. In this study, a new approach for ABSA is proposed through the guidance of contextual, lexical, and syntactic cues. First, a novel sub-network is introduced to represent a target in a sentence in ABSA by considering the whole context. Second, lexicon embedding is applied to incorporate additional lexical cues. Third, a new attention module, namely, dependency attention, is proposed to elaborate syntactic dependency cues between words in attention inference. Experimental results on four benchmark data sets demonstrate the effectiveness of our proposed approach to aspect-based sentiment analysis.

The use of the Choquet integral in data fusion processes allows for the effective modelling of interactions and dependencies between data features or criteria. Its application requires identification of the defining capacity (also known as fuzzy measure) values. The main limiting factor is the complexity of the underlying parameter learning problem, which grows exponentially in the number of variables. However, in practice we may have expert knowledge regarding which of the subsets of criteria interact with each other, and which groups are independent. In this paper we study hierarchical aggregation processes, architecturally similar to feed-forward neural networks, but which allow for the simplification of the fitting problem both in terms of the number of variables and monotonicity constraints. We note that the Mobius representation lets us identify a number of relationships between the overall fuzzy measure and the data pipeline structure. Included in our findings are simplified fuzzy measures that generalise both k-intolerant and k-interactive capacities.

Remote sensing image scene classification has attracted great attention because of its wide applications. Although convolutional neural network (CNN)-based methods for scene classification have achieved excellent results, the large-scale variation of the features and objects in remote sensing images limits the further improvement of the classification performance. To address this issue, we present multiscale representation for scene classification, which is realized by a global-local two-stream architecture. This architecture has two branches of the global stream and local stream, which can individually extract the global features and local features from the whole image and the most important area. In order to locate the most important area in the whole image using only image-level labels, a weakly supervised key area detection strategy of structured key area localization (SKAL) is specially designed to connect the above two streams. To verify the effectiveness of the proposed SKAL-based two-stream architecture, we conduct comparative experiments based on three widely used CNN models, including AlexNet, GoogleNet, and ResNet18, on four public remote sensing image scene classification data sets, and achieve the state-of-the-art results on all the four data sets. Our codes are provided in https://github.com/hw2hwei/SKAL.

Knowledge Graphs (KGs) are the most widely used representation of structured information about a particular domain consisting of billions of facts in the form of entities (nodes) and relations (edges) between them. Besides, the KGs also encapsulate the semantic type information of the entities. The last two decades have witnessed a constant growth of KGs in various domains such as government, scholarly data, biomedical domains, etc. KGs have been used in Machine Learning based applications such as entity linking, question answering, recommender systems, etc. Open KGs are mostly heuristically created, automatically generated from heterogeneous resources such as text, images, etc., or are human-curated. However, these KGs are often incomplete, i.e., there are missing links between the entities and missing links between the entities and their corresponding entity types. This thesis focuses on addressing these two challenges of link prediction for Knowledge Graph Completion (KGC): \textbf{(i)} General Link Prediction in KGs that include head and tail prediction, triple classification, and \textbf{(ii)} Entity Type Prediction. Most of the graph mining algorithms are proven to be of high complexity, deterring their usage in KG-based applications. In recent years, KG embeddings have been trained to represent the entities and relations in the KG in a low-dimensional vector space preserving the graph structure. In most published works such as the translational models, convolutional models, semantic matching, etc., the triple information is used to generate the latent representation of the entities and relations. In this dissertation, it is argued that contextual information about the entities obtained from the random walks, and textual entity descriptions, are the keys to improving the latent representation of the entities for KGC. The experimental results show that the knowledge obtained from the context of the entities supports the hypothesis. Several methods have been proposed for KGC and their effectiveness is shown empirically in this thesis. Firstly, a novel multi-hop attentive KG embedding model MADLINK is proposed for Link Prediction. It considers the contextual information of the entities by using random walks as well as textual entity descriptions of the entities. Secondly, a novel architecture exploiting the information contained in a pre-trained contextual Neural Language Model (NLM) is proposed for Triple Classification. Thirdly, the limitations of the current state-of-the-art (SoTA) entity type prediction models have been analysed and a novel entity typing model CAT2Type is proposed that exploits the Wikipedia Categories which is one of the most under-treated features of the KGs. This model can also be used to predict missing types of unseen entities i.e., the newly added entities in the KG. Finally, another novel architecture GRAND is proposed to predict the missing entity types in KGs using multi-label, multi-class, and hierarchical classification by leveraging different strategic graph walks in the KGs. The extensive experiments and ablation studies show that all the proposed models outperform the current SoTA models and set new baselines for KGC. The proposed models establish that the NLMs and the contextual information of the entities in the KGs together with the different neural network architectures benefit KGC. The promising results and observations open up interesting scopes for future research involving exploiting the proposed models in domain-specific KGs such as scholarly data, biomedical data, etc. Furthermore, the link prediction model can be exploited as a base model for the entity alignment task as it considers the neighbourhood information of the entities.

False data injection attacks (FDIAs) have recently become a major threat to smart grids. Most of existing FDIAS detection methods focus on modeling temporal relationship of time-series measurement data, yet pay less attention to spatial relationship between bus/line measurement data, and fail to consider the relationship between sub-grids. To address these issues, we propose a sub-grid-oriented microservice framework by integrating a well-designed spatial-temporal neural network for FDIA detection in AC-model power systems. First, a welldesigned neural network is developed to model the spatialtemporal relationship of bus/line measurement for sub-grids. Then, a microservice-based supervising network is proposed to integrate the representation features obtained from sub-grids for the collaborative detection of FDIAs. To evaluate the proposed framework, three types of FDIAs datasets are generated based on a public benchmark power grid. Case studies on the FDIA datasets show that our method outperforms the state-of-the-art methods for FDIA detection in these datasets.

The deep learning based approaches which have been repeatedly proven to bring benefits to visual recognition tasks usually make a strong assumption that the training and test data are drawn from similar feature spaces and distributions. However, such an assumption may not always hold in various practical application scenarios on visual recognition tasks. Inspired by the hierarchical organization of deep feature representation that progressively leads to more abstract features at higher layers of representations, we propose to tackle this problem with a novel feature learning framework, which is called GMFAD, with better generalization capability in a multilayer perceptron manner. We first learn feature representations at the shallow layer where shareable underlying factors among domains (e.g., a subset of which could be relevant for each particular domain) can be explored. In particular, we propose to align the domain divergence between domain pair(s) by considering both inter-dimension and inter-sample correlations, which have been largely ignored by many cross-domain visual recognition methods. Subsequently, to learn more abstract information which could further benefit transferability, we propose to conduct feature disentanglement at the deep feature layer. Extensive experiments based on different visual recognition tasks demonstrate that our proposed framework can learn better transferable feature representation compared with state-of-the-art baselines. Nanyang Technological University This research was supported in part by the Wallenberg-NTU Presidential Postdoctoral Fellowship, the NTU-PKU Joint Research Institute, a collaboration between the Nanyang Technological University and Peking University that is sponsored by a donation from the Ng Teng Fong Charitable Foundation, the Science and Technology Foundation of Guangzhou Huangpu Development District under Grant 2017GH22 and 201902 010028, and Sino-Singapore International Joint Research Institute (Project No. 206-A017023 and 206-A018001).

Heterogeneous information network (HIN) embedding aims at learning the low-dimensional representation of nodes while preserving structure and semantics in an HIN. Existing methods mainly focus on static networks, while a real HIN usually evolves over time with the addition (deletion) of multiple types of nodes and edges. Because even a tiny change can influence the whole structure and semantics, the conventional HIN embedding methods need to be retrained to get the updated embeddings, which is time-consuming and unrealistic. In this paper, we investigate the problem of dynamic HIN embedding and propose a novel Dynamic HIN Embedding model (DyHNE) with meta-path based proximity. Specifically, we introduce the meta-path based first- and second-order proximities to preserve structure and semantics in HINs. As the HIN evolves over time, we naturally capture changes with the perturbation of meta-path augmented adjacency matrices. Thereafter, we learn the node embeddings by solving generalized eigenvalue problem effectively and employ eigenvalue perturbation to derive the updated embeddings efficiently without retraining. Experiments show that DyHNE outperforms the state-of-the-arts in terms of effectiveness and efficiency.

Early classification of ongoing alarm floods in industrial monitoring systems is crucial to provide a safe and efficient operation. In this article, a data-driven approach is proposed to address the early classification problem with unlabeled historical data. To prioritize earlier activated alarms and take advantage of the triggering time information of alarms, a vector representation called exponentially attenuated component (EAC) is used to represent alarm floods. This makes alarm sequences fit for different powerful machine learning algorithms. A method based on the time information of unlabeled historical alarm floods is formulated to determine the attenuation coefficient for EAC representation. With the Gaussian mixture model, an efficient semi-supervised approach is proposed to provide an early classification of alarm floods using unlabeled historical data. The efficiency of the proposed method is validated by the Tennessee Eastman process benchmark and a real industrial dataset.

Information retrieval is a difficult process due to the overabundance of information on the web. Nowadays, search result responds to user queries with too many results although only a few are relevant. Therefore, the existing clustering methods that fail in clustering snippets (short texts) of web documents due to the low frequencies of document terms should be deeply investigated. One of the approaches that can be used to solve this problem is the expansion of document terms with semantically similar terms. Hence, a list of terms with their closest and accurate semantically similar words (word representation) must be built. This study aims to design and develop a new framework to enhance the performance of web search result clustering (WSRC). The research also presents a new unsupervised distributed word representation scheme where each word is represented by a vector of its semantically related words; such as scheme expands snippets and user queries. The proposed framework consists of several activities, such as (1) various standard datasets (Open Directory Project [ODP]-239 and MORESQUE) that are used for evaluating search result clustering algorithms for most cited dataset works, (2) text pre-processing, (3) document representation based on a new wiki-based k-nearest neighbors (KNN) representation method, (4) effect of the proposed model on the performance of traditional clustering methods (k-means, k-medoids, single-linkage, and complete-linkage) for WSRC, and (5) evaluation stage of the proposed method. Results indicate that enhanced clustering methods, according to the new wiki-KNN based representation method in comparison with the baseline methods, show a significant improvement in WSRC. Furthermore, the new data representation scheme has enhanced the overall performance of clustering methods.

Understanding the interconnected relationships of large-scale information networks like social, scholar and Internet of Things networks is vital for tasks like recommendation and fraud detection. The vast majority of the real-world networks are inherently heterogeneous and dynamic, containing many different types of nodes and edges and can change drastically over time. The dynamicity and heterogeneity make it extremely challenging to reason about the network structure. Unfortunately, existing approaches are inadequate in modeling real-life networks as they require extensive computational resources and do not scale well to large, dynamically evolving networks. We introduce LIME, a better approach for modeling dynamic and heterogeneous information networks. LIME is designed to extract high-quality network representation with significantly lower memory resources and computational time over the state-of-the-art. Unlike prior work that uses a vector to encode each network node, we exploit the semantic relationships among network nodes to encode multiple nodes with similar semantics in shared vectors. We evaluate LIME by applying it to three representative network-based tasks, node classification, node clustering and anomaly detection, performing on three large-scale datasets. Our extensive experiments demonstrate that LIME not only reduces the memory footprint by over 80\% and computational time over 2x when learning network representation but also delivers comparable performance for downstream processing tasks.

The present paper is concerned with the solutions of first-order linear fuzzy differential equations where differentiability of fuzzy functions is defined under an assumption of linear correlation between fuzzy numbers. Motivated by the fact that the structure of the space of linearly correlated fuzzy numbers strongly depends on the symmetry of the basic fuzzy number, here we address first-order linear fuzzy differential equations by distinguishing whether the basic fuzzy number is symmetric or not. In the non-symmetric case, a first-order linear fuzzy differential equation may be transformed into an equivalent system of ordinary differential equations related to the representation functions of the linearly correlated fuzzy number-valued function. In the symmetric case, according to the monotonicity of the diameter of the fuzzy solution, a first-order linear fuzzy differential equation may be transformed into a system of ordinary differential equations associated with the representation functions of the canonical form of the linearly correlated fuzzy number-valued function. In addition, using our extension method one may obtain solutions with either increasing or decreasing diameters. Several examples are provided in order to illustrate the proposed method.

Omni-directional images are becoming more prevalent for understanding the scene of all directions around a camera, as they provide a much wider field-of-view (FoV) compared to conventional images. In this work, we present a novel approach to represent omni-directional images and suggest how to apply CNNs on the proposed image representation. The proposed image representation method utilizes a spherical polyhedron to reduce distortion introduced inevitably when sampling pixels on a non-Euclidean spherical surface around the camera center. To apply convolution operation on our representation of images, we stack the neighboring pixels on top of each pixel and multiply with trainable parameters. This approach enables us to apply the same CNN architectures used in conventional Euclidean 2D images on our proposed method in a straightforward manner. Compared to the previous work, we additionally compare different designs of kernels that can be applied to our proposed method. We also show that our method outperforms in monocular depth estimation task compared to other state-of-the-art representation methods of omni-directional images. In addition, we propose a novel method to fit bounding ellipses of arbitrary orientation using object detection networks and apply it to an omni-directional real-world human detection dataset.

Nonlinear process modeling is a primary task in intelligent manufacturing, aiming at extracting high-value features from massive process data for further process analysis like process monitoring. However, it is still a challenge to develop nonlinear process models with robust representation capability for diverse process faults. From the new perspective of the correlation between process variables, this paper develops a nonlinear process modeling algorithm to adaptively preserve the features of both global and local inter-variable structures, in order to fully exploit inter-variable features for enhancing the nonlinear representation of process operating conditions. Specifically, a unidimensional convolutional operation with a self-attention mechanism is proposed to simultaneously extract global and local inter-variable structures, wherein different attentions can be adaptively adjusted to these two structures for the final aggregation of them. Besides, cooperating with a two-dimensional dynamic data extension, the unidimensional convolutional operation can represent the overall temporal relationship between process samples. Through stacking a collection of these convolutional operations, a ResNet-style convolutional neural network then is constructed to extract high-order nonlinear features. Experiments on the Tennessee Eastman process validate the effectiveness of the proposed algorithm for two vital process monitoring problems-fault detection and fault identification.

This paper presents an observer-based control design approach for a class of nonlinear discrete-time systems. The model nonlinearities are handled in two ways: 1) a Takagi-Sugeno fuzzy representation is used for nonlinearities that depend on measured states, and 2) nonlinearities that depend on unmeasured states are kept in their original form and handled using a slope-bound condition. The observer-based controller design conditions are given as linear matrix inequalities. The approach we propose significantly improves results in the literature by providing less restrictive design conditions. These improvements are illustrated in a detailed analytical and numerical comparison on a synthetic example; while a pendulum-on-a-cart example shows that the approach works both in simulation and in real-time experiments.

This paper establishes the calculus for linearly correlated fuzzy number-valued functions. In particular, a definition of derivative is introduced by using representation functions and a linear isomorphism when the basic fuzzy number is non-symmetric. Our definition is more general than that leading to concept of Frechet derivative. Derivative and Riemann integral are introduced using the canonical form of a linearly correlated fuzzy number-valued function when the basic fuzzy number is symmetric. Some relevant properties of derivatives and integrals of linearly correlated fuzzy number-valued functions are further investigated for symmetric and non-symmetric basic fuzzy number, respectively.

Understanding customers’ energy consumption at the individual appliances level is crucial for the planning and im-plementation of demand response (DR) programs. The appliances’ usage profiles can be disaggregated from whole-house energy consumption data using non-intrusive load monitoring (NILM) methods. The appliance load patterns of each customer are considera-bly different, which make it challenging to train a model with strong generalization ability. In this paper, a novel methodology using transfer knowledge between domains for NILM is proposed. A temporal convolutional network is developed to learn the dynamic features of individual appliance load. A domain adaption loss is used to quantify the domain distribution discrepancy between source and target domain representation. By jointly optimizing domain adaptation and energy disaggregation, an invariant representation across domains for the individual appliance states can be learned. Data experiments on ground truth data validate the accuracy and the robustness of the proposed model, and demonstrate its superior transferability and application potential under those scenarios of data shortage.

Audio-visual matching aims to learn the intrinsic correspondence between image and audio clip. Existing works mainly concentrate on learning discriminative features, while ignore the cross-modal heterogeneous issue between audio and visual modalities. To deal with this issue, we propose a novel Adversarial-Metric Learning (AML) model for audio-visual matching. AML aims to generate a modality-independent representation for each person in each modality via adversarial learning, while simultaneously learns a robust similarity measure for cross-modality matching via metric learning. By integrating the discriminative modality-independent representation and robust cross-modality metric learning into an end-to-end trainable deep network, AML can overcome the heterogeneous issue with promising performance for audio-visual matching. Experiments on the various audio-visual learning tasks, including audio-visual matching, audio-visual verification and audio-visual retrieval on benchmark dataset demonstrate the effectiveness of the proposed AML model. The implementation codes are available on https://github.com/MLanHu/AML.

Nowadays, mining opinion or sentiment from the online user-generated texts has become a research hot-spot. Although, a large amount of lexicon-based Chinese polarity detection works have been done, yet the existing methods have one common flaw that even the same word can have opposite polarities among different seed-lexicons, or polarity fuzziness. Therefore, in order to further enhance the performance of Chinese sentiment polarity detection, in this paper, we start from a two-aspect lexicon expansion, specifically, detecting sentiment polarity for new words and revising sentiment polarity for words already defined in seed-lexicons, so that the polarity fuzziness can be avoided. Then, we formulate a novel sentiment polarity detection framework for Chinese (SPDFC) with more effort to fine-grained sentiment processing, involved in symmetrical mapping, sentiment feature pruning and text representation. In this way, the word polarity can be directly taken as features, further penetrating into polarity detection phase. According to the experimental results, compared to other classical and state-of-the-art methods, the proposed the framework SPDFC can achieve the best overall performance from the perspective of Chinese polarity detection, sentiment feature pruning as well as text representation.

Apparent personality and emotion analysis are both central to affective computing. Existing works solve them individually. In this paper we investigate if such high-level affect traits and their relationship can be jointly learned from face images in the wild. To this end, we introduce PersEmoN, an end-to-end trainable and deep Siamese-like network. It consists of two convolutional network branches, one for emotion and the other for apparent personality. Both networks share their bottom feature extraction module and are optimized within a multi-task learning framework. Emotion and personality networks are dedicated to their own annotated dataset. Furthermore, an adversarial-like loss function is employed to promote representation coherence among heterogeneous dataset sources. Based on this, we also explore the emotion-to-apparent-personality relationship. Extensive experiments demonstrate the effectiveness of PersEmoN., Accepted to IEEE Transactions on Affective Computing

This paper presents an H ∞ event-triggered state-feedback controller design for continuous-time nonlinear systems via convex optimization techniques. The proposal is based on an exact discretization and its quasi-linear parameter varying representation. Thus, two sets of design conditions in terms of linear matrix inequalities by means of both parameter-dependent and quadratic Lyapunov functions are proposed. The proposed conditions also provide an estimate to system’s domain of attraction and an extra set of conditions is presented for a guaranteed minimum time between events. Well-known examples are employed to illustrate the effectiveness of the proposal.

In this study, we present a novel framework called Community-based sentiment Extraction and NeTwork Embedding for POI Recommendation (CENTER) in order to suggesting impressive POIs in an effective and efficient fashion. The CENTER framework integrates two essential techniques: (1) a latent probabilistic generative model called Community-based Sentiment Extraction (CSE), which can accurately capture the review contents in LBSNs by taking into consideration the characteristics of social communities. The parameters of the CSE model are inferred effectively by the Gibbs sampling method. The primary sentiments are obtained based on the probability distribution of sentiments; (2) a network embedding model called Sentiment-aware Nework Embedding for POI Recommendation (SNER) is employed to learn the representation of important factors including POIs, users and textual sentiments in a low-dimensional embedding space. The joint training is utilized to alternatively sample all sets of edges in a heterogeneous network. Extensive experiments were conducted on two large-scale real datasets, in order to evaluate the performance of CENTER. The results demonstrate that CENTER outperforms the state-of-the-art baseline methods in the effectiveness and efficiency of POI suggestion.

Graph and subspace clustering methods have become the mainstream of multi-view clustering due to their promising performance. However, (1) since graph clustering methods learn graphs directly from the raw data, when the raw data is distorted by noise and outliers, their performance may seriously decrease; (2) subspace clustering methods use a “two-step” strategy to learn the representation and affinity matrix independently, and thus may fail to explore their high correlation. To address these issues, we propose a novel multi-view clustering method via learning a Low-Rank Tensor Graph (LRTG). Different from subspace clustering methods, LRTG simultaneously learns the representation and affinity matrix in a single step to preserve their correlation. We apply Tucker decomposition and l2;1-norm to the LRTG model to alleviate noise and outliers for learning a “clean” representation. LRTG then learns the affinity matrix from this “clean” representation. Additionally, an adaptive neighbor scheme is proposed to find the K largest entries of the affinity matrix to form a flexible graph for clustering. An effective optimization algorithm is designed to solve the LRTG model based on the alternating direction method of multipliers. Extensive experiments on different clustering tasks demonstrate the effectiveness and superiority of LRTG over seventeen state-of-the-art clustering methods.

Neighboring frames are more correlated compared to frames from further temporal distances. In this paper, we aim to explore the temporal correlations among neighboring frames and exploit cross-layer multi-scale features for action recognition. We present a Temporal Cross-Layer Correlation (TCLC) framework for temporal correlation learning. First, we introduce a context-aware reconstruction block to enable the exploration of neighboring context. This prediction block aims to reconstruct the past frame and the future frame in a unified way. It learns to mine frame correlations and aggregate long sequences at the same time. We demonstrate that this neighborhood mining process enhances the discriminative ability of the network. Second, we propose a novel cross-layer attention and a center-guided attention mechanism to integrate features with contextual knowledge from multiple scales. Our method is a two-stage process for effective cross-layer feature learning. The first stage incorporates the cross-layer attention module to decide the importance weight of the convolutional layers. The second stage leverages the center-guided attention mechanism to aggregate local features from each layer for the generation of a final video representation. We leverage global centers to extract shared semantic knowledge among videos. We evaluate TCLC on three action recognition datasets, i.e., UCF-101, HMDB-51 and Kinetics. Our experimental results demonstrate the superiority of our proposed temporal correlation mining method.

Various palmprint recognition methods have been proposed and applied in security, particularly authentication. However, improving the performance of palmprint recognition with insufficient training samples per person is still a challenging task. The undersampling problem limits the application and popularization of palmprint recognition. In this article, by regularly sampling different local regions of the palmprint image, we learn complete and discriminative convolution features by using deep convolutional neural networks (DCNNs). With this powerful palmprint description, a joint constrained least-square regression (JCLSR) framework, which performs representation for each local region of the same palmprint image requiring all regular local regions of the palmprint image to have similar projected target matrices, is presented to exploit the commonality of different patches. The proposed method can well solve the undersampling classification problem in palmprint recognition. Experiments were conducted on the IITD, CASIA, noisy IITD, and PolyU multispectral palmprint databases. It can be seen from the experimental results that the proposed JCLSR consistently outperformed the classical palmprint recognition methods and some subspace learning-based methods for palmprint recognition.

Anomaly detection suffers from unbalanced data since anomalies are quite rare. Synthetically generated anomalies are a solution to such ill or not fully defined data. However, synthesis requires an expressive representation to guarantee the quality of the generated data. In this paper, we propose a two-level hierarchical latent space representation that distills inliers' feature-descriptors (through autoencoders) into more robust representations based on a variational family of distributions (through a variational autoencoder) for zero-shot anomaly generation. From the learned latent distributions, we select those that lie on the outskirts of the training data as synthetic-outlier generators. And, we synthesize from them, i.e., generate negative samples without seen them before, to train binary classifiers. We found that the use of the proposed hierarchical structure for feature distillation and fusion creates robust and general representations that allow us to synthesize pseudo outlier samples. And in turn, train robust binary classifiers for true outlier detection (without the need for actual outliers during training). We demonstrate the performance of our proposal on several benchmarks for anomaly detection., To appear in IEEE Trans. on Neural Networks and Learning Systems

Dynamic network analysis is a promising research area with a wide range of applications. Criminal network investigation is one of them. People may be interested in using dynamic network analysis to detect criminal communities in a dynamic social network, track the evolution of those communities, identify critical criminal members, or predict links between criminal members and others. One difficulty in applying dynamic network analysis to real-world data is that real-world dynamic networks may vary sparse, which can cause overfitting problem and compromise the performance of the proposed model. Another problem is that each node in a complex real-world network may have multiple features, making it complicated to compute the distance between nodes. We propose a higher-order multiple-feature-based community evolution model (HFCE) to address those two issues. The model uses a higher-order representation of the neighbouring information among nodes to alleviate the sparsity problem. It also introduces first-order similarity regularization to clarify the distance between nodes with multiple features. Experiment results show that the HFCE model outperforms five other popular dynamic network models (ESPRA, AFECT, GenLouvain, ECD and DYNMOGA) in terms of community on the real-world sparse dataset detection and link prediction precision. The HFCE model can also effectively track the evolution of the communities and identify the important nodes in the network over time, which makes it a desirable model in criminal network investigation.

Time series data mining has received significant attention over the past decade, and many approaches have focused on classification tasks where the goal is to define the label of a test time series, given labeled training data. Time series classification approaches can be broadly grouped into two categories as instance-based and feature-based methods. Instance-based approaches utilize similarity information in a nearest-neighbor setting to classify time series data. Although approaches from this category provide accurate results, their performance degrades with long and noisy time series. On the other hand, feature-based approaches extract features to deal with the limitations of instance-based approaches; however, these approaches work with predefined features and may not be successful in certain classification problems. This study proposes a time series classification approach that benefits from both scale-space theory and bag-of-features technique. The method starts with finding the scale-space extrema points (i.e. key points) of each time series according to the SiZer (SIgnificant ZERo crossings of the derivatives) method, and then proceeds to create the local features set around these points. After extraction of the local features from each key point, a bag-of-features representation for each time series is constructed as the summary of the key point characteristics. We evaluate the success of the proposed representation on time series classification problems from various domains. Our experimental results show that our proposal provides competitive results compared to widely used approaches in the literature.

Emotion recognition using electroencephalogram (EEG) signals is getting more and more attention in recent years. Since the EEG signals are noisy, non-linear and have non-stationary properties, it is a challenging task to develop an intelligent framework that can provide high accuracy for emotion recognition. In this paper, we propose a new model for emotion recognition that will be based on the creation of feature maps based on the topographic (TOPO-FM) and holographic (HOLO-FM) representation of EEG signal characteristics. Deep learning has been utilized as a feature extractor method on feature maps, and afterward extracted features are fused together for the classification process to recognize different kinds of emotions. The experiments are conducted on the four publicly available emotion datasets: DEAP, SEED, DREAMER, and AMIGOS. We demonstrated the effectiveness of our approaches in comparison with studies where authors used EEG signals that classify human emotions in the two-dimensional space. Experimental results show that the proposed methods can improve the emotion recognition rate on the different size datasets.

Computational drug repositioning technology aims to rediscover the potential use of drugs already on the market and can significantly accelerate the traditional drug development process, reducing significant drug development costs and drug development instability In this work, in order to capture valid and robust hidden feature representations of drugs and diseases, we introduce a new computational drug relocation model, HSSIGNN, based on hybrid similarity side information powered graph neural network, by drawing on the application of graph neural networks and Side information in recommender systems. Its advantage is to utilize the learning capability of graph neural networks to capture the effective hidden feature representation of drugs and diseases, which is used to infer the probability of whether a drug can treat the disease of interest, as a way to improve the generalization capability of the model. In addition, dimensionality reduction algorithms and side information of drugs and diseases are used to overcome the cold start problem encountered by traditional computational drug relocation models. Finally, the experimental results of the proposed model on two real drug–disease association datasets are analyzed to verify its superiority and effectiveness. Comprehensive experimentations on several real-world datasets show the efficiency of HSSIGNN.

In recent years, knowledge graph representation learning has prompted extensive research. Machine learning models are used to map entity and relational data in knowledge graphs to vector representations in low-dimensional spaces to predict and analyze potential relationships. Current works mainly focus on the knowledge representation of the triple structure and relationship path in knowledge graphs without fully utilizing external textual information to semantically supplement knowledge representation. However, the existing knowledge inventory, such as that for smart health and emotion care systems, is relatively meager, and structural knowledge is incomplete; therefore, knowledge graph completion is essential. In this article, we propose a novel joint representation learning model that introduces text description information and extracts reliable feature information from text data by using a convolutional neural network (CNN) model. Furthermore, being based on the attention mechanism, the proposed model distinguishes the characteristic credibility of different relationships, enhances the representation of the entity relationship structure vector in the existing knowledge graph, and obtains rich semantic information. Finally, the 2-D convolution operation is used to process the joint representation vectors of entities and relationships to obtain nonlinear features, and the knowledge graph is completed by completing the calculation of the score function of the joint representation vector of the entity and the relationship. Experiments performing tasks, such as link prediction and triple classification, on the FreeBase (FB15k), WordNet (WN18) and Yet Another Great Ontology (YAGO3-10) data sets reveal that our model performs better than the benchmark model and has some degree of scalability.