Search

We explored the paper augmentation technique where a tablet computer is placed under the paper to modify its content appearance and create the illusion of a static object appearing perceptually dynamic. Our approach is based on the Deformation Lamps (DL) technique commonly used with front projection. However, we applied it to rear projection, which removes the need for a fixed hardware setup or handling and wearing dedicated hardware. The key challenge was to ensure the appropriate mixing of light reflecting from the paper (to see what is printed on the paper) and the light passing though the paper (to see what the tablet screen is projecting) since (i) a tablet screen is a relatively weak light source when being used in rear projection and (ii) commonly, paper has high opacity so the content from one side is not visible on the other side. In order to achieve the appropriate mixing of light, we reduced the opacity of the paper by perforating it (thus allowing more light to pass through), and modified the algorithm for generating dynamic luminance animation by adding gamma and brightness correction. The outcomes provide a guide on how to modify the paper and the algorithm in order to achieve the DL effect with rear projection using a tablet computer.

Automatically detecting the positions of key-points (e.g., facial key-points or finger key-points) in an image is an essential problem in many applications, such as driver's gaze detection and drowsiness detection in automated driving systems. With the recent advances of Deep Neural Networks (DNNs), Key-Points detection DNNs (KP-DNNs) have been increasingly employed for that purpose. Nevertheless, KP-DNN testing and validation have remained a challenging problem because KP-DNNs predict many independent key-points at the same time -- where each individual key-point may be critical in the targeted application -- and images can vary a great deal according to many factors. In this paper, we present an approach to automatically generate test data for KP-DNNs using many-objective search. In our experiments, focused on facial key-points detection DNNs developed for an industrial automotive application, we show that our approach can generate test suites to severely mispredict, on average, more than 93% of all key-points. In comparison, random search-based test data generation can only severely mispredict 41% of them. Many of these mispredictions, however, are not avoidable and should not therefore be considered failures. We also empirically compare state-of-the-art, many-objective search algorithms and their variants, tailored for test suite generation. Furthermore, we investigate and demonstrate how to learn specific conditions, based on image characteristics (e.g., head posture and skin color), that lead to severe mispredictions. Such conditions serve as a basis for risk analysis or DNN retraining., to appear in ISSTA 2021

Using machine learning (ML) services, both service customers and providers need to monitor complex contractual constraints of ML service that are strongly related to ML models and data. Therefore, establishing and monitoring comprehensive ML contracts are crucial in ML serving. This paper demonstrates a set of features and utilities of the QoA4ML framework for ML contracts.

This companion paper supports the experimental replication of paper "Outfit Compatibility Prediction and Diagnosis with Multi-Layered Comparison Network", which is presented at ACM Multimedia 2019. We provide the software package for replicating the implementation of Multi-Layered Comparison Network (MCN), as well as the Polyvore-T dataset and baseline methods compared in the original paper. This paper contains the guides to reproduce the experiment results including outfit compatibility prediction, outfit diagnosis and automatic outfit revision.

Biological systems exhibit extraordinary levels of diversity. Embodied behaviours are the emergent property of many loosely-coupled parallel processes, from the microscopic level, e.g. materials, to more abstracted levels, e.g. organs and limbs. We summarise our investigations into using a synthetic methodology, i.e. an understanding-by-building approach, for designing the complex interactions of falling paper shapes. By studying how simple systems such as a falling paper shape behave, we can analyse the specific characteristics of the interaction between morphology and the environment, and how this leads to programmable non-trivial behaviours. We present current results and discuss the implications on the future of design in robotics.

Nowadays, recommendation systems play an indispensable role in many fields, including e-commerce, finance, economy, and gaming. There is emerging research on publication venue recommendation systems to support researchers when submitting their scientific work. Several publishers such as IEEE, Springer, and Elsevier have implemented their submission recommendation systems only to help researchers choose appropriate conferences or journals for submission. In this work, we present a demo framework to construct an effective recommendation system for paper submission. With the input data (the title, the abstract, and the list of possible keywords) of a given manuscript, the system recommends the list of top relevant journals or conferences to authors. By using state-of-the-art techniques in natural language understanding, we combine the features extracted with other useful handcrafted features. We utilize deep learning models to build an efficient recommendation engine for the proposed system. Finally, we present the User Interface (UI) and the architecture of our paper submission recommendation system for later usage by researchers.

Imprecise composite location references formed using ad hoc spatial expressions in English text makes the geocoding task challenging for both inference and evaluation. Typically such spatial expressions fill in unestablished areas with new toponyms for finer spatial referents. For example, the spatial extent of the ad hoc spatial expression "north of" or "50 minutes away from" in relation to the toponym "Dayton, OH" refers to an ambiguous, imprecise area, requiring translation from this qualitative representation to a quantitative one with precise semantics using systems such as WGS84. Here we highlight the challenges of geocoding such referents and propose a general formal representation that employs background knowledge, semantic approximations and rules, and fuzzy linguistic variables. We also discuss an appropriate evaluation technique for the task that is based on human contextualized and subjective judgment.

Keyphrases associated with research papers provide an effective way to find useful information in the large and growing scholarly digital collections. However, keyphrases are not always provided with the papers, but they need to be extracted from their content. In this paper, we explore keyphrase extraction formulated as sequence labeling and utilize the power of Conditional Random Fields in capturing label dependencies through a transition parameter matrix consisting of the transition probabilities from one label to the neighboring label. We aim at identifying the features that, by themselves or in combination with others, perform well in extracting the descriptive keyphrases for a paper. Specifically, we explore word embeddings as features along with traditional, document-specific features for keyphrase extraction. Our results on five datasets of research papers show that the word embeddings combined with document specific features achieve high performance and outperform strong baselines for this task.

The human visual system uses cast shadows to judge the three-dimensional layout of an object. The purpose of this installation is to demonstrate novel visual illusions of depth and transparency for paper materials, which is induced by the conventional light projection of cast shadow patterns. Thus, this installation focuses on perceptual rather than technical aspects. Illuminating a target object, the spatial vicinity of the object is darkened to produce the visual impression of a shadow of the object. By controlling the blurriness of the cast shadow and/or spatial distance between the object and its shadow, a perceptual change in the layout of the object is induced. The audience can interactively enjoy visual experiences wherein objects and letters on a paper perceptually float in the air. We also demonstrate that it is possible to edit the material appearance of a real object by manipulating the shape of the shadow; an opaque colored paper appear to be a transparent color film floating in the air.

With the increasing number of scientific publications, research paper recommendation has become increasingly important for scientists. Most researchers rely on keyword-based search or following citations in other papers, in order to find relevant research articles. And usually they spend a lot of time without getting satisfactory results. This study aims to propose a personalized research paper recommendation system, that facilitate this task by recommending papers based on users' explicit and implicit feedback. The users will be allowed to explicitly specify the papers of interest. In addition, user activities (e.g., viewing abstracts or full-texts) will be analyzed in order to enhance users' profiles. Most of the current research paper recommendation and information retrieval systems use the classical bag-of-words methods, which don't consider the context of the words and the semantic similarity between the articles. This study will use Recurrent Neural Networks (RNNs) to discover continuous and latent semantic features of the papers, in order to improve the recommendation quality. The proposed approach utilizes PubMed so far, since it is frequently used by physicians and scientists, but it can easily incorporate other datasets in the future.

Red Shift is a new design pattern for implementing parsers. The pattern draws ideas from traditional shift-reduce parsing as well as procedural PEG parsers. Red Shift parsers behaves like shift-reduce parsers, but eliminate ambiguity by always prioritizing reductions over shifts. To compensate the resulting lack of expressivity, reducers are not simple reduction rules but full-blown procedures written in a general-purpose host language. I found many advantages to this style of parser. In particular, we can generate high-quality error messages more easily; and compose different style of parsers. I also speculate about how Red Shift parsers may improve partial compilation in the context of an IDE.

For the past few years most published research on recommendation algorithms has been based on deep learning (DL) methods. Following common research practices in our field, these works usually demonstrate that a new DL method is outperforming other models not based on deep learning in offline experiments. This almost consistent success of DL based models is however not observed in recommendation-related machine learning competitions like the challenges that are held with the yearly ACM RecSys conference. Instead the winning solutions mostly consist of substantial feature engineering efforts and the use of gradient boosting or ensemble techniques. In this paper we investigate possible reasons for this surprising phenomenon. We consider multiple possible factors such as the characteristics and complexity of the problem settings, datasets, and DL methods; the background of the competition participants; or the particularities of the evaluation approach.

In order to defend neural networks against malicious attacks, recent approaches propose the use of secret keys in the training or inference pipelines of learning systems. While this concept is innovative and the results are promising in terms of attack mitigation and classification accuracy, the effectiveness relies on the secrecy of the key. However, this aspect is often not discussed. In this short paper, we explore this issue for the case of a recently proposed key-based deep neural network. White-box experiments on multiple models and datasets, using the original key-based method and our own extensions, show that it is currently possible to extract secret key bits with relatively limited effort.

Most existing Augmented Reality (AR) applications use either template (picture) markers or bar-code markers to overlay computer-generated graphics on the real world surfaces. The use of template markers is computationally expensive and unreliable. On the other hand, bar-code markers display only black and white blocks; thus, they look uninteresting and uninformative. In this short paper, we describe a new way to optically hide a QR code inside a tile based colour picture. Each AR marker is built from hundreds of small tiles (just like tiling a bathroom), and the unique gaps between the tiles are used to determine the elements of the hidden QR Code. This novel type of AR marker presents not only a realistic-looking colour picture but also contains self-Correcting information (stored in QR code). In this article, we demonstrate that this tile based colour picture with hidden QR code is relatively robust under various conditions and scaling. We believe many nowadays' AR challenges could be solved with this type of marker. AR-enabled medias could then be easily generated. For instance, it would be capable of storing and displaying virtual figures of an entire book or magazine. Thus, it provides a promising AR approach to be used in many different AR applications; and beyond, it may even replace the barcodes and QR Codes in some cases.

In this position paper we try to extend the discussion about the human augmentation of the mind through reasoning. Memory seems favored but both of them play a major role in the mind, we want to equilibrate it. Those two components of our cognitive system are actually quite intertwined with sometimes similar properties and goals. Both are failing, in many ways, that is why we are interested in it, a slight augmentation to those abilities would have huge impact. Therefore, we made a first attempt to propose an approach of a system capable of augmenting the reasoning of his user. First by sensing his context and detect a kind of cognitive bias for many domains through wearable devices and natural language processing. Then we propose a radical method to debiasing that could be used in different scenarios thanks to augmented reality. Finally, those propositions are a beginning of a much needed bigger work that confronts many challenges discussed at the end.

As robots start to enter our everyday lives, they will bring with them the risk of privacy invasions. Unlike videoconferencing, we might not have control of where the sensors on our robots look, and where the robots go. They might operating in or homes and offices while we are not there, opening the door to an invasion of our privacy that no previous technology has been capable of. How do people think about privacy in terms of robots? Are they worried about their privacy being violated and, if so, in what ways? How does the law view the issues around privacy and robotics? What are the pressing questions that we need to address now, before it's too late. This workshop will bring together researchers from a wide variety of communities to look at these questions, identify others, and help define the new area of Privacy-Sensitive Robotics. The workshop will result in a white paper that defines the pressing issues in Privacy-Sensitive Robotics, from a variety of perspectives, and will suggest a research agenda to address these problems.

Online learning to rank (OLTR) uses interaction data, such as clicks, to dynamically update rankers. OLTR has been thought to capture user intent change overtime - a task that is impossible for rankers trained on statistic datasets such as in offline and counterfactual learning to rank. However, this feature has never been demonstrated and empirically studied, as previous work only considered simulated online data with single user intent or real online data with no explicit notion of intents and how they change over interactions. In this paper, we address this gap by study the capability of OLTR algorithms to adapt to user intent change. Our empirical experiments show that the adaptation to intent change does vary across OLTR methods, and is also dependent on the amount of noise in the implicit feedback signal. This is an important result, as it highlights that intent change adaptation should be studied alongside online and offline performance. Investigating how OLTR algorithms adapt to intent change is challenging as current LTR datasets do not explicitly contain the required intent data. Along with the main findings reported in this paper related to intent change, we also contribute a methodology to investigate this aspect of OLTR methods. Specifically, we create a collection for OLTR with explicit intent change by adapting an existing TREC collection to this task. We further introduce methods to model and simulate click behaviour related to intent change. We further propose novel evaluation metrics tailored to study different aspects of how OLTR methods adapt to intent change.

Problem decomposition is an important part of many state-of-the-art Evolutionary Algorithms (EAs). The quality of the decomposition may be decisive for the EA effectiveness and efficiency. Therefore, in this paper, we focus on the recent proposition of Linkage Learning based on Local Optimization (3LO). 3LO is an empirical linkage learning (ELL) technique and is proven never to report the false linkage. False linkage is one of the possible linkage defects and occurs when linkage marks two independent genes as a dependent. Although thanks to the problem decomposition quality, the use of 3LO may lead to excellent results, its main disadvantage is its high computational cost. This disadvantage makes 3LO not applicable to state-of-the-art EAs that originally employed Statistical-based Linkage Learning (SLL) and frequently update the linkage information. Therefore, we propose the Direct Linkage Empirical Discovery technique (DLED) that preserves 3LO advantages, reduces its costs, and we prove that it is precise in recognizing the direct linkage. The concept of direct linkage, which we identify in this paper, is related to the quality of the decomposition of overlapping problems. The results show that incorporating DLED in three significantly different state-of-the-art EAs may lead to promising results.

A core challenge of evolutionary search is the need to balance between exploration of the search space and exploitation of highly fit regions. Quality-diversity search has explicitly walked this tightrope between a population's diversity and its quality. This paper extends a popular quality-diversity search algorithm, MAP-Elites, by treating the selection of parents as a multi-armed bandit problem. Using variations of the upper-confidence bound to select parents from under-explored but potentially rewarding areas of the search space can accelerate the discovery of new regions as well as improve its archive's total quality. The paper tests an indirect measure of quality for parent selection: the survival rate of a parent's offspring. Results show that maintaining a balance between exploration and exploitation leads to the most diverse and high-quality set of solutions in three different testbeds., Comment: Proceedings of the Genetic and Evolutionary Computation Conference, 9 pages, 6 figures, 3 tables

Discrete models such as cellular automata may be ported from one platform or language onto another to improve performances, for instance by rewriting legacy Matlab code into C++ or adding optimizations into a Python implementation. Although such transformations can offer benefits such as scalability or maintainability, they also have the risk of introducing bugs. While standard verification techniques can always be applied, this situation presents a unique opportunity since the two implementations can be directly compared based on their simulation runs. Although comparing average results across runs of a same configuration is a common practice, our paper shows that many bugs would not be detected at this aggregate level. We thus propose comparing implementations of cellular automata by analyzing their outputs as images. In this paper, we examine the detection of several implementation errors using five different techniques (supervised/unsupervised image processing, decision trees, random forests, or deep learning) across three different cellular automata models (forest fire, tumor, HIV). We show that in some models, random forests can detect 4 out of 5 erroneous runs, although the accuracy depends both on the model and on the nature of the errors.

This paper presents a novel method to retrieve similar scenes in soccer videos with weak annotations via multimodal use of bidirectional long short-term memory (BiLSTM). The significant increase in the number of different types of soccer videos with the development of technology brings valid assets for effective coaching, but it also increases the work of players and training staff. We tackle this problem with a nontraditional combination of pre-trained models for feature extraction and BiLSTMs for feature transformation. By using the pre-trained models, no training data is required for feature extraction. Then effective feature transformation for similarity calculation is performed by applying BiLSTM trained with weak annotations. This transformation allows for highly accurate capture of soccer video context from less annotation work. In this paper, we achieve an accurate retrieval of similar scenes by multimodal use of this BiLSTM-based transformer trainable with less human effort. The effectiveness of our method was verified by comparative experiments with state-of-the-art using actual soccer video dataset.

This paper presents Deepsy, a Natural Language-based synthesizer to assist source code analysis. It takes English descriptions of to-be-found code patterns as its inputs, and automatically produces ASTMatcher expressions that are directly usable by LLVM/Clang to materialize intended code analysis. The code analysis domain features profuse complexities in data types and operations, which make it elusive for prior rule-based synthesizers to tackle. On the other hand, machine learning-based solutions are neither applicable due to the scarcity of well labeled examples. This paper presents how Deepsy addresses the challenges by leveraging deep Natural Language Processing (NLP) and creating a new technique named dependency tree-based co-evolvement.Deepsy features an effective design that seamlessly integrates Natural Language dependency analysis into code analysis and meanwhile synergizes it with type-based narrowing and domain-specific guidance. Deepsy achieves over 70.0% expression-level accuracy and 85.1% individual API-level accuracy, significantly outperforming previous solutions.

This paper mainly studies the implementation of diverse conversation generation system based on end-to-end neural networks and sentence classification on sentence function. In existing work, the output of the conversational system is mostly in the form of one type of sentence, for instance, declarative sentence. The type of sentence function is not well distributed. There is still insufficient diversity in the output of conversational system, which could be unattractive to users. A good conversational system could well interact with users by generating diverse output, including asking and responding, driving conversations to go further. Generating different type of output sentence is necessary to conversational systems, which is also challenging. Therefore, this paper introduces the idea of diverse conversation into the generative system, and designs the Diverse Conversation Generation (DCG) model. The model adopts a sentence function classifier trained independently to supervise the model output with modified loss function and back-propagation. The DCG model increases the diversity of output sentence, which could guide user to chat more with the system, extend the quality of conversation, and improve the user experience. The model is experimented on two different sequence-to sequence models, evaluated with perplexity and classify entropy, achieves better performance compared with two base models.

Combining deep reinforcement learning with portfolio management is one of the research hotspots in the field of quantitative investment. In recent years, quantitative investment has become one of the research hotspots in the field of deep reinforcement learning. Combining portfolio management with deep reinforcement learning has become a hot research direction. The model in this paper combines the perception ability of deep learning and the online decision-making ability of reinforcement learning to construct an intelligent model of portfolio management based on deep reinforcement learning DDPG algorithm. This model addresses issues such as the limited ability of reinforcement learning models to perceive complex environments and the strong correlation between the internal features of state data, introducing the convolutional neural network into the DDPG model. The use of convolutional neural networks to process multi-dimensional state data improves the perception ability of the reinforcement learning model and solves the problem of data correlation. Therefore, the model in this paper can obtain the dynamic optimal investment portfolio within the specified trading time range. At the same time, this paper introduces the action vector into the state value representing the stock market environment, and solves the problem of static model. Experiments show that our model improves the income acquisition ability of the portfolio management model.

Model-Based testing (MBT) is a technique to derive test cases from models of the system under test (SUT), and the UML is the leading and de-facto standard language for system models both in industry and academia. This paper describes the results of a general systematic literature review (SLR) on the topic of generating test cases based on UML diagrams. By using selection and exclusion criteria, 62 primary studies were chosen from 443 identified publications, which ranged from 1999 to 2019. This study analysed the approaches and outcomes of the reviewed papers. The analysis is mainly focused on the model type, intermediate formats used and coverage criteria. Based on the review results, research trends and deficiencies are identified, and future works are proposed.

Based on the background of "ZTE event" in China-US trade friction, in order to effectively analyze the stock trend of communication industry in the securities market, this paper uses the long short-term memory (LSTM) neural network to establish a deep learning model, taking stock of Zhongxing Telecommunication Equipment Corporation as the specific research object, and predicts the stock price of ZTE within two years after the outbreak of "ZTE event" The prediction results are compared with the actual stock price and the experimental results show that the proposed method can effectively predict the development trend of the stock. Based on the research results, this paper analyzes the future development of China's communication industry under the background of China-US trade friction, and puts forward the corresponding industry development strategy, which provides reference for the development strategy of Chinese communication enterprises.

In recent years, with the rapid development of deep learning technology and its wide application in the field of computer vision, various image understanding tasks including semantic segmentation and instance segmentation have made great progress, and people's further demand for image understanding has spawned a more comprehensive task image panoptic segmentation. Image panoptic segmentation can be seen as the combination of semantic segmentation and instance segmentation. For uncountable object categories (called stuff), the pixel category is distinguished. For countable object categories (called things), not only the semantic category of the target is recognized, but also each instance is distinguished. This task can provide more comprehensive scene information, and can be widely used in the understanding of various natural scenes. This paper investigate the commonly used panoptic segmentation methods, including the basic shared feature extraction method, the information combination method between semantic segmentation and instance segmentation sub-tasks, and the learnable method to remove the overlap between instances. This paper also summarize the commonly used panoptic segmentation datasets and the evaluation metrics, then the experimental performance evaluation results of various methods on commonly used datasets are showed. Finally, this paper summarize the general direction of panoptic segmentation, and predict the future research direction.

The 3D point cloud data contains the semantic information of the measured scene object, and has the characteristics of massive, high density and high precision. It has become a main data type used to understand, analyze and semantically explain the 3D natural scene. This paper takes indoor 3D point cloud data as the research object and studies the semantic segmentation of 3D point cloud. Based on the in-depth study of PointNet++ network, this paper proposes a local information aggregation module based on graph network to optimize PointNet++ network. The local information aggregation module based on graph network proposed in this paper uses graph network learning to explore the mutual influence of points in a local area, and then aggregate local information to generate more representative local features. Finally, we evaluated our model on S3DIS and compared it quantitatively with PointNet++ to show the improved performance of the original algorithm.

Individual identification of rare animals in different scenarios helps to accurately grasp the number and scale of animal populations, so as to formulate appropriate protection measures and policies, which are beneficial to maintaining human ecological balance and sustainable development. In this paper, we study the individual identification methods of leopard in multiple scenes, obtain the leopard image data through various channels such as zoos and protected areas, and construct a dataset, which obtains 28,751 images of leopard. In order to extract more fine-grained features of the leopard, Res2net was added to the network, at the same time, the self-attention mechanism was added to improve the global correlation. A R2NMGN method based on the self-attention mechanism was proposed. The experimental results show that the method proposed in this paper is significantly better than other methods in individual recognition of leopard.

Wildlife is an important biological resource in China. Classifying images of wildlife through computer technology can help people identify wildlife, which is of great significance to help people understand and protect wildlife. Therefore, this issue is worth studying. Traditional methods mostly use standard Convolutional Neural Networks (CNN) to classify wild animal images, but these methods have disadvantages such as slow computing speed, long time consumption and low accuracy. With an attempt to address such issues, this paper proposes a method based on transfer-learning for classifying wild animal images. By using the pre-trained model it can save a lot of training time. The experimental results on Oregon Wildlife, using a public wildlife data set, show that the method proposed in this paper achieved 99.01% accuracy and is 57.82% more accurate than the standard Convolutional Neural Networks (CNN) method. Moreover, in terms of running time, the method presented in this paper has achieved higher efficiency, and the training time is 50% shorter than the standard method, which proves the superiority of the method proposed in this paper.

Oracle bone inscriptions, as the earliest writing system in China, are of great cultural and academic value. In recent years, deep learning method has been widely used in the process of oracle bone character examination and interpretation, and has achieved certain effect in oracle bone character recognition. However, little progress has been made in the unknown oracle bone character recognition. To solve these problems, this paper collects four types of characters: oracle bone script, jinwen, Seal script and official script as the data set, and uses the similarity between the characters and the depth network model to build the oracle bone character detection and recognition model. Then, this paper improved the Tiny-YOLOv4 model to realize the function of unknown oracular character detection and enhanced the detection speed by changing the NMS (non-maximum Suppression). The experiment proved that the detection and recognition accuracy of oracle bone characters reached 67.28%, and the detection speed of 2.281ms was improved by using NMS strategy.

In this paper, we employ a Convolutional Neural Network (CNN) in predicting physician gaze. This paper focuses on two aspects – one comparison between hand-crafted features and CNN-based learned features, and two in investigating the impact of fully-connected layers in an end-to-end CNN model. The pre-trained CNN model based on VGG16 through transfer learning is used as a feature extractor and a K-Nearest Neighbor and a Random Forest (RF) algorithm were used as the classifier of physician gaze. The CNN-RF and CNN–K-NN models were compared with the traditional end-to-end CNN model and through a series of experiments and statistical tests of significance, we show that the power of CNN comes from the features extraction part and that the fully connected layers of the CNN have comparable performance to the random forest and the k-NN classifiers. We also show that the CNN-based learned features provide substantial distinguishable power in classifying physician gaze.

Ransomware is a class of malware that gains unauthorized admittance to organizational and personal assets and blocks their access to legitimate users and staff. It accomplishes this by encrypting the algorithms with a strong encryption technique. The Hacker demands a ransom to allow access to the asset and revert the control to the authorized organizations. Many individuals and Organizations have been victim to such Ransomware Attacks. With the mounting growth of technology, such attacks have increased exponentially and the need for adequate preventive techniques are also increasing. This paper reviews the current state-of-the-art Ransomware detection techniques implemented to resist such attacks. While most surveys focus on ransomware detection using Machine learning techniques, this survey concentrates on Ransomware techniques that apply the sub-domains of Artificial Intelligence. This paper reviews detection techniques that employ Machine Learning, Deep Learning and Natural Language Processing techniques, which are sub-fields of AI for ransomware detection. Though the reviewed works have good detection rates, a few open concerns are identified and discussed. The primary objective of the review is to highlight concerns of the proposed system that prevent it from becoming effective in real-time systems.

Deep Learning and Natural Language Processing are branches of modern technology that are fast being used to solve myriads of problems that inflict us in our daily life. A particular problem that has the potential to benefit greatly from the capabilities of data mining and machine learning is the issue of unpredictability in the stock market environment. What makes this a difficult problem is that stock volume movements are influenced by a variety of factors some of which are inherently quantifiable while others such as trader sentiments are not. The system proposed in this paper combines Fundamental and Technical trading philosophies in predicting stock volume movements, during day trading, based on historical stock performance data, financial news, and trader sentiments. Financial news articles, for a stipulated time period, are collected and filtered based on the companies mentioned in the articles. For this paper, we have chosen to filter and retain articles about companies belonging to SENSEX 50. For gauging trader sentiments with respect to the news about a company or the company in general, Twitter tweets are considered as a data source. Sentiment analysis is performed for the news and cumulated tweets for a company separately to arrive at two polarity scores that indicate the sentiments carried. Since recurrent neural networks have the ability to store additional states which can be used as a memory store in order to maintain and dismiss information based on data patterns, an LSTM (Long Short Term Memory Networks) model is developed to predict the stock volume movement based on the polarity scores of the news and tweets, and OHLC (Open, High, Low, Close) price values.

Deaf people are a considerable part of the world population and communicate naturally using sign languages. However, although many countries adopt their sign language as an official language, there are linguistics barriers to accessing fundamental rights, especially access to health services, even more critical situation in the midst of the COVID-19 crisis. This situation has been the focus of some government policies that oblige essential service providers to provide sign language interpreters to assist Deaf people. However, this type of solution has high operating costs, mainly to serve the entire Deaf community in all environments. These setbacks motivate the investigation of methodologies and automated tools to support this type of problem. In this paper, we address this problem by proposing two-stream model for the recognition of the Brazilian Sign Language (Libras) in the health context. The proposed solution does not use any additional capture sensor or hardware, being entirely base on images or sequences of images (videos). The results show that the solution is able to recognize the Libras signs in the test dataset reasonably well, achieved an average accuracy of approximately 96,12% considering a scenario where the interpreter used in the test set was not used in the training set, which shows that there are good evidence that it can assist in the communication process with Deaf people. Besides, an additional contribution of this paper is the introduction of a new dataset in the Brazilian sign language (Libras) containing 5000 videos of 50 signs in the health context, which may assist the development and research of other solutions.

The constant expansion of e-commerce, recently boosted due to the coronavirus pandemic, has led to a huge increase in online shopping. More and more, customers demand online reviews of products and comments on the Web to make decisions about buying a product over another. In this context, sentiment analysis techniques constitute the traditional way to summarize user's opinions that criticizes or highlights the positive aspects of a product. Sentiment analysis of reviews usually relies on extracting positive and negative aspects of products, neglecting comparative opinions. Such opinions do not directly express a positive or negative view but contrast aspects of products from different competitors. In this paper, we present the first effort towards detecting comparative sentences in Portuguese. Identifying comparative sentences is a key task for companies to know how users are comparing a product with their competitors and is essential for developing sentiment summarization applications for the end user. In addition, we present a supervised approach to automatically detect Portuguese comparative sentences, classifying them into five distinct groups: (1) Non-Comparative, (2) Non-Equal Gradable, (3) Equative, (4) Superlative e (5) Non-Gradable. To that end, this paper provides three main contributions: (1) a Portuguese lexicon list with words used to make comparisons; (2) two new Portuguese datasets with comparative sentences; and (3) a hierarchical approach for detecting multiple comparisons and classify the sentences in different groups by using state-of-art classification algorithms, reaching an accuracy of 87%.

The recent growth of social networks led to an increase in the dissemination of hate speech online and, as a consequence, studies about hate speech detection have emerged. Part of them are based on supervised learning techniques, which require large labeled databases, however this is still a scarce resource. Most of the existing databases on hate speech consist of texts in English. Although there are some databases in Portuguese, they are rather small with few examples, a factor which can limit both the performance of simple deep neural networks and the usage of more complex architectures. In order to work around this setback, this paper analyses the usage of Data Augmentation in texts to improve the training performance of recursive neural networks (LSTM) and convolutional neural networks (CNN) applied to the hate speech detection task. Data Augmentation is a regularization process used on deep neural networks to avoid overfitting. These techniques are common in the field of computer vision, however, due to the complexity of natural languages, this process is not as frequently used in tasks involving texts. In this paper, we experimented on a public hate speech database in Portuguese using Data Augmentation techniques such as text generation from the original database to enhance the results. The experiments show an improvement in the results, demonstrating that the techniques are promising.

Trust can be employed for finding reliable information in Online Social Networks (OSNs). Since users in OSNs may intentionally change their behavior over time (in some cases for deceiving other users), modeling (pair-wise) trust relations in such complex environment is a challenging task. However, most of the existing trust prediction approaches assume that trust relations are fixed over time and they fail to capture the dynamic behavior of users in OSNs. In this paper, we propose a dynamic deep trust prediction model. As the impact of incidental emotions on trust has been proven in psychology studies, in this paper, we also study this impact on our trust prediction approach. First, we propose a novel deep structure that incorporates users' emotions and their textual contents in OSNs. Second, we use embeddings to represent the users and their self-descriptions provided. Finally, considering different time windows, we dynamically predict pair-wise trust relations. To evaluate our approach, we collected a large twitter dataset. The evaluation results demonstrate the effectiveness of our approach compared to the state-of-the-art approaches.

The captcha is a Turing test used to distinguish between machines and humans. It is considered as the verification code for the security on many websites. In recent years, deep learning has been widely used in the related field such as data analysis and computer vision. By applying deep learning method to captcha recognition, the vulnerabilities of the current captcha can be discovered, and the captcha can be further improved to reach a higher security level. This paper focuses on the captcha recognition based on deep learning, which is conducive to find the loopholes of current captcha to design more secure captcha. In this paper, the AlexNet neural network is analyzed in detail first, then the model is implemented to achieve captcha recognition for the simple and complex captcha data. The experiment results show that the recognition accuracy of the simple captcha and complex captcha data is close to 99%, and captcha recognition based on deep learning has a satisfied effect for recognition of captcha with different complexity.

With the rapid development of artificial intelligence technology and autonomous navigation technology, the unmanned surface vessel (USV) industry has developed accordingly, and it has played an important role in the fields of water quality monitoring, maritime inspection, and maritime safety assurance. However, USV is easily affected by the external lighting environment. In the case of insufficient lighting, the collected images have the characteristics of low brightness, low contrast and low resolution, and are extremely susceptible to external noise interference, making USV difficult obtain input requirements that meet the visual tasks such as target recognition and semantic segmentation. In this paper, we propose a deep learning-based low-light image enhancement and noise suppression method (LENet). Specifically, LENet is used to map the low-light image to the normal-light image through a deep Unet network, and CBM3D further suppresses the interference noise in the image to achieve the enhancement of the low-light image. We enhance the generalization ability and robustness of the deep network by embedding dilated convolutions and dense blocks in the deep Unet network. Structural similarity (SSIM) and norm are used as the loss function to further improve the quality of the enhanced image. The experimental results show that the deep network proposed in this paper improves the brightness and contrast of the images collected by the USV under insufficient lighting conditions, which can meet the input requirements of the USV visual task.

Crowd counting, which is one of the primary research lines of computer vision, has achieved significant advancement due to the affordable computational cost, and it is a particularly useful means of ensuring public security in public places, especially in crowded places. In recent years, crowd counting methods based on public places have emerged one after another given crowd congestion. Firstly, this paper introduces in detail from traditional approaches to deep learning approaches, focusing on crowd counting approaches based on the convolution neural network (CNN), and compares and analyzes the advantages and disadvantages of each approach. Next, this paper summarizes the commonly used datasets and the main indicators of evaluating crowd counting algorithms. Finally, this paper expounds on the challenges existing in the field of crowd counting and the possible research directions in the future.

Image steganalysis is a very important technology for forensics. Recent studies show that the idea of steganalysis based on Convolutional Neural Network (CNN) is feasible. In this paper, we propose a novel digital image steganalysis model based on CNN. Compared with the existing CNN-based methods, the CNN model proposed to this paper has two characteristics. First, in the front of the network, high-pass filter in SRM is used to initialize the convolution kernels, which is beneficial to learning steganography noise in the image. Second, in the middle of the network, the residual learning mechanism is used to enhance the convergence speed and stability of the network. Experiments on the standard data set show that the proposed CNN model can detect S-UNIWARD steganography algorithm with high accuracy.

Location estimation of Autonomous mobile robots is an essential and challenging task, especially for indoor applications. Despite the many solutions and algorithms that have been suggested in the literature to provide a precise localisation technique for mobile robots, it continues to be an open research problem and worth further study. In this paper, a predefined map with artificial colour code signature (CCs) beacons are used to build an effective algorithm to achieve an indoor localisation and position prediction of an omnidirectional mobile robot. This algorithm is primarily based on calculating the distance between the robot and the beacon using Pixy cameras, as vision sensors; then, estimating the position of the robot using a trilateration method. By comparing the results obtained in this paper with the mathematically obtained results, it is clearly shown that the robot effectively follows the localisation algorithm to estimate its pose (position and orientation), improving its localisation abilities in addition to obtaining its initial position. Furthermore, the limitations associated with using Pixy cameras are discussed in this paper as well.

In this paper, we present the outcomes of the first user study designed and evaluated in the context of the Smart City Robotics Challenge (SciRoc Challenge). The study presented in this paper has the main novelty of having been devised and implemented in a realistic environment: a robot competition where robot tasks were developed by participant teams, robots were fully autonomous, and user questionnaires were part of the competition score. Specifically, our study was performed over a scenario configured to instruct a robot to take an elevator of a shopping mall asking for customers support. Leveraging the dedicated questionnaire designed for the tested scenario, we validated the experimental hypothesis if user perception of robots' behaviour may be influenced by the user's gender. In the end, we discuss the results of our study.

COVID pandemic has changed the way in which we carry out our daily life and also have affected educational processes. Teaching and learning have changed from a most common face to face context to a blended or online context. This implies changes in the way to carry out the activates and have an impact in research projects such as RoboSTEAM. Such project, that applies Challenge Based Learning methodologies with application of Robotics and Mechatronics, requires to change its approach to show how it is possible to succeed in the new situation. This paper describes how the project has evolved, how it has been affected by COVID and the possible changes to carry out. Regarding this last issue remote labs and online tools are presented as solutions to support changes in the application of challenge-based learning methodology. This paper is supported by ROBOSTEAM Erasmus+ KA201 Project with reference 2018-1-ES01-KA201-050939 info:eu-repo/semantics/publishedVersion

In business domains, \textit{bundling} is one of the most important marketing strategies to conduct product promotions, which is commonly used in online e-commerce and offline retailers. Existing recommender systems mostly focus on recommending individual items that users may be interested in. In this paper, we target at a practical but less explored recommendation problem named bundle recommendation, which aims to offer a combination of items to users. To tackle this specific recommendation problem in the context of the \emph{virtual mall} in online games, we formalize it as a link prediction problem on a user-item-bundle tripartite graph constructed from the historical interactions, and solve it with a neural network model that can learn directly on the graph-structure data. Extensive experiments on three public datasets and one industrial game dataset demonstrate the effectiveness of the proposed method. Further, the bundle recommendation model has been deployed in production for more than one year in a popular online game developed by Netease Games, and the launch of the model yields more than 60\% improvement on conversion rate of bundles, and a relative improvement of more than 15\% on gross merchandise volume (GMV)., Comment: 8 pages, 10 figures, accepted paper on CIKM 2020

The theory of multiple intelligences is a new concept of intelligence based on the traditional theory of intelligence. it has an important impact on contemporary education and plays an important role in the design of deep learning. This paper discusses the positive influence of multiple intelligences theory on deep learning. Through case analysis, this paper illustrates the feasibility of the application of multiple intelligences theory in deep learning design from four aspects: creating learning situations, using multiple teaching strategies, carrying out collaborative learning, and designing multiple evaluation methods.

A goal model family is a set of related goal models that conform to the same metamodel, with commonalities and variabilities between models. Such families stem from the evolution of initial models into several versions over time and/or the variation of models over the space dimension (e.g., products). In contexts where there are several versions/variations of a goal model, analyzing a set of related models with typical similarities, one model at a time, often involves redundant computations and may require repeated user assistance (e.g., for interactive analysis) and laborious activities. This paper proposes the use of union models as first-class artifacts to analyze families of goal models, in order to improve performance of language-specific analysis procedures. The paper empirically evaluates the performance gain resulting from adapting (or lifting) an existing analysis technique specific to the Goal-oriented Requirement Language (GRL) to a family of GRL models, all at once using a union model, compared to analyzing individual models. Our experiments show, based on the use of the IBM CPLEX optimizer, the usefulness and performance gains of using union models to perform a computationally expensive analysis, namely quantitative backward propagation, on families of GRL models.

In this paper, the problem of person-independent facial expression recognition (FER) is addressed using the 3D geometry information extracted from the 3D shape of the face. The 3D data can provide more information and more robust features under some conditions than the 2D data. State-of-the-art 3D FER methods are often based on a single descriptor which may fail to handle the large inter-class and intra-class variability of the human facial expressions. For this reason, we propose to explore the usage of covariance matrices of features, instead of directly using the local features for the 3D FER task. Since covariance matrices are elements of the non-linear manifold of Symmetric Positive Definite (SPD) matrices, we propose, in this paper, a manifold-based classification using a Graph-Matching Kernel method. The proposed Kernel is appropriate for SVM-based image classification, and thus is suitable for FER application. We evaluate the performance of the proposed method on the BU-3DFE and the Bosphorus datasets, and demonstrate its superiority compared to the state-of-the-art methods.

This paper contrasts and analyzes the image enhancement processing algorithm, median filtering and histogram equalization in the face recognition preprocessing process. The advantages and use methods of the OpenCV machine vision library are analyzed. The analyzed algorithm is implemented in the development environment of VC ++ 6.0 and MATLAB, and experiments are conducted on the ORL face library. The experimental results show that the algorithm used in this paper has obvious effect on processing face samples, has strong pertinence, and is suitable for preprocessing face images, and the OpenCV machine vision library can be used as a development tool for face recognition systems.