Applying fact-based verification to online misinformation detection can be a complex task due to the challenges of varying input lengths and the lack of sufficient data to train fact-based verification models for misinformation detection. We collected and created two specific datasets consisting of news articles and a set of fact articles (fact pool) which are used to develop a novel online misinformation detection pipeline based on information retrieval and claim verification. In addition to the AveriTec dataset, we collected and published the Misbar 1 1 https://anonymous.4open.science/r/Misinformation-detection-using-automatic-fact-based-news-verification-65D3 . dataset, which consists of expert-annotated long news articles and verification articles. We also propose a novel eXplainable Misinformation Detection method with Fact Verification, called XMDFaVer, that detects online misinformation through several stages: text summarization to condense lengthy news articles into short claims, a question generation and answering model to clarify the claim with reference to the associated fact pool, and finally a classifier to assess the claim against the factual data and discern the authenticity of the news articles. XMDFaVer demonstrates strong and consistent performance across most test cases, outperforming standard and fact-based baselines. The results also show that using more fact articles improves detection accuracy, confirming the effectiveness and robustness of the approach.

Abstract Video-based emotion change estimation using spatiotemporal features is an emerging area of research, enabling various practical applications. Existing methods often overlook continual analysis of the changes in emotions during spontaneous conversations. In this study, considering a set of spatial and temporal attributes, a novel one-dimensional weighted emotion model is proposed to accurately scale the emotion intensity levels. Further, a tri-stream deep framework, called TriCe , is proposed to estimate the emotion changes using three complementary modal information from videos, namely key-frame, video and voice. In TriCe, each modal information is extracted separately and combined using an adaptive fusion technique, which is then used to recognize the emotion level. Experimental results on a newly constructed benchmark emotion dataset demonstrate that our TriCe framework is feasible in estimating emotion changes. Comparison with other existing works demonstrates that the TriCe sets the benchmark for emotion estimation.

Abstract Medical image analysis using computer-based algorithms has attracted considerable attention from the research community and achieved tremendous progress in the last decade. With recent advances in computing resources and availability of large-scale medical image datasets, many deep learning models have been developed for disease diagnosis from medical images. However, existing techniques focus on sub-tasks, e.g., disease classification and identification, individually, while there is a lack of a unified framework enabling multi-task diagnosis. Inspired by the capability of Vision Transformers in both patch-based and image-based representation learning, we propose in this paper a new method, namely Multi-task Vision Transformer (MVC) for simultaneously classifying chest X-ray images and identifying affected regions from the input data. Our method is built upon the Vision Transformer but extends its learning capability in a multi-task setting. We evaluated our proposed method and compared it with existing baselines on a benchmark dataset of COVID-19 chest X-ray images. Experimental results verified the superiority of the proposed method over the baselines on both the image classification and affected region identification tasks.

Investigating Answer Validation Using Noise Identification and Classification in Goal-Oriented Dialogues

Abstract Over the years, researchers have applied various deep learning techniques to automatically recognise plant diseases from both raster and spectral images. The primary focus of the existing studies is developing individual species-specific or disease-specific models, where the former recognises diseases of single crop type and the latter recognises single diseases of single or multiple crop types. Building one global model to recognise diseases of multiple crops has also been widely explored, where a class is treated as a crop-disease combination. While training individual species-specific or disease-specific deep models is labour-intensive, embracing a vast number of crop species and inherent diseases present on this planet makes the model cumbersome. In order to address this problem, a more intuitive and feasible family-based plant disease characterisation approach with botanical reasoning is proposed in this study. This approach demonstrates the feasibility of six state-of-the-art deep neural networks through a set of extensive experiments incorporating six key strategies. The results on a newly built family-based plant disease dataset confirm that the proposed novel approach is convincing to be applied in a plant family-based disease recognition problem. Further, this study creates future opportunities for more intuitive plant disease data collection and benchmark classification model development.

Federated Learning (FL) is a machine learning technique that enables multiple local clients holding individual datasets to collaboratively train a model, without exchanging the clients' datasets. Conventional FL approaches often assign a fixed workload (local epoch) and step size (learning rate) to the clients during the client-side local model training and utilize all collaborating trained models' parameters evenly during the server-side global model aggregation. Consequently, they frequently experience problems with data heterogeneity and high communication costs. In this paper, we propose a novel FL approach to mitigate the above problems. On the client side, we propose an adaptive model update approach that optimally allocates a needful number of local epochs and dynamically adjusts the learning rate to train the local model and regularizes the conventional objective function by adding a proximal term to it. On the server side, we propose a robust model aggregation strategy that potentially supplants the local outlier updates (models' weights) prior to the aggregation. We provide the theoretical convergence results and perform extensive experiments on different data setups over the MNIST, CIFAR-10, and Shakespeare datasets, which manifest that our FL scheme surpasses the baselines in terms of communication speedup, test-set performance, and global convergence.

Evaluating human dialogue is a complex task, as our conversation are never structured. There are, however, cases where there is some structure in our conversation, e.g., in a typical call center, dialogue between a call center agent and customer revolves around certain topics of conversation. These dialogues can be evaluated based on some pre-specified criteria as well as sub-criteria. This evaluation is typically done manually, which can be be time-consuming, motivating the need for an automated system that employs Artificial Intelligence (AI) algorithms to evaluate dialogues efficiently. In this paper, we have proposed a novel dialogue-evaluation framework that leverages recent advancements in deep learning research. The contributions of this work are two fold. Firstly, we introduce a straightforward end-to-end framework – CallAI, for evaluating dialogues in any domain, based on some predefined hierarchical criteria. Secondly, we present a novel algorithm – TAABLM, utilizing a novel combination of aspect-based learning along with traditional TF-IDF features for text. We show in this paper, that TAABLM outperforms conventional baselines such as BERT, LSTM, etc, delivering improved performance in automated dialogue evaluation, whereas CallAI offers a simple yet elegant framework for an AI-based solution to hierarchical dialogue evaluation. We demonstrate the efficacy of our proposed framework and proposed algorithm on three datasets, where we quantify performance in terms of an aggregated dialogue-score as well as in terms of either accuracy or AuROC metrics.

Interpretability often seeks domain-specific facts, which is understandable to human, from deep-learning (DL) or other machine-learning (ML) models of black-box nature. This is particularly important to establish transparency in ML model's inner-working and decision-making, so that a certain level of trust is achieved when a model is deployed in a sensitive and mission-critical context, such as health-care. Model-level transparency can be achieved when its components are transparent and are capable of explaining reason of a decision, for a given input, which can be linked to domain-knowledge. This article used convolutional neural network (CNN), with sinc-convolution as its constrained first-layer, to explore if such a model's decision-making can be explained, for a given task, by observing the sinc-convolution's sinc-kernels. These kernels work like band-pass filters, having only two parameters per kernel - lower and upper cutoff frequencies, and optimised through back-propagation. The optimised frequency-bands of sinc-kernels may provide domain-specific insights for a given task. For a given input instance, the effects of sinc-kernels was visualised by means of explanation vector, which may help to identify comparatively significant frequency-bands, that may provide domain-specific interpretation, for the given task. In addition, a CNN model was further optimised by considering the identified subset of prominent sinc frequency-bands as the constrained first-layer, which yielded comparable or better performance, as compared to its all sinc-bands counterpart, as well as, a classical CNN. A minimal CNN structure, achieved through such an optimisation process, may help design task-specific interpretable models. To the best of our knowledge, the idea of sinc-convolution layer's task-specific significant sinc-kernel-based network optimisation is the first of its kind. Additionally, the idea of explanation-vector-based joint time-frequency representation to analyse time-series signals is rare in the literature. The above concept was validated for two tasks, ECG beat-classification (five-class classification task), and R-peak localisation (sample-wise segmentation task).

Deep-learning-based QRS-detection algorithms often require essential post-processing to refine the output prediction-stream for R-peak localisation. The post-processing involves basic signal-processing tasks including the removal of random noise in the model's prediction stream using a basic Salt and Pepper filter, as well as, tasks that use domain-specific thresholds, including a minimum QRS size, and a minimum or maximum R-R distance. These thresholds were found to vary among QRS-detection studies and empirically determined for the target dataset, which may have implications if the target dataset differs such as the drop of performance in unknown test datasets. Moreover, these studies, in general, fail to identify the relative strengths of deep-learning models and the post-processing to weigh them appropriately. This study identifies the domain-specific post-processing, as found in the QRS-detection literature, as three steps based on the required domain knowledge. It was found that the use of minimal domain-specific post-processing is often sufficient for most of the cases and the use of additional domain-specific refinement ensures superior performance, however, it makes the process biased towards the training data and lacks generalisability. As a remedy, a domain-agnostic automated post-processing is introduced where a separate recurrent neural network (RNN)-based model learns required post-processing from the output generated from a QRS-segmenting deep learning model, which is, to the best of our knowledge, the first of its kind. The RNN-based post-processing shows superiority over the domain-specific post-processing for most of the cases (with shallow variants of the QRS-segmenting model and datasets like TWADB) and lags behind for others but with a small margin ( ≤ 2%). The consistency of the RNN-based post-processor is an important characteristic which can be utilised in designing a stable and domain agnostic QRS detector.

Existing Federated Learning (FL) algorithms generally suffer from high communication costs and data heterogeneity due to the use of conventional loss function for local model update and the equal consideration of each local model for global model aggregation. In this paper, we propose a novel FL approach to address the above issues. For local model update, we propose a disentangled Information Bottleneck (IB) principle-based loss function. For global model aggregation, we suggest a model selection strategy based on Mutual Information (MI). Particularly, we design a Lagrangian-based loss function using the IB principle and “disentanglement” for maximizing MI between the ground truth and model prediction and minimizing MI between the intermediate representations. We calculate MI ratio between the ground truth and model prediction, and between the original input and ground truth to select the effective models for aggregation. We analyze the theoretical optimal cost of the loss function and manifest optimal convergence rate, and quantify the outlier robustness of the aggregation scheme. Experiments demonstrate the superiority of the proposed FL approach, in terms of testing performance and communication speedup (i.e., 3.00-14.88 times for IID MNIST, 2.5-50.75 times for non-IID MNIST, 1.87-18.40 times for IID CIFAR-10, and 1.24-2.10 times for non-IID MIMIC-III).

Summary With the development of internet technologies, social media and mobile devices, short texts have become an increasingly popular medium among users to communicate with friends, search information and review products. Measuring the similarity between short texts is a fundamental task due to its importance in many applications, such as text retrieval, topic discovery, and event detection. However, short texts generally comprise sparse, noisy, and ambiguous information. Hence, effectively measuring the distance between short texts is a challenging task. In this paper, we exploit the advantageous corpus‐wide word co‐occurrence information into document‐level feature enrichment to mitigate the challenges caused by the sparseness of short texts for distance measurement. We propose a novel context‐aware weighted Biterm method for short text Distance Measurement (BDM). In BDM, we extract biterms (ie, word pairs) from a short text corpus and exploit a biterm topic model to determine the global weights of biterms in the corpus. We then determine the local importance of a biterm in different contexts (ie, short texts) based on the corpus‐level biterm weight. The distance between two short texts is computed using the context‐aware weighted biterms. Experimental results on three real‐world datasets demonstrate better accuracy and effectiveness of the proposed BDM.

Federated Averaging (FedAvg) is the baseline Federated Learning (FL) algorithm that applies the stochastic gradient descent for local model training and the arithmetic averaging of the local models’ parameters for global model aggregation. Succeeding FL works commonly utilize the arithmetic averaging scheme of FedAvg for the aggregation. However, such arithmetic averaging is prone to the outlier model-updates, especially when the clients’ data are non-Independent and Identically Distributed (non-IID). As such, the classical aggregation approach suffers from the dominance of the outlier updates and, consequently, causes high communication costs towards producing a decent global model. In this letter, we propose a robust aggregation strategy to alleviate the above issues. In particular, we propose first pruning the node-wise outlier updates (weights) from the local trained models and then performing the aggregation on the selected effective weights-set at each node. We provide the theoretical result of our method and conduct extensive experiments on the MNIST, CIFAR-10, and Shakespeare datasets with IID and non-IID settings, which demonstrate that our aggregation approach outperforms the state-of-the-art methods in terms of communication speedup, test-set performance and training convergence.

Understanding anunknown facial emotionthat emerges in the future underpins significant impacts in various domains. Knowing the fact that emotional states grow in vocabulary, new emotional states need to be adapted while the existing knowledge of known emotional states is preserved. While human beings spontaneously perform this task, the challenge is, how to devise a deep learning technique that can effectively recognize an unknown emotion category in the future. Although the deep convolutional neural network has shown excellent emotion recognition performances in the past, it is conventionally a predefined multi-way classifier showing little resilience towards adding a new emotion class. Considering the aforementioned challenge, in this paper, we propose a generic deep convolutional neural network-based architecture that constantly absorbs the upcoming emotion categories and recognizes them effectively. We further propose an indicator loss, which is associated with the distillation mechanism that preserves the existing knowledge. In order to demonstrate the feasibility of our proposed method, we evaluated our model using benchmark emotion datasets. The results confirm that the proposed approach is superior in recognizing unknown emotional states compared to continual learning benchmarks. Further, our proposed method demonstrates higher accuracy, compared to the transfer learning baselines.

Recognizing emotions in context from an image has become an emerging topic in the recent past due to its high practical demand in various domains. Performance of the existing works is limited as they predominantly utilized the entire image and the primary human subject as major cues to recognize the emotions from images. However, in addition to the primary human subject, other human subjects in a scene also play a vital role in determining the image's overall emotional state. In this work, a novel visual feature type is introduced based on the features extracted from other human subjects in order to enhance emotion recognition performance. A novel deep learning based hybrid framework is also proposed to effectively integrate the proposed feature with the visual cues from the entire image and primary human subject. The extensive experiments carried out on a subset of the Emotic dataset reveal that the newly introduced visual feature type contributes to the overall emotional state of the scene. Furthermore, the proposed framework achieved 45.11% average accuracy for discrete emotion categories, which is a significant improvement in the emotion recognition performance compared to existing techniques that use emotion information from the entire image and primary human subject.

Federated Learning (FL) is an emerging research field that yields a global trained model from different local clients without violating data privacy. Existing FL techniques often ignore the effective distinction between local models and the aggregated global model when doing the client-side weight update, as well as the distinction of local models for the server-side aggregation. In this article, we propose a novel FL approach with resorting to mutual information (MI). Specifically, in client-side, the weight update is reformulated through minimizing the MI between local and aggregated models and employing Negative Correlation Learning (NCL) strategy. In server-side, we select top effective models for aggregation based on the MI between an individual local model and its previous aggregated model. We also theoretically prove the convergence of our algorithm. Experiments conducted on MNIST, CIFAR-10, ImageNet, and the clinical MIMIC-III datasets manifest that our method outperforms the state-of-the-art techniques in terms of both communication and testing performance.

User engagement has recently received significant attention in understanding decay and expansion of communities in social networks. However, the problem of user engagement hasn't been fully explored in terms of users' specific interests and structural cohesiveness altogether. Therefore, we fill the gap by investigating the problem of community engagement from the perspective of attributed communities. Given a set of keywords W, a structure cohesive parameter k, and a budget parameter l, our objective is to find l number of users who can induce a maximal expanded community. Meanwhile, every community member must contain the given keywords in W and the community should meet the specified structure cohesiveness constraint k. We introduce this problem as best-Anchored Vertex set Exploration (AVE).To solve the AVE problem, we develop a Filter-Verify framework by maintaining the intermediate results using multiway tree, and probe the best anchored users in a best search way. To accelerate the efficiency, we further design a keyword-aware anchored and follower index, and also develop an index-based efficient algorithm. The proposed algorithm can greatly reduce the cost of computing anchored users and their followers. Additionally, we present two bound properties that can guarantee the correctness of our solution. Finally, we demonstrate the efficiency of our proposed algorithms and index. We measure the effectiveness of attributed community-based community engagement model by conducting extensive experiments on five real-world datasets.

This paper proposes a Mixed-Integer Linear Programming (MILP) formulation for Soft Graph Clustering that can be applied to both weighted and unweighted graphs and is polynomial in size. It can provide proven optimal solutions or solutions with proven optimality gap. The solutions will partition the set of vertices in a graph into K (possibly overlapping) clusters, for K predetermined. The MILP approach can also simultaneously allocate membership proportion for vertices that lie in multiple clusters, and handle restrictions such as cardinality constraints on the size of the overlaps and equal balance of cluster memberships. Our approach requires neither a threshold on edge weights to be predetermined nor the finding of clique neighbourhoods of any size. It does not require the maximum number of clusters to be predetermined. We compare our approach numerically with four existing algorithms: one that requires the finding of clique neighbourhoods, one that performs well on weighted graphs only, one that requires the maximum number of clusters be predetermined, and one that is statistical analysis-based; and discuss in what way our approach outperform these methods.

Detecting emotional states of human from videos is essential in order to automate the process of profiling human behaviour, which has applications in a variety of domains, such as social, medical and behavioural science. Considerable research has been carried out for binary classification of emotions using facial expressions. However, a challenge exists to automate the feature extraction process to recognise the various intensities or levels of emotions. The intensity information of emotions is essential for tasks such as sentiment analysis. In this work, we propose a metric-based intensity estimation mechanism for primary emotions, and a deep hybrid convolutional neural network-based approach to recognise the defined intensities of the primary emotions from spontaneous and posed sequences. Further, we extend the intensity estimation approach to detect the basic emotions. The frame level facial action coding system annotations and the intensities of action units associated with each primary emotion are considered for deriving the various intensity levels of emotions. The evaluation on benchmark datasets demonstrates that our proposed approach is capable of correctly classifying the various intensity levels of emotions as well as detecting them.

Summary In a data sharing group, each user can upload, modify, and access group files and a user is required to generate a new signature for the modified file after modification. There is a situation that two or more users modify the same file at almost the same time, which should be avoided as it gives rise to a signature conflict. However, the existing schemes do not take it into consideration. In this paper, we proposed a new mechanism SeShare for data storing based on blockchain to realize signature uniqueness, which solves the problem of generating signatures for the same file meanwhile by different group users. Specifically, we record every signature of a file in a blockchain in chronological order, and only one user is allowed to add new signature at the end of the blockchain when modification conflicts occur. On the other hand, to provide a secure data sharing service, SeShare introduces an efficient public auditing scheme for file integrity verification when a group user leaves the group. We also prove the security of the proposed scheme and evaluate the performance at the end of this paper. Our experimental results demonstrate the efficiency of public auditing for user leaving.

In data sharing schemes, the file owners should obtain rewards by sharing files with others as they put energy in these files. Therefore, we proposed an incentive data sharing scheme in this paper which encourages users to share data and also supports customization. Customization allows the owners to decide the threshold of access, the importance of each attributive classification which determines users' priority level of file modification and file ownership obtaining when the original owner leaves according to the priority level value. To support a convincing customized data sharing scheme, we introduce the knowledge of blockchain and construct a suitable access structure based on weighted attributes. The blockchain is used to ensure the fairness in incentive. Based on weighted attributes, an attribute set is disposed to a numerical value and the owner of the attribute set is able to obtain the file when the value is not less than the threshold, which is different from the normal access control policy. We prove the security from integrity, privacy and the availability of access key. The performance of the proposed scheme is evaluated at the end of this paper.

This letter proposes an effective method to improve object location in Chamfer template matching (CTM) based object detection using image segmentation. In our method, object bounding boxes are iteratively adjusted to fit with the object images obtained from image segmentation in a probabilistic model. The proposed method was tested with state-of-the-art CTM-based object detectors. Experimental results have shown the proposed method improved the location accuracy of the object detectors and reduce the false alarms rate.

Summary Malware replicates itself and produces offspring with the same characteristics but different signatures by using code obfuscation techniques. Current generation anti‐virus engines employ a signature‐template type detection approach where malware can easily evade existing signatures in the database. This reduces the capability of current anti‐virus engines in detecting malware. In this paper, we propose a stepwise binary logistic regression‐based dimensionality reduction techniques for malware detection using application program interface (API) call statistics. Finding the most significant malware feature using traditional wrapper‐based approaches takes an exponential complexity of the dimension (m) of the dataset with a brute‐force search strategies and order of (m‐1) complexity with a backward elimination filter heuristics. The novelty of the proposed approach is that it finds the worst case computational complexity which is less than order of (m‐1). The proposed approach uses multi‐linear regression and thep‐value of each individual API feature for selection of the most uncorrelated and significant features in order to reduce the dimensionality of the large malware data and to ensure the absence of multi‐collinearity. The stepwise logistic regression approach is then employed to test the significance of the individual malware feature based on their corresponding Wald statistic and to construct the binary decision the model. When the selected most significant APIs are used in a decision rule generation systems, this approach not only reduces the tree size but also improves classification performance. Exhaustive experiments on a large malware data set show that the proposed approach clearly exceeds the existing standard decision rule, support vector machine‐based template approach with complete data and provides a better statistical fitness. Copyright © 2016 John Wiley & Sons, Ltd.

A multilayer watermarking system is a system that is able to embed watermarks to a host media signal repeatedly in an overlaying manner, without incurring troubles in extracting the watermarks in each layer. In this paper, we present a novel patchwork-based audio watermarking algorithm that can embed and extract watermark bits successfully in such a multilayer framework. In the proposed method, a new watermark embedding algorithm is designed to ensure that the embedded watermarks in a certain layer do not affect the detection of watermarks in other layers. Adding multiple layers of watermark bits inevitably reduces the perceptual quality. However, to minimize the perceptual quality degradation in multilayer watermarking, the audio fragments for watermark embedding are selected from a set of specially arranged discrete cosine transform coefficients of the host audio signal. Watermark embedding is achieved by modifying the mean values of selected sample fragments. With the use of an embedding error buffer, the proposed system can withstand a wide range of common attacks. To maintain the balance between the perceptual quality and robustness, watermark embedding strength is adjusted according to the specific layer used. The proposed multilayer scheme ensures the independence of the processing in different layers. The effectiveness of the proposed system is demonstrated and verified by extensive simulation results.

Evolutionary algorithms (EAs) have recently been suggested as a candidate for solving big data optimization problems that involve a very large number of variables and need to be analyzed in a short period of time. However, EAs face a scalability issue when dealing with big data problems. Moreover, the performance of EAs critically hinges on the utilized parameter values and operator types, thus it is impossible to design a single EA that can outperform all others in every problem instance. To address these challenges, we propose a heterogeneous framework that integrates a cooperative co-evolution method with various types of memetic algorithms. We use the cooperative co-evolution method to split the big problem into subproblems in order to increase the efficiency of the solving process. The subproblems are then solved using various heterogeneous memetic algorithms. The proposed heterogeneous framework adaptively assigns, for each solution, different operators, parameter values and a local search algorithm to efficiently explore and exploit the search space of the given problem instance. The performance of the proposed algorithm is assessed using the Big Data 2015 competition benchmark problems that contain data with and without noise. Experimental results demonstrate that the proposed algorithm, with the cooperative co-evolution method, performs better than without the cooperative co-evolution method. Furthermore, it obtained very competitive results for all tested instances, if not better, when compared to other algorithms using lower computational times.

Offensive Internet chats, particularly the child‐exploiting type, tend to follow a documented psychological behavioral pattern. Researchers have identified some important stages in this pattern. The psychological stages broadly include befriending, information exchange, grooming, and approach. Similarities among the posts of a chat play an important role in differentiating as well as in identifying these stages. In this article a novel similarity measure is constructed which gives high Inter‐post‐similarity among the chat‐posts within a particular behavioral stage and low inter‐post‐similarity across different behavioral stages. A psychological stage corpus‐based dictionary is constructed from mining the terms associated with each stage. The dictionary works as a background knowledge‐base to support the similarity measure. To find the inter‐post similarity a modified sentence similarity measure is used. The proposed measure gives improved recognition of inter‐stage and intra‐stage similarity among the chat posts compared with other types of similarity measures. The pairwise inter‐post similarity is used for clustering chat‐posts into the psychological stages. Results of experiments demonstrate that the new clustering method gives better results than some current clustering methods.

Big data analytics and business analytics are disruptive technology and innovative solution for enterprise development. However, what is the relationship between big data analytics and business analytics? What is the relationship between business analytics and enterprise information systems (EIS)? How can business analytics enhance the development of EIS? These are still big issues for EIS development. This paper addresses these three issues by proposing an ontology of business analytics, presenting an analytics service-oriented architecture (ASOA) and applying ASOA to EIS, where our surveyed data analysis showed that the proposed ASOA can enhance to develop EIS. This paper also discusses the interrelationship between data analysis and business analytics, and between data analytics and big data analytics. The proposed approaches in this paper will facilitate research and development of EIS, business analytics, big data analytics, and business intelligence.

Loss function plays an important role in data classification. Manyloss functions have been proposed and applied to differentclassification problems. This paper proposes a new so called thesmoothed 0-1 loss function, that could be considered as anapproximation of the classical 0-1 loss function. Due to thenon-convexity property of the proposed loss function, globaloptimization methods are required to solve the correspondingoptimization problems. Together with the proposed loss function, wecompare the performance of several existing loss functions in theclassification of noisy data sets. In this comparison, differentoptimization problems are considered in regards to the convexity andsmoothness of different loss functions. The experimental resultsshow that the proposed smoothed 0-1 loss function works better ondata sets with noisy labels, noisy features, and outliers.

The expectation maximization (EM) is the standard training algorithm for hidden Markov model (HMM). However, EM faces a local convergence problem in HMM estimation. This paper attempts to overcome this problem of EM and proposes hybrid metaheuristic approaches to EM for HMM. In our earlier research, a hybrid of a constraint-based evolutionary learning approach to EM (CEL-EM) improved HMM estimation. In this paper, we propose a hybrid simulated annealing stochastic version of EM (SASEM) that combines simulated annealing (SA) with EM. The novelty of our approach is that we develop a mathematical reformulation of HMM estimation by introducing a stochastic step between the EM steps and combine SA with EM to provide better control over the acceptance of stochastic and EM steps for better HMM estimation. We also extend our earlier work and propose a second hybrid which is a combination of an EA and the proposed SASEM, (EA-SASEM). The proposed EA-SASEM uses the best constraint-based EA strategies from CEL-EM and stochastic reformulation of HMM. The complementary properties of EA and SA and stochastic reformulation of HMM of SASEM provide EA-SASEM with sufficient potential to find better estimation for HMM. To the best of our knowledge, this type of hybridization and mathematical reformulation have not been explored in the context of EM and HMM training. The proposed approaches have been evaluated through comprehensive experiments to justify their effectiveness in signal modeling using the speech corpus: TIMIT. Experimental results show that proposed approaches obtain higher recognition accuracies than the EM algorithm and CEL-EM as well.

This article investigates a novel machine learning approach applying consensus clustering in conjunction with classification for the data mining of very large and highly dimensional ECG data sets. To obtain robust and stable clusterings, consensus functions can be applied for clustering ensembles combining a multitude of independent initial clusterings. Direct applications of consensus functions to highly dimensional ECG data sets remain computationally expensive and impracticable. We introduce a multistage scheme including various procedures for dimensionality reduction, consensus clustering of randomized samples, followed by the use of a fast supervised classification algorithm. Applying the Hybrid Bipartite Graph Formulation combined with rank ordering and SMO we obtained an area under the receiver operating curve of 0.987. The performance of the classification algorithm at the final stage is crucial for the effectiveness of this technique. It can be regarded as an indication of the reliability, quality and stability of the combined consensus clustering.

This study empirically evaluates the effectiveness of different feature types for the classification of the first language of an author. In particular, it examines the utility of psycholinguistic features, extracted by the Linguistic Inquiry and Word Count (LIWC) tool, that have not previously been applied to the task of author profiling. As LIWC is a tool that has been developed in the psycholinguistic field rather than the computational linguistics field, it was hypothesized that it would be effective, both as a single type feature set because of its psycholinguistic basis, and in combination with other feature sets, because it should be sufficiently different to add insight rather than redundancy. It was found that LIWC features were competitive with previously used feature types in identifying the first language of an author, and that combined feature sets including LIWC features consistently showed better accuracy rates and average F measures than were achieved by the same feature sets without the LIWC features. As a secondary issue, this study also examined how effectively first language classification scaled up to a larger number of possible languages. It was found that the classification scheme scaled up effectively to the entire 16 language collection from the International Corpus of Learner English, when compared with results achieved on just 5 languages in previous research.

Robocup is a popular test bed for AI programs around the world. Robosoccer is one of the two major parts of Robocup, in which AIBO entertainment robots take part in the middle sized soccer event. The three key challenges that robots need to face in this event are manoeuvrability, image recognition and decision making skills. This paper focuses on the decision making problem in Robosoccer -- The goal keeper problem. We investigate whether reinforcement learning (RL) as a form of semi-supervised learning can effectively contribute to the goal keeper's decision making process when penalty shot and two attacker problem are considered. Currently, the decision making process in Robosoccer is carried out using rule-base system. RL also is used for quadruped locomotion and navigation purpose in Robosoccer using AIBO. In this paper, we propose a reinforcement learning based approach that uses a dynamic state-action mapping using back propagation of reward and space quantized Q-learning (SQQL) for the choice of high level functions in order to save the goal. The novelty of our approach is that the agent learns while playing and can take independent decision which overcomes the limitations of rule-base system due to fixed and limited predefined decision rules. Performance of the proposed method has been verified against the bench mark data set made with Upenn'03 code logic. It was found that the efficiency of our SQQL approach in goalkeeping was better than the rule based approach. The SQQL develops a semi-supervised learning process over the rule-base system's input-output mapping process, given in the Upenn'03 code.

In this paper, a novel method for profiling phishing activity from an analysis of phishing emails is proposed. Profiling is useful in determining the activity of an individual or a particular group of phishers. Work in the area of phishing is usually aimed at detection of phishing emails. In this paper, we concentrate on profiling as distinct from detection of phishing emails. We formulate the profiling problem as a multi-label classification problem using the hyperlinks in the phishing emails as features and structural properties of emails along with who is (i.e. DNS) information on hyperlinks as profile classes. Further, we generate profiles based on classifier predictions. Thus, classes become elements of profiles. We employ a boosting algorithm (AdaBoost) as well as SVM to generate multi-label class predictions on three different datasets created from hyperlink information in phishing emails. These predictions are further utilized to generate complete profiles of these emails. Results show that profiling can be done with quite high accuracy using hyperlink information.

Discovery of interesting rules describing the behavioural patterns of smokers' quitting intentions is an important task in the determination of an effective tobacco control strategy. In this paper, we investigate a compact and simplified rule discovery process for predicting smokers' quitting behaviour that can provide feedback to build an scientific evidence-based adaptive tobacco control policy. Standard decision tree (SDT) based rule discovery depends on decision boundaries in the feature space which are orthogonal to the axis of the feature of a particular decision node. This may limit the ability of SDT to learn intermediate concepts for high dimensional large datasets such as tobacco control. In this paper, we propose a cluster based rule discovery model (CRDM) for generation of more compact and simplified rules for the enhancement of tobacco control policy. The cluster-based approach builds conceptual groups from which a set of decision trees (a decision forest) are constructed. Experimental results on the tobacco control data set show that decision rules from the decision forest constructed by CRDM are simpler and can predict smokers' quitting intention more accurately than a single decision tree.

Feature selection is an important research problem in machine learning and data mining applications. This paper proposes a hybrid wrapper and filter feature selection algorithm by introducing the filter's feature ranking score in the wrapper stage to speed up the search process for wrapper and thereby finding a more compact feature subset. The approach hybridizes a Mutual Information (MI) based Maximum Relevance (MR) filter ranking heuristic with an Artificial Neural Network (ANN) based wrapper approach where Artificial Neural Network Input Gain Measurement Approximation (ANNIGMA) has been combined with MR (MR-ANNIGMA) to guide the search process in the wrapper. The novelty of our approach is that we use hybrid of wrapper and filter methods that combines filter's ranking score with the wrapper-heuristic's score to take advantages of both filter and wrapper heuristics. Performance of the proposed MR-ANNIGMA has been verified using bench mark data sets and compared to both independent filter and wrapper based approaches. Experimental results show that MR-ANNIGMA achieves more compact feature sets and higher accuracies than both filter and wrapper approaches alone.

Distributed Denial of Service (DDoS) attacks pose an increasing threat to the current internet. The detection of such attacks plays an important role in maintaining the security of networks. In this paper, we propose a novel adaptive clustering method combined with feature ranking for DDoS attacks detection. First, based on the analysis of network traffic, preliminary variables are selected. Second, the Modified Global K-means algorithm (MGKM) is used as the basic incremental clustering algorithm to identify the cluster structure of the target data. Third, the linear correlation coefficient is used for feature ranking. Lastly, the feature ranking result is used to inform and recalculate the clusters. This adaptive process can make worthwhile adjustments to the working feature vector according to different patterns of DDoS attacks, and can improve the quality of the clusters and the effectiveness of the clustering algorithm. The experimental results demonstrate that our method is effective and adaptive in detecting the separate phases of DDoS attacks.

Abstract The impact of frame semantic enrichment of texts on the task of factoid question answering (QA) is studied in this paper. In particular, we consider different techniques for answer processing with frame semantics: the level of semantic class identification and role assignment to texts, and the fusion of frame semantic‐based answer‐processing approaches with other methods used in the Text REtrieval Conference (TREC). The impact of each of these aspects on the overall performance of a QA system is analyzed in this paper. The TREC 2004 and TREC 2006 factoid question sets were used for the experiments. These demonstrate that the exploitation of encapsulated frame semantics in FrameNet in a shallow semantic parsing process can enhance answer‐processing performance in factoid QA systems. This improvement is dependent on the level of semantic annotation, the frame semantic alignment method, and the method of fusing frame semantic‐based answer‐processing models with other existing models. A more comprehensively annotated environment with all different part‐of‐speech target predicates provides a higher chance of correct factoid answer retrieval where semantic alignment is based on both semantic classes and a relaxed set of semantic roles for answer span identification. Our experiments on fusion techniques of frame semantic‐based and entity‐based answer‐processing models show that merging answer lists with respect to their scores and redundancy by exploiting a fusion function leads to a more effective overall factoid QA system compared to the use of individual models.

This paper attempts to overcome the tendency of the expectation-maximization (EM) algorithm to locate a local rather than global maximum when applied to estimate the hidden Markov model (HMM) parameters in speech signal modeling. We propose a hybrid algorithm for estimation of the HMM in automatic speech recognition (ASR) using a constraint-based evolutionary algorithm (EA) and EM, the CEL-EM. The novelty of our hybrid algorithm (CEL-EM) is that it is applicable for estimation of the constraint-based models with many constraints and large numbers of parameters (which use EM) like HMM. Two constraint-based versions of the CEL-EM with different fusion strategies have been proposed using a constraint-based EA and the EM for better estimation of HMM in ASR. The first one uses a traditional constraint-handling mechanism of EA. The other version transforms a constrained optimization problem into an unconstrained problem using Lagrange multipliers. Fusion strategies for the CEL-EM use a staged-fusion approach where EM has been plugged with the EA periodically after the execution of EA for a specific period of time to maintain the global sampling capabilities of EA in the hybrid algorithm. A variable initialization approach (VIA) has been proposed using a variable segmentation to provide a better initialization for EA in the CEL-EM. Experimental results on the TIMIT speech corpus show that CEL-EM obtains higher recognition accuracies than the traditional EM algorithm as well as a top-standard EM (VIA-EM, constructed by applying the VIA to EM).

With the increasing popularity of digital products, there is a strong desire to protect the rights of owners against illegal redistribution. Traditional encryption schemes alone do not provide a comprehensive solution to digital rights management, since they do not prevent users who are authorized to use a digital product for their own use from transferring the cleartext content to unauthorized users. However, traceability schemes can be used to trace the illegitimate redistributors effectively. Two types of traceability schemes have been proposed in the literature - traceability codes (TA codes), and codes with the identifiable parent properties (IPP codes). TA codes are special IPP codes, and many TA codes implement an efficient identification algorithm which can determine at least one redistributor. However, many IPP codes are not TA codes, in which case, no efficient identification algorithms are available. In this paper, we generalize the definition of TA codes to derive a new family of traceability codes that is much larger than the family of traditional TA codes. By using existing decoding algorithms with respect to the Lee distance, an efficient identification algorithm is proposed for generalized TA codes. Furthermore, we show that the identification algorithm of generalized TA codes can find more redistributors than those of traditional TA codes.

In this paper we investigate a hybrid model based on the Discrete Gradient method and an evolutionary strategy for determining the weights in a feed forward artificial neural network. Also we discuss different variants for hybrid models using the Discrete Gradient method and an evolutionary strategy for determining the weights in a feed forward artificial neural network. The Discrete Gradient method has the advantage of being able to jump over many local minima and find very deep local minima. However, earlier research has shown that a good starting point for the discrete gradient method can improve the quality of the solution point. Evolutionary algorithms are best suited for global optimisation problems. Nevertheless they are cursed with longer training times and often unsuitable for real world application. For optimisation problems such as weight optimisation for ANNs in real world applications the dimensions are large and time complexity is critical. Hence the idea of a hybrid model can be a suitable option. In this paper we propose different fusion strategies for hybrid models combining the evolutionary strategy with the discrete gradient method to obtain an optimal solution much quicker. Three different fusion strategies are discussed: a linear hybrid model, an iterative hybrid model and a restricted local search hybrid model. Comparative results on a range of standard datasets are provided for different fusion hybrid models.

Narrative or story telling has long been used to structure and organise human experience. In contrast to logical models of reasoning, narrative models enable complex situations to be understood and recalled by humans readily. There is also some indication that narrative models represent the way in which jurors weigh up the veracity of legal evidence. In this work a narrative model is integrated into a logical reasoning model for the purpose of advancing a learning environment that promises to be engaging and effective. The narrative model includes a representation of the point of a story and a simple story grammar. The learning environment is designed to enable the automated generation of plausible scenarios representing a variety of family law property division cases told from the point of view of numerous characters.

This paper reports on an experimental approach to find a modularized artificial neural network solution for the UCI letters recognition problem. Our experiments have been carried out in two parts. We investigate directed task decomposition using expert knowledge and clustering approaches to find the subtasks for the modules of the network. We next investigate processes to combine the modules effectively in a single decision process. After having found suitable modules through task decomposition we have found through further experimentation that when the modules are combined with decision tree supervision, their functional error is reduced significantly to improve their combination through the decision process that has been implemented as a small multilayered perceptron. The experiments conclude with a modularized neural network design for this classification problem that has increased learning and generalization characteristics. The test results for this network are markedly better than a single or stand alone network that has a fully connected topology.

Knowledge discovery from databases (KDD) exercises in law have typically attempted to derive knowledge about decision making processes in the legal domain automatically from datasets. This is made difficult in that real data that represents aspects of a decision process in law is commonly stored as text and rarely stored in structured databases. The central claim advanced here is that KDD processes can be usefully applied to existing datasets of client and demographic data in order to provide feedback for the effective operation of organizations within the legal system. However, the cost of data mining suites and the scarcity of specialized personnel for these tools mitigates against their use. In this study data mining with Association Rules (AR) has been performed on a data-set of over 380,000 records from a legal aid agency. Methods to visualise patterns in order to suggest and test plausible hypotheses from the data have been developed. The tool, called WebAssociate is entirely web based. Domain experts using the tool report favorable responses.

MOTIVATION: The increasing use of DNA microarray-based tumor gene expression profiles for cancer diagnosis requires mathematical methods with high accuracy for solving clustering, feature selection and classification problems of gene expression data. RESULTS: New algorithms are developed for solving clustering, feature selection and classification problems of gene expression data. The clustering algorithm is based on optimization techniques and allows the calculation of clusters step-by-step. This approach allows us to find as many clusters as a data set contains with respect to some tolerance. Feature selection is crucial for a gene expression database. Our feature selection algorithm is based on calculating overlaps of different genes. The database used, contains over 16 000 genes and this number is considerably reduced by feature selection. We propose a classification algorithm where each tissue sample is considered as the center of a cluster which is a ball. The results of numerical experiments confirm that the classification algorithm in combination with the feature selection algorithm perform slightly better than the published results for multi-class classifiers based on support vector machines for this data set. AVAILABILITY: Available on request from the authors.

This paper describes the MODDE (Model of Decision support system Design and Evaluation) framework in some detail. The work is in progress and is being currently applied to the EMBRACE project being developed for the Refugee Review Tribunal (RRT) of Australia. Refugee law is the general legal area we are working in, while the specific domain under investigation is that of the decision makers at the RRT. EMBRACE is a decision support system being designed to assist the RRT in maintaining consistency of decisions, and preserve discretion of decision makers as well as making it easier to cope with high volumes of work in decreasing time frames. The use of the MODDE framework is intended to facilitate systematic attention to important features of decision making in our specific legal domain and to provide a sound basis upon which to evaluate a part of the system intrinsic to user acceptance.

The majority of legal knowledge based systems (LKBS) in commercial use are rule based and target domains of law characterized by large and complex statutes where modelling discretion is not a central concern. Furthermore, to date, few LKBS execute on the World Wide Web. Despite this, LKBS designed for a web environment can make law more universally accessible and transparent. Tools required to facilitate the development of web based systems include a web based expert system shell, conceptual tools that allow for the identification of appropriate domains for web implementation, modeling tools for discretionary domains and architectures for virtual discourse. We present a shell called WebShell that uses two knowledge modelling techniques; decision trees for procedural type tasks and argument trees for tasks that are more discretionary. Rather than translate decision tree knowledge into rules for a conventional inference engine, we map the decision trees into sets we call sequence transition networks. These sets can readily be stored in relational database format in a way that simplifies the inference engine design. Although WebShell facilitates the deployment of LKBS in a web environment, it does not encourage negotiation and virtual discourse. An argumentation shell program, Argument Developer is presented that encourages participants in a virtual discursive community to understand each other's perspectives and reach decisions by consensus.