In this paper, we developed the latest IoT honeypots to capture IoT malware currently on the loose, analyzed IoT malware with new features such as persistent infection, developed malware removal methods to be provided to IoT device users. Furthermore, as attack behaviors using IoT devices become more diverse and sophisticated every year, we conducted research related to various factors involved in understanding the overall picture of attack behaviors from the perspective of incident responders. As the final stage of countermeasures, we also conducted research and development of IoT malware disabling technology to stop only IoT malware activities in IoT devices and IoT system disabling technology to remotely control (including stopping) IoT devices themselves.

Since the outbreak of IoT malware “Mirai,” several incidents have occurred in which IoT devices have been infected with malware. The malware targets IoT devices whose Telnet and SSH services are accessible from the Internet and whose ID/Password settings are not strong enough. Several IoT malware families, including Mirai, are also known that restrict access to Telnet and other services to keep the devices from being infected by other malware after infection. However, tens of thousands of devices in Japan can be still accessed Telnet services over the Internet according to network scan results. Does this imply that these devices can avoid malware infection by setting strong enough passwords, and thus cannot be used as a stepping stone for cyber attacks? In February 2019, we initiated the National Operation Toward IoT Clean Environment (NOTICE) project in Japan to investigate IoT devices with weak credentials and notify the device users. In this study, we analyze the results of the NOTICE project from February 2021 to May 2021 and the results of the large-scale darknet monitoring to reveal whether IoT devices with weak credentials are infected with malware or not. Moreover, we analyze the IoT devices with weak credentials to find out the factors that prevent these devices from being infected with malware and to assess the risk of abuse for cyber attacks. From the results of the analysis, it is discovered that approximately 2,000 devices can be easily logged in using weak credentials in one month in Japan. We also clarify that no device are infected with Mirai and its variants malware due to lack of functions used for malware infection excluding only one host. Finally, even the devices which are logged in by NOTICE project are not infected with Mirai, we find that at least 80% and 93% of the devices can execute arbitrary scripts and can send packets to arbitrary destinations respectively.

Memory corruption can modify the kernel data of an operating system kernel through exploiting kernel vulnerabilities that allow privilege escalation and defeats security mechanisms. To prevent memory corruption, the several security mechanisms are proposed. Kernel address space layout randomization randomizes the virtual address layout of the kernel. The kernel control flow integrity verifies the order of invoking kernel codes. The additional kernel observer focuses on the unintended privilege modifications. However, illegal writing of kernel data is not prevented by these existing security mechanisms. Therefore, an adversary can achieve the privilege escalation and the defeat of security mechanisms. This study proposes a kernel data protection mechanism (KDPM), which is a novel security design that restricts the writing of specific kernel data. The KDPM adopts a memory protection key (MPK) to control the write restriction of kernel data. The KDPM with the MPK ensures that the writing of privileged information for user processes and the writing of kernel data related to the mandatory access control. These are dynamically restricted during the invocation of specific system calls and the execution of specific kernel codes. Further, the KDPM is implemented on the latest Linux with an MPK emulator. The evaluation results indicate the possibility of preventing the illegal writing of kernel data. The KDPM showed an acceptable performance cost, measured by the overhead, which was from 2.96% to 9.01% of system call invocations, whereas the performance load on the MPK operations was 22.1ns to 1347.9ns. Additionally, the KDPM requires 137 to 176 instructions for its implementations.

Information leakage is a significant threat to organizations, and effective measures are required to protect information assets. As confidential files can be leaked through various paths, a countermeasure is necessary to prevent information leakage from various paths, from simple drag-and-drop movements to complex transformations such as encryption and encoding. However, existing methods are difficult to take countermeasures depending on the information leakage paths. Furthermore, it is also necessary to create a visualization format that can find information leakage easily and a method that can remove unnecessary parts while leaving the necessary parts of information leakage to improve visibility. This paper proposes a new information leakage countermeasure method that incorporates file tracking and visualization. The file tracking component recursively extracts all events related to confidential files. Therefore, tracking is possible even when data have transformed significantly from the original file. The visualization component represents the results of file tracking as a network graph. This allows security administrators to find information leakage even if a file is transformed through multiple events. Furthermore, by pruning the network graph using the frequency of past events, the indicators of information leakage can be more easily found by security administrators. In experiments conducted, network graphs were generated for two information leakage scenarios in which files were moved and copied. The visualization results were obtained according to the scenarios, and the network graph was pruned to reduce vertices by 17.6% and edges by 10.9%.

With the rise of social network service (SNS) in recent years, the security of SNS users' private information has been a concern for the public. However, due to the anonymity of SNS, identity impersonation is hard to be detected and prevented since users are free to create an account with any username they want. This could lead to cybercrimes like fraud because impersonation allows malicious users to steal private information. Until now, there are few studies about this problem, and none of them can perfectly handle this problem. In this paper, based on an idea from previous work, we combine blockchain technology and security protocol to prevent impersonation in SNS. In our scheme, the defects of complex and duplicated operations in the previous work are improved. And the authentication work of SNS server is also adjusted to resist single-point, attacks. Moreover, the smart contract is introduced to help the whole system runs automatically. Afterward, our proposed scheme is implemented and tested on an Ethereum test network and the result suggests that it is acceptable and suitable for nowadays SNS network.

Two problems occur when an authorization server is utilized for a use case where a different security profile needs to be applied to a unique client request for accessing a distinct type of an API, such as open banking. A security profile can be applied to a client request by using the settings of an authorization server and client. However, this method can only apply the same security profile to all client requests. Therefore, multiple authorization servers or isolated environments, such as realms of an authorization server, are needed to apply a different security profile. However, this increases managerial costs for the authorization server administration. Moreover, new settings and logic need to be added to an authorization server if the existing client settings are inadequate for applying a security profile, which requires modification of an authorization server's source code. We aims to propose the policy-based method that resolves these problems. The proposed method does not completely rely on the settings of a client and can determine an applied security profile using a policy and the context of the client's request. Therefore, only one authorization server or isolated environment, such as a realm of an authorization server, is required to support multiple different security profiles. Additionally, the proposed method can implement a security profile as a pluggable software module. Thus, the source code of the authorization server need not be modified. The proposed method and Financial-grade application programming interface (FAPI) security profiles were implemented in Keycloak, which is an open-source identity and access management solution, and evaluation scenarios were executed. The results of the evaluation confirmed that the proposed method resolves these problems. The implementation has been contributed to Keycloak, making the proposed method and FAPI security profiles publicly available.

Although security and privacy technologies are incorporated into every device and service, the complexity of these concepts confuses non-expert users. Prior research has shown that non-expert users ask strangers for advice about digital media use online. In this study, to clarify the security and privacy concerns of non-expert users in their daily lives, we investigated security- and privacy-related question posts on a Question-and-Answer (Q&A) site for non-expert users. We conducted a thematic analysis of 445 question posts. We identified seven themes among the questions and found that users asked about cyberattacks the most, followed by authentication and security software. We also found that there was a strong demand for answers, especially for questions related to privacy abuse and account/device management. Our findings provide key insights into what non-experts are struggling with when it comes to privacy and security and will help service providers and researchers make improvements to address these concerns.

Theoretically secure cryptosystems, digital signatures may not be secure after being implemented on Internet of Things (IoT) devices and PCs because of side-channel attacks (SCA). Because RSA key generation and ECDSA require GCD computations or modular inversions, which are often computed using the binary Euclidean algorithm (BEA) or binary extended Euclidean algorithm (BEEA), the SCA weaknesses of BEA and BEEA become a serious concern. Constant-time GCD (CT-GCD) and constant-time modular inversion (CTMI) algorithms are effective countermeasures in such situations. Modular inversion based on Fermat's little theorem (FLT) can work in constant time, but it is not efficient for general inputs. Two CTMI algorithms, named BOS and BY in this paper, were proposed by Bos, Bernstein and Yang, respectively. Their algorithms are all based on the concept of BEA. However, one iteration of BOS has complicated computations, and BY requires more iterations. A small number of iterations and simple computations during one iteration are good characteristics of a constant-time algorithm. Based on this view, this study proposes new short-iteration CT-GCD and CTMI algorithms over 𝔽_p borrowing a simple concept from BEA. Our algorithms are evaluated from a theoretical perspective. Compared with BOS, BY, and the improved version of BY, our short-iteration algorithms are experimentally demonstrated to be faster.

This paper introduces significant improvements over the existing cryptanalysis approaches on Salsa20 and ChaCha stream ciphers. For the first time, we reduced the attack complexity on Salsa20/8 to the lowest possible margin. We introduced an attack on ChaCha7.25. It is the first attack of its type on ChaCha7.25/20. In our approach, we studied differential cryptanalysis of the Salsa20 and ChaCha stream ciphers based on a comprehensive analysis of probabilistic neutral bits (PNBs). The existing differential cryptanalysis approaches on Salsa20 and ChaCha stream ciphers first study the differential bias at specific input and output differential positions and then search for probabilistic neutral bits. However, the differential bias and the set of PNBs obtained in this method are not always the ideal combination to conduct the attack against the ciphers. The researchers have not focused on the comprehensive analysis of the probabilistic neutrality measure of all key bits concerning all possible output difference positions at all possible internal rounds of Salsa20 and ChaCha stream ciphers. Moreover, the relationship between the neutrality measure and the number of inverse quarter rounds has not been scrutinized yet. To address these study gaps, we study the differential cryptanalysis based on the comprehensive analysis of probabilistic neutral bits on the reduced-round Salsa20 and ChaCha. At first, we comprehensively analyze the neutrality measure of 256 key bits positions. Afterward, we select the output difference bit position with the best average neutrality measure and look for the corresponding input differential with the best differential bias. Considering all aspects, we present an attack on Salsa20/8 with a time complexity of 2^241.62 and data complexity of 2^31.5, which is the best-known single bit differential attack on Salsa20/8 and then, we introduced an attack on ChaCha7.25 rounds with a time complexity of 2^254.011 and data complexity of 2^51.81.

The recent decision by the National Institute of Standards and Technology (NIST) to standardize lattice-based cryptography has further increased the demand for security analysis. The Ring-Learning with Error (Ring-LWE) problem is a mathematical problem that constitutes such lattice cryptosystems. It has many algebraic properties because it is considered in the ring of integers, R, of a number field, K. These algebraic properties make the Ring-LWE based schemes efficient, although some of them are also used for attacks. When the modulus, q, is unramified in K, it is known that the Ring-LWE problem, to determine the secret information s ∈ R/qR, can be solved by determining s (mod q) ∈ 𝔽_q^f for all prime ideals q lying over q. The χ²-attack determines s (mod q) ∈𝔽_q^f using chi-square tests over R/q ≅ 𝔽_q^f. The χ²-attack is improved in the special case where the residue degree f is two, which is called the two-residue-degree χ²-attack. In this paper, we extend the two-residue-degree χ²-attack to the attack that works efficiently for any residue degree. As a result, the attack time against a vulnerable field using our proposed attack with parameter (q,f)=(67, 3) was 129 seconds on a standard PC. We also evaluate the vulnerability of the two-power cyclotomic fields.

The proliferation of coronavirus disease (COVID-19) has prompted changes in business models. To ensure a successful transition to non-face-to-face and electronic communication, the authenticity of data and the trustworthiness of communication partners are essential. Trust services provide a mechanism for preventing data falsification and spoofing. To develop a trust service, the characteristics of the service and the scope of its use need to be determined, and the relevant legal systems must be investigated. Preparing a document to meet trust service provider requirements may incur significant expenses. This study focuses on electronic signatures, proposes criteria for classification, classifies actual documents based on these criteria, and opens a discussion. A case study illustrates how trusted service providers search a document highlighting areas that require approval. The classification table in this paper may prove advantageous at the outset when business decisions are uncertain, and there is no clear starting point.

In recent years, the market size for internet advertising has been increasing with the expansion of the Internet. Among the internet advertising technologies, affiliate services, which are a performance-based service, use cookies to track and measure the performance of affiliates. However, for the purpose of safeguarding personal information, cookies tend to be regulated, which leads to concerns over whether normal tracking by cookies works as intended. Therefore, in this study, the recent problems from the perspectives of affiliates, affiliate service providers, and advertisers are extracted, and a framework of cookie-independent measuring engagement method using access logs is proposed and open issues are discussed for future affiliate services.

The use of reports in action has grown significantly in recent decades as data has become digitized. However, traditional statistical methods no longer work due to the uncontrollable expansion and complexity of raw data. Therefore, it is crucial to clean and analyze financial data using modern machine learning methods. In this study, the quarterly reports (i.e. 10Q filings) of publicly traded companies in the United States were analyzed by utilizing data mining methods. The study used 8905 quarterly reports of companies from 2019 to 2022. The proposed approach consists of two phases with a combination of three different machine learning methods. The first two methods were used to generate a dataset from the 10Q filings with extracting new features, and the last method was used for the classification problem. Doc2Vec method in Gensim framework was used to generate vectors from textual tags in 10Q filings. The generated vectors were clustered using the K-means algorithm to combine the tags according to their semantics. By this way, 94000 tags representing different financial items were reduced to 20000 clusters consisting of these tags, making the analysis more efficient and manageable. The dataset was created with the values corresponding to the tags in the clusters. In addition, PriceRank metric was added to the dataset as a class label indicating the price strength of the companies for the next financial quarter. Thus, it is aimed to determine the effect of a company's quarterly reports on the market price of the company for the next period. Finally, a Convolutional Neural Network model was utilized for the classification problem. To evaluate the results, all stages of the proposed hybrid method were compared with other machine learning techniques. This novel approach could assist investors in examining companies collectively and inferring new, significant insights. The proposed method was compared with different approaches for creating datasets by extracting new features and classification tasks, then eventually tested with different metrics. The proposed approach performed comparatively better than the other machine learning methods to predict future price strength based on past reports with an accuracy of 84% on the created 10Q filings dataset.

With the growing global demand for seafood, sustainable aquaculture is attracting more attention than conventional natural fishing, which causes overfishing and damage to the marine environment. However, a major problem facing the aquaculture industry is the cost of feeding, which accounts for about 60% of a fishing expenditure. Excessive feeding increases costs, and the accumulation of residual feed on the seabed negatively impacts the quality of water environments (e.g., causing red tides). Therefore, the importance of raising fishes efficiently with less food by optimizing the timing and quantity of feeding becomes more evident. Thus, we developed a system to quantitate the amount of fish activity for the optimal feeding time and feed quantity based on the images taken. For quantitation, optical flow that is a method for tracking individual objects was used. However, it is difficult to track individual fish and quantitate their activity in the presence of many fishes. Therefore, all fish in the filmed screen were considered as a single school and the amount of change in an entire screen was used as the amount of the school activity. We divided specifically the entire image into fixed regions and quantitated by vectorizing the amount of change in each region using optical flow. A vector represents the moving distance and direction. We used the numerical data of a histogram as the indicator for the amount of fish activity by dividing them into classes and recording the number of occurrences in each class. We verified the effectiveness of the indicator by quantitating the eating and not eating movements during feeding. We evaluated the performance of the quantified indicators by the support vector classification, which is a form of machine learning. We confirmed that the two activities can be correctly classified.

With a rapidly escalating number of sophisticated cyber-attacks, protecting Internet of Things (IoT) networks against unauthorized activity is a major concern. The detection of malicious attack traffic is thus crucial for IoT security to prevent unwanted traffic. However, existing traditional malicious traffic detection systems which relied on supervised machine learning approach need a considerable number of benign and malware traffic samples to train the machine learning models. Moreover, in the cases of zero-day attacks, only a few labeled traffic samples are accessible for analysis. To deal with this, we propose a few-shot malicious IoT traffic detection system with a prototypical graph neural network. The proposed approach does not require prior knowledge of network payload binaries or network traffic signatures. The model is trained on labeled traffic data and tested to evaluate its ability to detect new types of attacks when only a few labeled traffic samples are available. The proposed detection system first categorizes the network traffic as a bidirectional flow and visualizes the binary traffic flow as a color image. A neural network is then applied to the visualized traffic to extract important features. After that, using the proposed few-shot graph neural network approach, the model is trained on different few-shot tasks to generalize it to new unseen attacks. The proposed model is evaluated on a network traffic dataset consisting of benign traffic and traffic corresponding to six types of attacks. The results revealed that our proposed model achieved an F1 score of 0.91 and 0.94 in 5-shot and 10-shot classification, respectively, and outperformed the baseline models.

Different types of malicious attacks have been increasing simultaneously and have become a serious issue for cybersecurity. Most attacks leverage domain URLs as an attack communications medium and compromise users into a victim of phishing or spam. We take advantage of machine learning methods to detect the maliciousness of a domain automatically using three features: DNS-based, lexical, and semantic features. The proposed approach exhibits high performance even with a small training dataset. The experimental results demonstrate that the proposed scheme achieves an approximate accuracy of 0.927 when using a random forest classifier.

This paper first takes IoT as an example to provide the motivation for eliminating the single point of trust (SPOT) in a CA-based private PKI. It then describes a distributed public key certificate-issuing infrastructure that eliminates the SPOT and its limitation derived from generating signing keys. Finally, it proposes a method to address its limitation by all certificate issuers.

The art gallery problem is to find a set of guards who together can observe every point of the interior of a polygon P. We study a chromatic variant of the problem, where each guard is assigned one of k distinct colors. A chromatic guarding is said to be conflict-free if at least one of the colors seen by every point in P is unique (i.e., each point in P is seen by some guard whose color appears exactly once among the guards visible to that point). In this paper, we consider vertex-to-point guarding, where the guards are placed on vertices of P, and they observe every point of the interior of P. The vertex-to-point conflict-free chromatic art gallery problem is to find a colored-guard set such that (i) guards are placed on P's vertices, and (ii) any point in P can see a guard of a unique color among all the visible guards. In this paper, it is shown that determining whether there exists a conflict-free chromatic vertex-guard set for a polygon with holes is NP-hard when the number of colors is k=2.

With the support of emerging technologies such as 5G, machine learning, edge computing and Industry 4.0, the Internet of Things (IoT) continues to evolve and promote the construction of future networks. Existing work on IoT mainly focuses on its practical applications, but there is little research on modeling the interactions among components in IoT systems and verifying the correctness of the network deployment. Therefore, the Calculus of the Internet of Things (CaIT) has previously been proposed to formally model and reason about IoT systems. In this paper, the CaIT calculus is extended by introducing broadcast communications. For modeling convenience, we provide explicit operations to model node mobility as well as the interactions between sensors (or actuators) with the environment. To support the use of UPPAAL to verify the temporal properties of IoT networks described by the CaIT calculus, we establish a relationship between timed automata and the CaIT calculus. Using UPPAAL, we verify six temporal properties of a simple “smart home” example, including Boiler On Manually, Boiler Off Automatically, Boiler On Automatically, Lights On, Lights Mutually, and Windows Simultaneously. The verification results show that the “smart home” can work properly.

Wireless technology improvements have been continually increasing, resulting in greater needs for system design and implementation to accommodate all newly emerging standards. As a result, developing a system that ensures compatibility with numerous wireless systems has sparked interest. As a result of their flexibility and scalability over alternative wireless design options, software-defined radios (SDRs) are highly motivated for wireless device modelling. This research paper delves into the difficulties of designing a reconfigurable multi modulation baseband modulator for SDR systems that can handle a variety of wireless protocols. This research paper has proposed an area-efficient Reconfigurable Baseband Modulator (RBM) model to accomplish multi modulation scheme and resolve the adaptability and flexibility issues with the wide range of wireless standards. This also presents the feasibility of using a multi modulation baseband modulator to maximize adaptability with the least possible computational complexity overhead in the SDR system for next-generation wireless communication systems and provides parameterization. Finally, the re-configurability is evaluated concerning the appropriate symbols generations and analyzed its performance metrics through hardware synthesize results.

Class imbalance is one of the challenges faced in the field of machine learning. It is difficult for traditional classifiers to predict the minority class data. If the imbalanced data is not processed, the effect of the classifier will be greatly reduced. Aiming at the problem that the traditional classifier tends to the majority class data and ignores the minority class data, imbalanced data over-sampling method based on iterative self-organizing data analysis technique algorithm(ISODATA) clustering is proposed. The minority class is divided into different sub-clusters by ISODATA, and each sub-cluster is over-sampled according to the sampling ratio, so that the sampled minority class data also conforms to the imbalance of the original minority class data. The new imbalanced data composed of new minority class data and majority class data is classified by SVM and Random Forest classifier. Experiments on 12 datasets from the KEEL datasets show that the method has better G-means and F-value, improving the classification accuracy.

In machine learning, data augmentation (DA) is a technique for improving the generalization performance of models. In this paper, we mainly consider gradient descent of linear regression under DA using noisy copies of datasets, in which noise is injected into inputs. We analyze the situation where noisy copies are newly generated and injected into inputs at each epoch, i.e., the case of using on-line noisy copies. Therefore, this article can also be viewed as an analysis on a method using noise injection into a training process by DA. We considered the training process under three training situations which are the full-batch training under the sum of squared errors, and full-batch and mini-batch training under the mean squared error. We showed that, in all cases, training for DA with on-line copies is approximately equivalent to the ℓ₂ regularization training for which variance of injected noise is important, whereas the number of copies is not. Moreover, we showed that DA with on-line copies apparently leads to an increase of learning rate in full-batch condition under the sum of squared errors and the mini-batch condition under the mean squared error. The apparent increase in learning rate and regularization effect can be attributed to the original input and additive noise in noisy copies, respectively. These results are confirmed in a numerical experiment in which we found that our result can be applied to usual off-line DA in an under-parameterization scenario and can not in an over-parametrization scenario. Moreover, we experimentally investigated the training process of neural networks under DA with off-line noisy copies and found that our analysis on linear regression can be qualitatively applied to neural networks.

Aiming at the problem of low classification accuracy of surface defects of lithium battery pole pieces by traditional classification methods, an image classification algorithm for surface defects of lithium battery pole piece based on deep learning is proposed in this paper. Firstly, Wavelet Threshold and Histogram Equalization are used to preprocess the detect image to weaken influence of noise in non-defect regions and enhance defect features. Secondly, a VGG-InceptionV2 network with better performance is proposed by adding InceptionV2 structure to the improved VGG network structure. Then the original data set is expanded by rotating, flipping and contrast adjustment, and the optimal value of the model hyperparameters is determined by experiments. Finally, the model in this paper is compared with VGG16 and GoogLeNet to realize the recognition of defect types. The results show that the accuracy rate of the model in this paper for the surface pole piece defects of lithium batteries is 98.75%, and the model parameters is only 1.7M, which has certain significance for the classification of lithium battery surface pole piece defects in industry.

In recent years, there has been a growing demand for urban digitization using cameras and light detection and ranging (LiDAR). Shadows are a condition that affects measurement the most. Therefore, shadow detection technology is essential. In this study, we propose shadow detection utilizing the LiDAR intensity that depends on the surface properties of objects but not on irradiation from other light sources. Unlike conventional LiDAR-intensity-aided shadow detection methods, our method embeds the un-correlation between luminance and LiDAR intensity in each position into the optimization. The energy, which is defined by the un-correlation between luminance and LiDAR intensity in each position, is minimized by graph-cut segmentation to detect shadows. In evaluations on KITTI and Waymo datasets, our shadow-detection method outperformed the previous methods in terms of multiple evaluation indices.

The thermal imaging pedestrian segmentation system has excellent performance in different illumination conditions, but it has some drawbacks(e.g., weak pedestrian texture information, blurred object boundaries). Meanwhile, high-performance large models have higher latency on edge devices with limited computing performance. To solve the above problems, in this paper, we propose a real-time thermal infrared pedestrian segmentation method. The feature extraction layers of our method consist of two paths. Firstly, we utilize the lossless spatial downsampling to obtain boundary texture details on the spatial path. On the context path, we use atrous convolutions to improve the receptive field and obtain more contextual semantic information. Then, the parameter-free attention mechanism is introduced at the end of the two paths for effective feature selection, respectively. The Feature Fusion Module (FFM) is added to fuse the semantic information of the two paths after selection. Finally, we accelerate method inference through multi-threading techniques on the edge computing device. Besides, we create a high-quality infrared pedestrian segmentation dataset to facilitate research. The comparative experiments on the self-built dataset and two public datasets with other methods show that our method also has certain effectiveness. Our code is available at https://github.com/mcjcs001/LEIPNet.

Visual saliency prediction has improved dramatically since the advent of convolutional neural networks (CNN). Although CNN achieves excellent performance, it still cannot learn global and long-range contextual information well and lacks interpretability due to the locality of convolution operations. We proposed a saliency prediction model based on multi-prior enhancement and cross-modal attention collaboration (ME-CAS). Concretely, we designed a transformer-based Siamese network architecture as the backbone for feature extraction. One of the transformer branches captures the context information of the image under the self-attention mechanism to obtain a global saliency map. At the same time, we build a prior learning module to learn the human visual center bias prior, contrast prior, and frequency prior. The multi-prior input to another Siamese branch to learn the detailed features of the underlying visual features and obtain the saliency map of local information. Finally, we use an attention calibration module to guide the cross-modal collaborative learning of global and local information and generate the final saliency map. Extensive experimental results demonstrate that our proposed ME-CAS achieves superior results on public benchmarks and competitors of saliency prediction models. Moreover, the multi-prior learning modules enhance images express salient details, and model interpretability.

Question answering (QA) systems are designed to answer questions based on given information or with the help of external information. Recent advances in QA systems are overwhelmingly contributed by deep learning techniques, which have been employed in a wide range of fields such as finance, sports and biomedicine. For generative QA in open-domain QA, although deep learning can leverage massive data to learn meaningful feature representations and generate free text as answers, there are still problems to limit the length and content of answers. To alleviate this problem, we focus on the variant YNQA of generative QA and propose a model CasATT (cascade prompt learning framework with the sentence-level attention mechanism). In the CasATT, we excavate text semantic information from document level to sentence level and mine evidence accurately from large-scale documents by retrieval and ranking, and answer questions with ranked candidates by discriminative question answering. Our experiments on several datasets demonstrate the superior performance of the CasATT over state-of-the-art baselines, whose accuracy score can achieve 93.1% on IR&QA Competition dataset and 90.5% on BoolQ dataset.

This paper addresses the novel task of detecting chorus sections in English and Japanese lyrics text. Although chorus-section detection using audio signals has been studied, whether chorus sections can be detected from text-only lyrics is an open issue. Another open issue is whether patterns of repeating lyric lines such as those appearing in chorus sections depend on language. To investigate these issues, we propose a neural-network-based model for sequence labeling. It can learn phrase repetition and linguistic features to detect chorus sections in lyrics text. It is, however, difficult to train this model since there was no dataset of lyrics with chorus-section annotations as there was no prior work on this task. We therefore generate a large amount of training data with such annotations by leveraging pairs of musical audio signals and their corresponding manually time-aligned lyrics; we first automatically detect chorus sections from the audio signals and then use their temporal positions to transfer them to the line-level chorus-section annotations for the lyrics. Experimental results show that the proposed model with the generated data contributes to detecting the chorus sections, that the model trained on Japanese lyrics can detect chorus sections surprisingly well in English lyrics, and that patterns of repeating lyric lines are language-independent.

We present a reconfigurable deep learning pedestrian detection system for surveillance systems that detect people with shadows in different lighting and heavily occluded conditions. This work proposes a region-based CNN, combined with CMOS and thermal cameras to obtain human features even under poor lighting conditions. The main advantage of a reconfigurable system with respect to processor-based systems is its high performance and parallelism when processing large amount of data such as video frames. We discuss the details of hardware implementation in the proposed real-time pedestrian detection algorithm on a Zynq FPGA. Simulation results show that the proposed integrated approach of R-CNN architecture with cameras provides better performance in terms of accuracy, precision, and F1-score. The performance of Zynq FPGA was compared to other works, which showed that the proposed architecture is a good trade-off in terms of quality, accuracy, speed, and resource utilization.

Recently, many AI-aided layout design systems are developed to reduce tedious manual intervention based on deep learning. However, most methods focus on a specific generation task. This paper explores a challenging problem to obtain multiple layout design generation (LDG), which generates floor plan or urban plan from a boundary input under a unified framework. One of the main challenges of multiple LDG is to obtain reasonable topological structures of layout generation with irregular boundaries and layout elements for different types of design. This paper formulates the multiple LDG task as an image-to-image translation problem, and proposes a conditional generative adversarial network (GAN), called LDGAN, with adaptive modules. The framework of LDGAN is based on a generator-discriminator architecture, where the generator is integrated with conditional convolution constrained by the boundary input and the attention module with channel and spatial features. Qualitative and quantitative experiments were conducted on the SCUT-AutoALP and RPLAN datasets, and the comparison with the state-of-the-art methods illustrate the effectiveness and superiority of the proposed LDGAN.