Search

Background Due to the COVID-19 pandemic, the demand for remote electrocardiogram (ECG) monitoring has increased drastically in an attempt to prevent the spread of the virus and keep vulnerable individuals with less severe cases out of hospitals. Enabling clinicians to set up remote patient ECG monitoring easily and determining how to classify the ECG signals accurately so relevant alerts are sent in a timely fashion is an urgent problem to be addressed for remote patient monitoring (RPM) to be adopted widely. Hence, a new technique is required to enable routine and widespread use of RPM, as is needed due to COVID-19. Objective The primary aim of this research is to create a robust and easy-to-use solution for personalized ECG monitoring in real-world settings that is precise, easily configurable, and understandable by clinicians. Methods In this paper, we propose a Personalized Monitoring Model (PMM) for ECG data based on motif discovery. Motif discovery finds meaningful or frequently recurring patterns in patient ECG readings. The main strategy is to use motif discovery to extract a small sample of personalized motifs for each individual patient and then use these motifs to predict abnormalities in real-time readings of that patient using an artificial logical network configured by a physician. Results Our approach was tested on 30 minutes of ECG readings from 32 patients. The average diagnostic accuracy of the PMM was always above 90% and reached 100% for some parameters, compared to 80% accuracy for the Generalized Monitoring Models (GMM). Regardless of parameter settings, PMM training models were generated within 3-4 minutes, compared to 1 hour (or longer, with increasing amounts of training data) for the GMM. Conclusions Our proposed PMM almost eliminates many of the training and small sample issues associated with GMMs. It also addresses accuracy and computational cost issues of the GMM, caused by the uniqueness of heartbeats and training issues. In addition, it addresses the fact that doctors and nurses typically do not have data science training and the skills needed to configure, understand, and even trust existing black box machine learning models.

PurposeThe best algorithm that was implemented on this Brazilian dataset was artificial immune system (AIS) algorithm. But the time and cost of this algorithm are high. Using asexual reproduction optimization (ARO) algorithm, the authors achieved better results in less time. So the authors achieved less cost in a shorter time. Their framework addressed the problems such as high costs and training time in credit card fraud detection. This simple and effective approach has achieved better results than the best techniques implemented on our dataset so far. The purpose of this paper is to detect credit card fraud using ARO.Design/methodology/approachIn this paper, the authors used ARO algorithm to classify the bank transactions into fraud and legitimate. ARO is taken from asexual reproduction. Asexual reproduction refers to a kind of production in which one parent produces offspring identical to herself. In ARO algorithm, an individual is shown by a vector of variables. Each variable is considered as a chromosome. A binary string represents a chromosome consisted of genes. It is supposed that every generated answer exists in the environment, and because of limited resources, only the best solution can remain alive. The algorithm starts with a random individual in the answer scope. This parent reproduces the offspring named bud. Either the parent or the offspring can survive. In this competition, the one which outperforms in fitness function remains alive. If the offspring has suitable performance, it will be the next parent, and the current parent becomes obsolete. Otherwise, the offspring perishes, and the present parent survives. The algorithm recurs until the stop condition occurs.FindingsResults showed that ARO had increased the AUC (i.e. area under a receiver operating characteristic (ROC) curve), sensitivity, precision, specificity and accuracy by 13%, 25%, 56%, 3% and 3%, in comparison with AIS, respectively. The authors achieved a high precision value indicating that if ARO detects a record as a fraud, with a high probability, it is a fraud one. Supporting a real-time fraud detection system is another vital issue. ARO outperforms AIS not only in the mentioned criteria, but also decreases the training time by 75% in comparison with the AIS, which is a significant figure.Originality/valueIn this paper, the authors implemented the ARO in credit card fraud detection. The authors compared the results with those of the AIS, which was one of the best methods ever implemented on the benchmark dataset. The chief focus of the fraud detection studies is finding the algorithms that can detect legal transactions from the fraudulent ones with high detection accuracy in the shortest time and at a low cost. That ARO meets all these demands.

Human communication is not limited to verbal speech but is infinitely more complex, involving many non-verbal cues such as facial emotions and body language. This paper aims to quantitatively show the impact of non-verbal cues, with primary focus on facial emotions, on the results of multi-modal sentiment analysis. The paper works with a dataset of Spanish video reviews. The audio is available as Spanish text and is translated to English while visual features are extracted from the videos. Multiple classification models are made to analyze the sentiments at each modal stage i.e. for the Spanish and English textual datasets as well as the datasets obtained upon coalescing the English and Spanish textual data with the corresponding visual cues. The results show that the analysis of Spanish textual features combined with the visual features outperforms its English counterpart with the highest accuracy difference, thereby indicating an inherent correlation between the Spanish visual cues and Spanish text which is lost upon translation to English text.

Due to the increasing cost of data storage and computation, more and more graphs (e.g., web graphs, social networks) are outsourced and analyzed in the cloud. However, there is growing concern on the privacy of these outsourced graphs at the hands of untrusted cloud providers. Unfortunately, simple label anonymization cannot protect nodes from being re-identified by adversary who knows the graph structure. To address this issue, existing works adopt the k-automorphism model, which constructs $$(k-1)$$ symmetric vertices for each vertex. It has two disadvantages. First, it significantly enlarges the graphs, which makes graph mining tasks such as subgraph matching extremely inefficient and sometimes infeasible even in the cloud. Second, it cannot protect the privacy of attributes in each node. In this paper, we propose a new privacy model (k, t)-privacy that combines the k-automorphism model for graph structure with the t-closeness privacy model for node label generalization. Besides a stronger privacy guarantee, the paper also optimizes the matching efficiency by (1) an approximate label generalization algorithm TOGGLE with $$(1+\epsilon )$$ approximation ratio and (2) a new subgraph matching algorithm PGP on succinct k-automorphic graphs without decomposing the query graph.

Co-location patterns in spatial dataset are the interesting collection of dissimilar objects which are located in proximity. We keep similar objects in an entity set and maintain that no two objects in a co-location pattern belong to an entity set. Location proximity is based on Euclidean distance measure. However, algorithms for mining patterns in transactional datasets are not directly applicable to spatial datasets for mining co-location patterns. Conventional methods are not applicable to distributed temporal data and many applications generating spatial dataset are inherently distributive in nature. In this paper, a Map-Reduce based approach is proposed to find all co-location patterns from a spatial dataset distributed over nodes. This approach is modularized one and consists of four algorithms. With the first three algorithms in the first approach and by proposing an algorithm for dynamic datasets, this paper contains another approach for the co-location patterns set, that also updates in an incremental manner (not from scratch) whenever certain changes occur in the dataset. Experimental results on larger datasets are also presented.

Influence maximization is a classic network optimization problem, which has been widely used in the field of viral marketing. The influence maximization problem aims to find a fixed number of active nodes. After a specific propagation model, the number of active nodes reaches the maximum. However, the existing influence maximization algorithms are overly pursuing certain indicators of efficiency or accuracy, which cannot be well accepted by some researchers. This paper proposes an effective algorithm to balance the accuracy and efficiency of the influence maximization problem called local two-hop search algorithm (LTHS). The core of the proposed algorithm is a node not only be affected by one-hop neighbor nodes, but also by two-hop neighbor nodes. Firstly, this paper selects initial seed nodes according to the characteristics of the node degree. Generally, the high degree of nodes regards as influential nodes. Secondly, this paper proposes a node two-hop influence evaluate function called two-hop diffusion value (THDV), which can evaluate node influence more accurately. Furthermore, in order to seek higher efficiency, this paper proposes a method to reduce the network scale. This paper conducted full experiments on five real-world social network datasets, and compared with other four well-known algorithms. The experimental results show that the LTHS algorithm is better than the comparison algorithms in terms of efficiency and accuracy.

Internet, & more unambiguously the creation of WWW in the early 1990s, helped people to build an interconnected global platform where information can be stored, shared, and consumed by anyone with an electronic device which has the ability to connect to the Web. This provides a way of putting together lots of information, ideas, and opinion. An interactive platform was born to post content, messages, and opinions under one roof, and the platform is known as social media. Social media has acquired massive popularity and importance that why today almost everyone can't stay away from it. Social media is not only a medium for people to express their thoughts, moreover, but it is also a very powerful tool which can be used by businesses to focus on new and existing customers and increase profit with the help of social media analytics. This paper starts with a discussion on social media with its significance & pitfalls. Later on, this paper presents a brief introduction of sentiment analysis in social media and give an experimental work on sentiment analysis in a social game review.

Data preprocessing is an important component of machine learning pipelines, which requires ample time and resources. An integral part of preprocessing is data transformation into the format required by a given learning algorithm. This paper outlines some of the modern data processing techniques used in relational learning that enable data fusion from different input data types and formats into a single table data representation, focusing on the propositionalization and embedding data transformation approaches. While both approaches aim at transforming data into tabular data format, they use different terminology and task definitions, are perceived to address different goals, and are used in different contexts. This paper contributes a unifying framework that allows for improved understanding of these two data transformation techniques by presenting their unified definitions, and by explaining the similarities and differences between the two approaches as variants of a unified complex data transformation task. In addition to the unifying framework, the novelty of this paper is a unifying methodology combining propositionalization and embeddings, which benefits from the advantages of both in solving complex data transformation and learning tasks. We present two efficient implementations of the unifying methodology: an instance-based PropDRM approach, and a feature-based PropStar approach to data transformation and learning, together with their empirical evaluation on several relational problems. The results show that the new algorithms can outperform existing relational learners and can solve much larger problems., Accepted in MLJ

Multi-dimensional arrays (also known as raster data or gridded data) play a key role in many, if not all science and engineering domains where they typically represent spatio- temporal sensor, image, simulation output, or statistics “datacubes”. As classic database technology does not support arrays adequately, such data today are maintained mostly in silo solutions, with architectures that tend to erode and not keep up with the increasing requirements on performance and service quality. Array Database systems attempt to close this gap by providing declarative query support for flexible ad-hoc analytics on large n-D arrays, similar to what SQL offers on set-oriented data, XQuery on hierarchical data, and SPARQL and CIPHER on graph data. Today, Petascale Array Database installations exist, employing massive parallelism and distributed processing. Hence, questions arise about technology and standards available, usability, and overall maturity. Several papers have compared models and formalisms, and benchmarks have been undertaken as well, typically comparing two systems against each other. While each of these represent valuable research to the best of our knowledge there is no comprehensive survey combining model, query language, architecture, and practical usability, and performance aspects. The size of this comparison differentiates our study as well with 19 systems compared, four benchmarked to an extent and depth clearly exceeding previous papers in the field; for example, subsetting tests were designed in a way that systems cannot be tuned to specifically these queries. It is hoped that this gives a representative overview to all who want to immerse into the field as well as a clear guidance to those who need to choose the best suited datacube tool for their application. This article presents results of the Research Data Alliance (RDA) Array Database Assessment Working Group (ADA:WG), a subgroup of the Big Data Interest Group. It has elicited the state of the art in Array Databases, technically supported by IEEE GRSS and CODATA Germany, to answer the question: how can data scientists and engineers benefit from Array Database technology? As it turns out, Array Databases can offer significant advantages in terms of flexibility, functionality, extensibility, as well as performance and scalability—in total, the database approach of offering “datacubes” analysis-ready heralds a new level of service quality. Investigation shows that there is a lively ecosystem of technology with increasing uptake, and proven array analytics standards are in place. Consequently, such approaches have to be considered a serious option for datacube services in science, engineering and beyond. Tools, though, vary greatly in functionality and performance as it turns out. &nbsp

The comparison of trace variants of business processes opens the door for a fine-grained analysis of the distinctive features inherent in the executions of a process in an organization. The current approaches for trace variant analysis do not consider the situation where a process model is present, and therefore, it can guide the derivation of the trace variants by considering high-level structures present in the process model. In this paper we propose a fresh alternative to trace variant analysis, which proposes a generalized notion of trace variant that incorporates concurrency and iteration. This way, the analyst may be relieved from analyzing trace variants that are essentially the same, if these aspects are disregarded. We propose a general algorithm for model based trace variant analysis which is grounded in encoding the problem into SAT, and a family of heuristic alternatives including a very light sampling technique that represents a good trade-off between quality of the trace variants identified, and the complexity of the analysis. All the techniques of the paper are implemented in two open-source tools, and experiments with publicly available benchmarks are reported. This work has been supported by Farman institute at ENS Paris-Saclay, Spain and by MINECO, Spain and FEDER, Spain funds under grant TIN2017-86727-C2-1-R.

Time series are ubiquitous in data mining applications. Similar to other types of data, annotations can be challenging to acquire, thus preventing from training time series classification models. In this context, clustering methods can be an appropriate alternative as they create homogeneous groups allowing a better analysis of the data structure. Time series clustering has been investigated for many years and multiple approaches have already been proposed. Following the advent of deep learning in computer vision, researchers recently started to study the use of deep clustering to cluster time series data. The existing approaches mostly rely on representation learning (imported from computer vision), which consists of learning a representation of the data and performing the clustering task using this new representation. The goal of this paper is to provide a careful study and an experimental comparison of the existing literature on time series representation learning for deep clustering. In this paper, we went beyond the sole comparison of existing approaches and proposed to decompose deep clustering methods into three main components: (1) network architecture, (2) pretext loss, and (3) clustering loss. We evaluated all combinations of these components (totaling 300 different models) with the objective to study their relative influence on the clustering performance. We also experimentally compared the most efficient combinations we identified with existing non-deep clustering methods. Experiments were performed using the largest repository of time series datasets (the UCR/UEA archive) composed of 128 univariate and 30 multivariate datasets. Finally, we proposed an extension of the class activation maps method to the unsupervised case which allows to identify patterns providing highlights on how the network clustered the time series.

Pattern mining is well established in data mining research, especially for mining binary datasets. Surprisingly, there is much less work about numerical pattern mining and this research area remains under-explored. In this paper we propose Mint, an efficient MDL-based algorithm for mining numerical datasets. The MDL principle is a robust and reliable framework widely used in pattern mining, and as well in subgroup discovery. In Mint we reuse MDL for discovering useful patterns and returning a set of non-redundant overlapping patterns with well-defined boundaries and covering meaningful groups of objects. Mint is not alone in the category of numerical pattern miners based on MDL. In the experiments presented in the paper we show that Mint outperforms competitors among which IPD, RealKrimp, and Slim.

Deep neural networks (DNNs) have introduced novel and useful tools to the machine learning community. Other types of classifiers can potentially make use of these tools as well to improve their performance and generality. This paper reviews the current state of the art for deep learning classifier technologies that are being used outside of deep neural networks. Non-neural network classifiers can employ many components found in DNN architectures. In this paper, we review the feature learning, optimization, and regularization methods that form a core of deep network technologies. We then survey non-neural network learning algorithms that make innovative use of these methods to improve classification performance. Because many opportunities and challenges still exist, we discuss directions that can be pursued to expand the area of deep learning for a variety of classification algorithms.

Fact-checking on the Web has become the main mechanism through which we detect the credibility of the news or information. Existing fact-checkers verify the authenticity of the information (support or refute the claim) based on secondary sources of information. However, existing approaches do not consider the problem of model updates due to constantly increasing training data due to user feedback. It is therefore important to conduct user studies to correct models' inference biases and improve the model in a life-long learning manner in the future according to the user feedback. In this paper, we present FaxPlainAC, a tool that gathers user feedback on the output of explainable fact-checking models. FaxPlainAC outputs both the model decision, i.e., whether the input fact is true or not, along with the supporting/refuting evidence considered by the model. Additionally, FaxPlainAC allows for accepting user feedback both on the prediction and explanation. Developed in Python, FaxPlainAC is designed as a modular and easily deployable tool. It can be integrated with other downstream tasks and allowing for fact-checking human annotation gathering and life-long learning., 5 pages, 4 figures, accepted as a DEMO paper in CIKM 2021

Funding Information: The research in this paper was supported by the EU H2020 project SERUMS: Securing Medical Data in Smart Patient-Centric Healthcare Systems (grant code 826278). Healthcare data sharing platforms have been gaining prominence over the last decade, especially with the emergence of technologies dedicated to increase system security and users’ privacy. Moreover, these platforms are becoming less centralised as time progresses, with need for more data from a variety of locations and settings to be transferred between authorised parties. These requirements also include legal and ethical concerns when creating such solution. Through data sharing, organisations can gain access to previously unknown information or higher quality data, share research findings, and make decisions based on larger (and hopefully more representative) datasets. Such platform should be resilient to attack or loss of data and be able to recover quickly and efficiently from unexpected events. This paper focuses on the blend of emerging technologies (data lake and blockchain) in a design to provide secure and resilient data sharing to only those patients and healthcare professionals authorised to access it across multiple European countries. Postprint

The Internet of Things (IoT) tsunami, public embracement, and the ubiquitous adoption of smart devices are affecting virtually every industry, directly or indirectly. The success of the current and future landscape of IoT and connected devices requires service provision characterized by scalability, ubiquity, reliability, and high-performance, among others. In order to achieve this attribution, the integration of IoT and Cloud Computing (CC), known as cloud IoT, has emerged as a new paradigm providing advanced services specific to aggregating, storing, and processing data generated by IoT. While the convergence of IoT and Cloud brings opportunities, it suffers from specific limitations such as bandwidth, latency, and connectivity. The increasing need for supporting interaction between cloud and IoT led to Edge and Fog Computing (FC) in which computing and storage resources are located not only in the cloud but also at the edges near the source of data. The hierarchical and collaborative edge–fog–cloud architecture brings tremendous benefits as it enables us to distribute the intelligence and computation – including Artificial Intelligence (AI), Machine Learning (ML), and big data analytics – to achieve an optimal solution while satisfying the given constraints e.g., delay-energy tradeoff. Due to the hierarchical, cross-layer, and distributed nature of this model, achieving an osmotic and effective convergence of IoT, edge, fog, and cloud computing requires overcoming many challenges with respect to design and implementation, as well as deployment and evaluation. This paper provides a comprehensive insight into the edge-fog-cloud computing paradigm by providing a blend of discussions on all important aspects of the underlying technologies to offer opportunities for more holistic studies and to accelerate knowledge acquisition. To gain a deep understanding of edge–fog–cloud, we will begin this paper by providing an in-depth tutorial and presenting the main requirements, state-of-the-art reference architectures, building blocks, components, protocols, applications, and other similar computing paradigms, including their similarities and differences. Following this, a holistic reference architecture for edge–fog–cloud IoT is presented and the major corresponding design and deployment considerations (e.g., service models, infrastructure design, provisioning, resource allocation, offloading, service migration, performance evaluation, and security concerns) are discussed. Next, we will take a look at the role of privacy-preserving, distributed, and collaborative analytics as well as the interaction between edge, fog, and cloud computing. Finally, we will overview the main challenges in the field of edge–fog–cloud computing that need to be tackled to realize the full potential of IoT.

The introduction of novel Datalog +/- fragments with good theoretical properties, together with the growing use of enterprise knowledge graphs motivated the development of Vadalog, a knowledge graph management system developed at the University of Oxford. It adopts Warded Datalog +/- as the core of its language for knowledge representation and reasoning, which exhibits a very good tradeoff between computational complexity of reasoning and expressive power, capturing PTIME data complexity while allowing ontological reasoning and full recursion. In this paper, we provide a detailed illustration of the Vadalog system, presenting: the essentials of the first implementation of Warded Datalog +/-; a comprehensive overview of the architecture with specific focus on runtime execution model, memory management, graph traversal strategies and join algorithms; and a detailed experimental evaluation. This paper is a substantially expanded version of the AMW 2019 paper “Datalog-based reasoning for Knowledge Graphs”. To stand apart from previous works on the topic, our focus in this work shall be a comprehensive presentation of the architecture of the Vadalog system and showing how our techniques work together to provide a full-fledged KGMS. In particular, roughly half of this paper is new material created particularly for this comprehensive architectural view. This includes a new series of experiments designed to shed light on architectural choices and alternatives.

International audience; In this paper, we tackle the task of similar question retrieval (QR) which is essential for Community Question Answering (cQA) and aims to retrieve historical questions that are semantically equivalent to the new queries. Over time, with the sharp increase of community archives and the accumulation of duplicated questions, the QR problem has become increasingly challenging due to the shortness of the community questions as well as the word mismatch problem as users can formulate the same query using different wording. Although many efforts have been devoted to address this problem, existing methods mostly relied on supervised models which significantly depend on massive training data sets and manual feature engineering. Such methods are chiefly constrained by their specificities that ignore the word order and do not capture enough syntactic and semantic information in questions. In this paper, we rely on Neural Networks (NNs) which use a deep analysis of words and questions to take into consideration the semantics as well as the structure of questions to predict the semantic text similarity. We propose a deep learning approach based on a Siamese architecture with Long Short-Term Memory (LSTM) networks, augmented with an attention mechanism to let the model give different words different attention while modeling questions. We also explore the use of Convolutional Neural Networks (CNN) nested within the Siamese architecture to retrieve relevant questions. Different similarity measures were tested to predict the semantic similarity between the the pairs of questions. To evaluate the proposed approach, we conducted experiments on large-scale datasets in English and Arabic.

Blockchain technology enables the execution of collaborative business processes involving mutually untrusted parties. Existing tools allow such processes to be modeled using high-level notations and compiled into smart contracts that can be deployed on blockchain platforms. However, these tools do not provide mechanisms to cope with the flexibility requirements inherent to open and dynamic collaboration environments. In particular, existing tools adopt a static role binding approach wherein roles are bound to actors upfront when a process instance is created. Also, these tools do not allow participants to collectively make choices regarding alternative sub-processes or branches in the process model, at runtime. This paper presents a model for dynamic binding of actors to roles in collaborative processes and an associated binding policy specification language. The proposed language is endowed with a Petri net semantics, thus enabling policy consistency verification. Furthermore, the paper introduces a model for consensus-based control-flow flexibility, wherein participants in a process can collectively agree on how to steer the business process within the boundaries defined by control-flow agreement policies. The paper also outlines an approach to compile policy specifications into smart contracts for enforcement. An experimental evaluation shows that the cost of policy enforcement increases linearly with the number of roles, control-flow elements, and policy constraints.

A fundamental problem in conformance checking is aligning event data with process models. Unfortunately, existing techniques for this task are either complex, or can only be applicable to restricted classes of models. This in practice means that for large inputs, current techniques often fail to produce a result. In this paper we propose a method to compute alignments for unconstrained process models, which relies on the use of relaxation labelling techniques on top of a partial order representation of the process model. The technique proposed in this paper precomputes information used in the search for alignments, and is able to produce real alignments that may be close to optimal ones by combining the aforementioned techniques with a locally applied A strategy. Remarkably, the implementation on the proposed technique achieves a speed-up of several orders of magnitude with respect to the approaches in the literature (either optimal, sup-optimal or approximate), often with a reasonable trade-off on the cost of the obtained alignment. This work has been supported by MINECO and FEDER funds under grant TIN2017-86727-C2-1-R.

Along with the rapid development of socio-technical systems, people are playing an increasingly important role in information system and have actually become an essential system component. However, unlike technology-based attacks that have been investigated for decades, social engineering attacks have not been efficiently addressed. In particular, due to the interdisciplinary nature of social engineering, there is a lack of consensus on its definition, hindering the further development of this research field. In this paper, we propose a comprehensive and fundamental ontology of social engineering based on a systematic review of existing social engineering taxonomies and ontologies in order to provide a theoretical foundation for social engineering analysis. The essential contributions of this paper include: (1) propose a comprehensive ontology of social engineering and precisely specify ontological definitions of its essential concepts based on Situation Calculus; (2) enumerate and summarize a set of social engineering techniques and present their fine-grained classification based on the proposed ontology; (3) incorporate psychology and sociology knowledge into social engineering analysis, encapsulating such knowledge in terms of a formalized ontology. We have evaluated our ontology based on a set of real social engineering attacks, the results of which show the usefulness of our proposal.

Traditional manufacturing and production systems are in the throes of a digital transformation. By blending the real and virtual production worlds, it is now possible to connect all parts of the production process: devices, products, processes, systems and people, in an informational ecosystem. This paper examines the underpinning issues that characterize the challenges for transforming traditional manufacturing to a Cyber Physical Production System. Such a transformation constitutes a major endeavour for requirements engineers who need to identify, specify and analyse the effects that a multitude of assets need to be transformed towards a network of collaborating devices, information sources, and human actors. The paper reports on the e-CORE approach which is a systematic, analytical and traceable approach to Requirements Engineering founded on a ‘capability-oriented’ paradigm with particular emphasis on ‘early requirements’. The utility of the approach is demonstrated using two use cases from an industrial-size application in the automotive industry and specifically the case of requirements of Fiat Chrysler Automobiles.

The emergence of Location-based social networks (LBSNs) in recent years has boosted improvements in Recommender Systems for a new and specific task: the recommendation of points-of-interest (POI). Despite all the blunt advances recently observed, the area still lacks an updated and consolidated view about the main limitations, common assumptions, and directions we are following as a community. Thus, this paper aims to provide an updated picture of POI recommendation, identifying relevant efforts, results, contributions, and limitations. In this sense, through a systematic mapping, we selected 74 relevant papers published in the last three years (2017, 2018, and 2019) in the main conferences and journals of the area. Most of these studies have focused on the general POI recommendation problem, although we still could identify a significant number of efforts addressing specific problems in the area, such as Time-Aware, Next POI, and In/Out-of-Town Recommendations. Also, we found user contexts, such as social network and geolocation, as recurring data types used to improve preference elicitation when exploring different methods (e.g., collaborative filtering, factorization, probabilistic, link-based, and hybrid methods). As major limitations, first, we identified that these studies prioritize accuracy over other quality dimensions, despite the consensus in the RS community that accuracy alone is not enough to assess the practical effectiveness of Recommender Systems. Further, we found a low intersection of metrics and datasets used to evaluate the proposed solutions, along with a large number of metrics used in a few distinct studies. These observations point out the potential damage to reproducibility and straightforward comparison of results in the area, motivating us to propose an extensible POI recommendation benchmark. Through this benchmark, we showed that when proper evaluations are carried out, by considering different datasets, metrics, and baselines, identifying which algorithm is the best one becomes non-trivial. Finally, we highlight as a promising future work the in-depth exploitation of textual data since just a few evaluated studies marginally use this rich data source.

In their recent paper in this journal, Delgrande, Renne and Sack study qualitative probability and discuss its role in Artificial Intelligence. Building on related work by de Finetti, Scott, Segerberg and others, the authors provide general axioms for qualitative probability. In this paper we investigate Koopman's conditional qualitative probability from the computational viewpoint. An important part of his work, published in the Annals of Mathematics in 1940-1941, deals with finite conjunctions K of statements of the form “the probability of a given h does not exceed the probability of b given k”, with a , b , h , k elements of a boolean algebra. Upon coding these elements by boolean formulas, we provide a decision procedure to check the consistency of any such K . As an immediate consequence, also inferences in Koopman's probability theory are shown to be computable. These problems of qualitative probability theory significantly generalize Boole's (typically quantitative) problem of estimating the possible probabilities of a new event given the probabilities of other events. Boole's classical problem today is known as the optimization version of the probabilistic satisfiability problem PSAT. In 1986 Nilsson published an influential paper on this subject in this journal. The scope of our results is much larger than that of PSAT, because Koopman's conjunctions K also formalize the key notion of independence. Some familiarity with boolean logic and finite boolean algebras is the only prerequisite for this paper.

Answer Set Programming (ASP) is a well-established paradigm for declarative programming and nonmonotonic reasoning. ASP allows for flexible modeling using rules. ASP rules induce a set of intended models called answer sets. Incoherence, the non-existence of answer sets, is therefore a feature of ASP, indicating that the rules admit no intended models. However, this feature can also be problematic in certain circumstances: errors that cause incoherence are notoriously difficult to debug, and query answering will not provide any meaningful answers for incoherent programs. Paracoherent semantics have been suggested as a remedy. They extend the classical notion of answer sets to draw meaningful conclusions also from incoherent programs. However, paracoherent semantics have essentially been inapplicable in practice, due to the lack of efficient algorithms and implementations. In this paper, this lack is addressed, and several different algorithms to compute semi-stable and semi-equilibrium models are proposed and implemented within an answer set solving framework. A key role in the framework is played by syntactic program transformations that allow for characterizing paracoherent semantics in terms of the answer sets of transformed programs. Apart from existing transformations from the literature, a novel transformation is also proposed, which provides an alternative characterization of paracoherent semantics in terms of (extended) externally supported models. Notably, the new transformation is more compact than the existing ones, and brings performance benefits. An extensive empirical performance comparison among the algorithms on benchmarks from ASP competitions and a real-world use case is given as well. It shows not only that the methods developed in this paper lead to practically effective systems, but also show a clear advantage of the methods that rely on (extended) externally supported models.

AIOps(Artificial Intelligence Operations) is emerging as one of the most important technology in industrial automation, and accurate time series prediction plays a crucial role in it. However, due to the non-linear and non-stationary characteristics of Ops(operations) data, traditional time series forecasting models can not effectively extract good enough sequence data features and result in poor forecasting accuracy. To fully extract the Ops data information and construct a robust prediction model, this paper proposes a hybrid model based on recursive empirical mode decomposition (REMD) and memory wavelet neural network (MWNN) to improve the forecasting accuracy of Ops data. In REMD-MWNN, we first use REMD to decompose the Ops data into multiple intrinsic modal functions (IMF) at different time scales. Then, to make full use of the historical information of the Ops data and reduce the running time, we designed a new memory recurrent neural network, namely MWNN. Next, use MWNN to predict multiple inherent IMFs respectively to obtain the predicted value of the corresponding subsequence. Finally, the final prediction result is obtained by reconstructing the prediction value of each subsequence. In the comprehensive experiment, we selected the Ops data of an open education platform. The experimental results show that, compared with other algorithms, the model proposed in this paper is highly competitive in predicting future changes and capturing the evolution mode of hidden factors. The experimental data involved in this paper can be downloaded from this website: https://github.com/1600383075/REMD-MWNN/tree/master/data .

In this paper, we propose a deep collaborative filtering recommendation model, which consists of an attention-based representation learning component and a multi-input matching function learning component. This model takes interaction matrix based on implicit feedback as data source to construct representations of long-term user preferences and item latent features. In the representation learning, a time-aware attention network is used, which uses the embedding vectors of the predicted item, recent historical interaction items, and the interaction time of recent historical interaction items to estimate the weights of different historical interaction items to short-term user preferences modeling. Then, the dynamic user preference representation can be obtained by combining short-term preferences with long-term preferences. In the matching function learning, a multi-input deep learning model is used. Its input includes not only the dynamic user preference representation and the item latent feature representation, but also the linear interaction between the two representations, so that the model has more powerful feature interactions learning ability. Experimental results on four datasets from different domains show that our method is largely superior to the state-of-the-art collaborative filtering methods, and the novel techniques we propose in this paper are effective in improving recommendation performance.

Cardiovascular disease (CVD) is a general term for a series of heart or blood vessels abnormality that serves as a global leading reason for death. The earlier the abnormal heart rhythm is discovered, the less severe the sequela and the faster the recovery. Electrocardiogram (ECG), as a main way to detect the electrical activity of heart, is a very important harmless means of predicting and diagnosing CVDs. However, ECG signal has characteristics of complex and high chaos, making it time-consuming and exhausting to interpret ECG signal even for experts. Hence, computer-aided methods are required to relief human burden and reduce errors caused by tiredness, inter- and intra-difference. Deep learning shows outstanding performance on ECG classification studies recent few years. Its hierarchical architecture enables higher-level features obtained and its strong ability to feature extraction contributes to classification project. Latest studies can achieve higher accuracy and efficiency than manual classification by experts. In this paper, we review the existing studies of deep learning applied in ECG diagnosis according to four typical algorithms: stacked auto-encoders, deep belief network, convolutional neural network and recurrent neural network. We first introduced the mechanism, development and application of the algorithms. Then we review their applications in ECG diagnosis systematically, discussing their highlights and limitations. Our view about future potential development of deep learning in ECG diagnosis is stated in the final part of this paper.

In plenty of real-life tasks, strongly supervised information is hard to obtain, such that there is not sufficient high-quality supervision to make traditional learning approaches succeed. Therefore, weakly supervised learning has drawn considerable attention recently. In this paper, we consider the problem of learning from incomplete and inaccurate supervision, where only a limited subset of training data is labeled but potentially with noise. This setting is challenging and of great importance but rarely studied in the literature. We notice that in many applications, the limited labeled data are usually with one-sided noise. For instance, considering the bug detection task in the software system, the identified buggy codes are indeed with defects whereas the codes that have been checked many times or newly fixed may still have other flaws due to the complexity of the system. We propose a novel method which is able to effectively alleviate the negative influence of one-sided label noise with the help of a vast number of unlabeled data. Excess risk analysis is provided as theoretical justifications on the usefulness of incomplete and one-sided inaccurate supervision. We conduct experiments on synthetic, benchmark datasets, and real-life tasks to validate the effectiveness of the proposed approach.

Mobile Deep Learning (MDL) has emerged as a privacy-preserving learning paradigm for mobile devices. This paradigm offers unique features such as privacy preservation, continual learning and low-latency inference to the building of personal mobile sensing applications. However, squeezing Deep Learning to mobile devices is extremely challenging due to resource constraint. Traditional Deep Neural Networks (DNNs) are usually over-parametered, hence incurring huge resource overhead for on-device learning. In this paper, we present a novel on-device deep learning framework named MDLdroidLite that transforms traditional DNNs into resource-efficient model structures for on-device learning. To minimize resource overhead, we propose a novel Release-and-Inhibit Control (RIC) approach based on Model Predictive Control theory to efficiently grow DNNs from tiny to backbone. We also design a gate-based fast adaptation mechanism for channel-level knowledge transformation to quickly adapt new-born neurons with existing neurons, enabling safe parameter adaptation and fast convergence for on-device training. Our evaluations show that MDLdroidLite boosts on-device training on various PMS datasets with 28x to 50x less model parameters, 4x to 10x less floating number operations than the state-of-the-art model structures while keeping the same accuracy level.

Reorganizing bus frequencies to cater for actual travel demands can signicantly save the cost of the public transport system. This paper studies the bus frequency optimization problem considering the user satisfaction. Specically, for the rst time to our best knowledge, we study how to schedule the buses such that the total number of passengers who could receive their bus services within the waiting time threshold can be maximized. We propose two variants of the problem, FAST and FASTCO, to cater for different application needs and prove that both are NP-hard. To solve FAST effectively and efciently, we rst present an index-based (1 1/e)-approximation algorithm. By exploiting the locality property of routes in a bus network, we further propose a partition-based greedy method for that achieves a (1)(1 1/e) approximation ratio. Then we propose a progressive partition-based greedy method for to further boost the efciency while achieving a (1)(1 1/e) approximation ratio. For the FASTCO problem, two greedy-based heuristic methods are proposed. Experiments on a real city-wide bus dataset in Singapore have been conducted to verify the efciency, effectiveness, and scalability of our methods in addressing FAST and FASTCO.

Search engines are crucial for information gathering systems (IGS). New challenges face search engines concerning automatic learning from user requests. In this paper, a new hybrid intelligent system is proposed to enhance the search process. Based on a Multilayer Fuzzy Inference System (MFIS), the first step is to implement a scalable system to relay logical rules in order to produce three classifications for search behavior, user profiles, and query characteristics from analysis of navigation log files. These three outputs from the MFIS are used as inputs for the second step, an Adaptive Neuro-Fuzzy Inference System (ANFIS). The training process of the ANFIS replaced the rules by adjusting the weights in order to find the most relevant result for the search query. This proposed system, called MFIS-ANFIS, is implemented as an experimental system. The system performance is evaluated using quantitative and comparative analysis. MFIS-ANFIS aimed to be the core of intelligent and reliable search process.

Cross-manifold clustering is a hard topic and many traditional clustering methods fail because of the cross-manifold structures. In this paper, we propose a Multiple Flat Projections Clustering (MFPC) to deal with cross-manifold clustering problems. In our MFPC, the given samples are projected into multiple subspaces to discover the global structures of the implicit manifolds. Thus, the cross-manifold clusters are distinguished from the various projections. Further, our MFPC is extended to nonlinear manifold clustering via kernel tricks to deal with more complex cross-manifold clustering. A series of non-convex matrix optimization problems in MFPC are solved by a proposed recursive algorithm. The synthetic tests show that our MFPC works on the cross-manifold structures well. Moreover, experimental results on the benchmark datasets show the excellent performance of our MFPC compared with some state-of-the-art clustering methods., Comment: 12 pages, 58 figures

Dynamic ridesharing refers to services that arrange one-time shared rides on short notice. It underpins various real-world intelligent transportation applications such as car-pooling, food delivery and last-mile logistics. A core operation in dynamic ridesharing is the "insertion operator". Given a worker and a feasible route which contains a sequence of origin-destination pairs from previous requests, the insertion operator inserts a new origin-destination pair from a newly arrived request into the current route such that certain objective is optimized. Common optimization objectives include minimizing the \rev{maximum/sum} flow time of all requests and minimizing the total travel time of the worker. Despite its frequent usage, the insertion operator has a time complexity of $O(n^3)$ , where $n$ is the number of all requests assigned to the worker. The cubic running time of insertion fundamentally limits the efficiency of urban-scale dynamic ridesharing based applications. In this paper, we propose a novel partition framework and a dynamic programming based insertion with a time complexity of $O(n^2)$ . We further improve the time efficiency of the insertion operator to $O(n)$ harnessing efficient index structures, such as fenwick tree. Evaluations on two real-world large-scale datasets show that our methods can accelerate insertion by 1.5 to 998.1 times.

Recommender Systems (RS) currently represent a fundamental tool in online services, especially with the advent of Online Social Networks (OSN). In this case, users generate huge amounts of contents and they can be quickly overloaded by useless information. At the same time, social media represent an important source of information to characterize contents and users’ interests. RS can exploit this information to further personalize suggestions and improve the recommendation process. In this paper we present a survey of Recommender Systems designed and implemented for Online and Mobile Social Networks, highlighting how the use of social context information improves the recommendation task, and how standard algorithms must be enhanced and optimized to run in a fully distributed environment, as opportunistic networks. We describe advantages and drawbacks of these systems in terms of algorithms, target domains, evaluation metrics and performance evaluations. Eventually, we present some open research challenges in this area.

Recommender systems personalize content by recommending items to users. Item recommendation algorithms are evaluated by metrics that compare the positions of truly relevant items among the recommended items. To speed up the computation of metrics, recent work often uses sampled metrics where only a smaller set of random items and the relevant items are ranked. This paper investigates such sampled metrics in more detail and shows that they are inconsistent with their exact counterpart, in the sense that they do not persist relative statements, for example, recommender A is better than B , not even in expectation. Moreover, the smaller the sample size, the less difference there is between metrics, and for very small sample size, all metrics collapse to the AUC metric. We show that it is possible to improve the quality of the sampled metrics by applying a correction, obtained by minimizing different criteria. We conclude with an empirical evaluation of the naive sampled metrics and their corrected variants. To summarize, our work suggests that sampling should be avoided for metric calculation, however if an experimental study needs to sample, the proposed corrections can improve the quality of the estimate.

In this paper, we propose the Data Flow Interface (DFI) as a way to make it easier for data processing systems to exploit high-speed networks without the need to deal with the complexity of RDMA. By lifting the level of abstraction, DFI factors out much of the complexity of network communication and makes it easier for developers to declaratively express how data should be efficiently routed to accomplish a given distributed data processing task. As we show in our experiments, DFI is able to support a wide variety of data-centric applications with high performance at a low complexity for the applications.

Machine Learning has emerged in the last years as the main solution to many of nowadays’ data-based decision problems. However, while new and more powerful algorithms and the increasing availability of computational resources contributed to a widespread use of Machine Learning, significant challenges still remain. Two of the most significant nowadays are the need to explain a model’s predictions, and the significant costs of training and re-training models, especially with large datasets or in streaming scenarios. In this paper we address both issues by proposing an approach we deem predictive and user-centric. It is predictive in the sense that it estimates the benefit of re-training a model with new data, and it is user-centric in the sense that it implements an explainable interface that produces interpretable explanations that accompany predictions. The former allows to reduce necessary resources (e.g. time, costs) spent on retraining models when no improvements are expected, while the latter allows for human users to have additional information to support decision-making. We validate the proposed approach with a group of public datasets and present a real application scenario.

Density Peak Clustering (DPC), a popular density-based clustering approach, has received considerable attention from the research community primarily due to its simplicity and fewer-parameter requirement. However, the resultant clusters obtained using DPC are influenced by the sensitive parameter $d_c$, which depends on data distribution and requirements of different users. Besides, the original DPC algorithm requires visiting a large number of objects, making it slow. To this end, this paper investigates index-based solutions for DPC. Specifically, we propose two list-based index methods viz. (i) a simple List Index, and (ii) an advanced Cumulative Histogram Index. Efficient query algorithms are proposed for these indices which significantly avoids irrelevant comparisons at the cost of space. For memory-constrained systems, we further introduce an approximate solution to the above indices which allows substantial reduction in the space cost, provided that slight inaccuracies are admissible. Furthermore, owing to considerably lower memory requirements of existing tree-based index structures, we also present effective pruning techniques and efficient query algorithms to support DPC using the popular Quadtree Index and R-tree Index. Finally, we practically evaluate all the above indices and present the findings and results, obtained from a set of extensive experiments on six synthetic and real datasets. The experimental insights obtained can help to guide in selecting a befitting index.

In an intelligent surveillance system, detecting and predicting diverse collective crowd behaviors has emerged as a challenging problem for efficient crowd management. In real-world scenarios, potential disasters and hazards can be averted by considering crowd psychology for predicting crowd behaviors. This paper proposes an approach that exploits the psychological and cognitive aspects of human behavior in determining nine diverse crowd behaviors. The proposed approach is a combination of two cognitive deep learning frameworks and a psychological fuzzy computational model that utilizes OCC theory of emotions, OCEAN five-factor model of personality and visual attention for detecting crowd behaviors. Experiments are performed on different datasets and the results prove that our approach is successful in detecting and predicting crowd behavior in confronting situations and also outperforms the state-of-the-art methods. In particular, considering psychological aspects and cognition in determining crowd behavior is beneficial for rectifying the semantic ambiguity in identifying crowd behaviors.

This paper presents DistRDF2ML, the generic, scalable, and distributed framework for creating in-memory data preprocessing pipelines for Spark-based machine learning on RDF knowledge graphs. This framework introduces software modules that transform large-scale RDF data into ML-ready fixed-length numeric feature vectors. The developed modules are optimized to the multimodal nature of knowledge graphs. DistRDF2ML provides aligned software design and usage principles as common data science stacks that offer an easy-to-use package for creating machine learning pipelines. The modules used in the pipeline, the hyper-parameters and the results are exported as a semantic structure that can be used to enrich the original knowledge graph. The semantic representation of metadata and machine learning results offers the advantage of increasing the machine learning pipelines’ reusability, explainability, and reproducibility. The entire framework of DistRDF2ML is open source, integrated into the holistic SANSA stack, documented in scala-docs, and covered by unit tests. DistRDF2ML demonstrates its scalable design across different processing power configurations and (hyper-)parameter setups within various experiments. The framework brings the three worlds of knowledge graph engineers, distributed computation developers, and data scientists closer together and offers all of them the creation of explainable ML pipelines using a few lines of code.

Change point detection plays a fundamental role in many real-world applications, where the goal is to analyze and monitor the behaviour of a data stream. In this paper, we study change detection in binary streams. To this end, we use a likelihood ratio between two models as a measure for indicating change. The first model is a single bernoulli variable while the second model divides the stored data in two segments, and models each segment with its own bernoulli variable. Finding the optimal split can be done in \( \mathcal {O} \mathopen {}\left( n\right) \) time, where n is the number of entries since the last change point. This is too expensive for large n. To combat this we propose an approximation scheme that yields \((1 - \epsilon )\) approximation in \( \mathcal {O} \mathopen {}\left( \epsilon ^{-1} \log ^2 n\right) \) time. The speed-up consists of several steps: First we reduce the number of possible candidates by adopting a known result from segmentation problems. We then show that for fixed bernoulli parameters we can find the optimal change point in logarithmic time. Finally, we show how to construct a candidate list of size \( \mathcal {O} \mathopen {}\left( \epsilon ^{-1} \log n\right) \) for model parameters. We demonstrate empirically the approximation quality and the running time of our algorithm, showing that we can gain a significant speed-up with a minimal average loss in optimality.

The recent years have witnessed the growing popularity of indoor location-based services (LBS) in practice and research. Among others, indoor shortest path query (ISPQ) is of fundamental importance for indoor LBS. However, existing works on ISPQ ignore indoor temporal variations, e.g., the open and close times associated with entities like doors and rooms. In this paper, we define a new type of query called Indoor Temporal-variation aware Shortest Path Query (ITSPQ). It returns the valid shortest path based on the up-to-date indoor topology at the query time. A set of techniques is designed to answer ITSPQ efficiently. We design a graph structure (IT-Graph) that captures indoor temporal variations. To process ITSPQ using IT-Graph, we design two algorithms that check a door's accessibility synchronously and asynchronously. Furthermore, we propose a novel index structure (IT-Index) that extends the state-of-the-art index significantly by storing dynamic door-to-door distances in a compact distance cube associated with tree nodes. When processing ITSPQ using IT-Index, we make use of the distance cube to avoid time-consuming indoor distance computation on-the-fly. We evaluate the proposed techniques using extensive experiments on synthetic and real data. The results show that our IT-Index based method is the most efficient for processing ITSPQ at a modest cost of index memory consumption.

Convolutional Networks have dominated the field of computer vision for the last ten years, exhibiting extremely powerful feature extraction capabilities and outstanding classification performance. The main strategy to prolong this trend relies on further upscaling networks in size. However, costs increase rapidly while performance improvements may be marginal. We hypothesise that adding heterogeneous sources of information may be more cost-effective to a CNN than building a bigger network. In this paper, an ensemble method is proposed for accurate image classification, fusing automatically detected features through Convolutional Neural Network architectures with a set of manually defined statistical indicators. Through a combination of the predictions of a CNN and a secondary classifier trained on statistical features, better classification performance can be cheaply achieved. We test multiple learning algorithms and CNN architectures on a diverse number of datasets to validate our proposal, making public all our code and data via GitHub. According to our results, the inclusion of additional indicators and an ensemble classification approach helps to increase the performance in 8 of 9 datasets, with a remarkable increase of more than 10% precision in two of them., Comment: 16 pages

Nowadays natural language generation (NLG) is used in everything from news reporting and chatbots to social media management. Recent advances in machine learning have made it possible to train NLG systems that seek to achieve human-level performance in text writing and summarisation. In this paper, we propose such a system in the context of Wikipedia and evaluate it with Wikipedia readers and editors. Our solution builds upon the ArticlePlaceholder, a tool used in 14 under-resourced Wikipedia language versions, which displays structured data from the Wikidata knowledge base on empty Wikipedia pages. We train a neural network to generate an introductory sentence from the Wikidata triples shown by the ArticlePlaceholder, and explore how Wikipedia users engage with it. The evaluation, which includes an automatic, a judgement-based, and a task-based component, shows that the summary sentences score well in terms of perceived fluency and appropriateness for Wikipedia, and can help editors bootstrap new articles. It also hints at several potential implications of using NLG solutions in Wikipedia at large, including content quality, trust in technology, and algorithmic transparency.

As the Internet of Things (IoT) is becoming increasingly popular, we have experienced more security breaches that are associated with the connection of vulnerable IoT devices. Therefore, it is crucial to employ intrusion detection techniques to mitigate attacks that exploit IoT security vulnerabilities. However, due to the limited capabilities of IoT devices and the specific protocols used, conventional intrusion detection mechanisms may not work well for IoT environments. In this paper, we propose a novel intrusion detection model that uses machine learning to effectively detect cyber-attacks and anomalies in resource-constraint IoT networks. Through a set of optimizations including removal of multicollinearity, sampling, and dimensionality reduction, our model can identify the most important features to detect intrusions using much fewer training data and less training time. Extensive experiments were performed on the CICIDS2017 and NSL-KDD datasets respectively to evaluate the proposed approach. The experimental results on two popular datasets show that our model has a high detection rate and a low false alarm rate. It outperforms existing models in multiple performance metrics and is consistent in classifying major cyber-attacks, respectively. Most importantly, unlike traditional resource-intensive intrusion detection systems, the proposed model is lightweight and can be deployed on IoT nodes with limited power and storage capabilities.

Distributed database systems are gaining importance due to the production in massive amount of data. The efficacy of such systems is highly dependent upon the design of the system. To increase the effectiveness and efficiency of distributed databases, two processes are mainly employed i.e. fragmentation and allocation. Fragmentations can be vertical or horizontal. This work focuses on vertical fragmentation design methods. In this paper, a novel differential bond energy (DBE) algorithm is proposed with objective to determine optimal partition point. The performance of proposed algorithm is compared with classical bond energy algorithm (BEA) on basis of global affinity measure (GAM) value. Results are depicted in form of line graphs. The mean difference in GAM values for both algorithms are also illustrated. The experimental results portrays that DBE is suitable for vertical fragmentation of high dimensional problems as it attain high GAM value as compared to BEA on various datasets.

Burst is a common pattern in data streams which is characterized by a sudden increase in terms of arrival rate followed by a sudden decrease. Burst detection has attracted extensive attention from the research community. In this paper, we propose a novel sketch, namely BurstSketch, to detect bursts accurately in real time. BurstSketch first uses the technique Running Track to select potential burst items efficiently, and then monitors the potential burst items and capture the key features of burst pattern by a technique called Snapshotting. Experimental results show that our sketch achieves a 1.75 times higher recall rate than the strawman solution.

Various strategies for active learning have been proposed in the machine learning literature. In uncertainty sampling, which is among the most popular approaches, the active learner sequentially queries the label of those instances for which its current prediction is maximally uncertain. The predictions as well as the measures used to quantify the degree of uncertainty, such as entropy, are traditionally of a probabilistic nature. Yet, alternative approaches to capturing uncertainty in machine learning, alongside with corresponding uncertainty measures, have been proposed in recent years. In particular, some of these measures seek to distinguish different sources and to separate different types of uncertainty, such as the reducible (epistemic) and the irreducible (aleatoric) part of the total uncertainty in a prediction. The goal of this paper is to elaborate on the usefulness of such measures for uncertainty sampling, and to compare their performance in active learning. To this end, we instantiate uncertainty sampling with different measures, analyze the properties of the sampling strategies thus obtained, and compare them in an experimental study.

The performance requirement of deep learning inevitably brings up with the expense of high computational complexity and memory requirements, to make it problematic for the deployment on resource-constrained devices. Edge computing, which distributed organizes the computing node close to the data source and end-device, provides a feasible way to tackle the high-efficiency demand and substantial computational load. Whereas given edge device is resource-constrained and energy-sensitive, designing effective neural network architecture for specific edge device is urgent in the sense that deploys the deep learning application by the edge computing solution. Undoubtedly manually design the high-performing neural architectures is burdensome,let alone taking account of the resource-constraint for the specific platform. Fortunately, the success of Neural Architecture Search techniques come up with hope recently. This paper dedicate todirectly employ multi-objective NAS on the resource-constrained edge devices. We first propose the framework of multi-objectiveNAS on edge device, which comprehensively considers the performance and real-world efficiency. Our improved MobileNet-V2 search space also strikes the scalability and practicality, so that a series of Pareto-optimal architectures are received. Benefits from the directness and specialization during search procedure,our experiment on JETSON NANO shows the comparable resultwith the state-of-the-art models on ImageNet.