Many license identification tools have been proposed to support OSS reuse. License identification tools automatically identify OSS licenses declared in source files. Ninka is one of the most accurate license identification tools. Because OSS licenses are often newly created or inherited, rules built into license identification tools need to be created and updated on a regular basis. However, when a large number of unknown licenses are detected in large OSS products, it is not easy to manually create new rules. In our previous studies, we proposed a method for clustering license statements that Ninka determined to be unknown. In this paper, we propose a method to automatically generate license rules from the clustered license statements. Our approach further filters the license statements from the created clusters to extract sequential patterns and converts the extracted patterns into regular expressions. We conducted conduct a case study where our method is applied to 1, 821, 3, 561 and 2, 838 unknown license statement files respectively collected from FreeBSD v10.3.0, Linux Kernel v4.4.6, and Debian v7.8.0, to confirm the usefulness of our method. As a result, we confirmed that our method successfully generated license rules that take into consideration the orthographical variants and that our method also efficiently identified licenses with a small number of license rules. Furthermore, we found that adding the license rules generated by our method to Ninka improves the licensing rule performance.

Various types of indoor positioning methods have been proposed. Accordingly, various ground-truth measurement methods have also been used. The selection of the ground-truth measurement method is crucial when appropriately evaluating the indoor positioning methods. In recent years, data-driven methods have been actively proposed, making the ground-truth measurement even more critical. Most existing survey papers only focus on classifying indoor positioning methods and datasets for evaluation. Our survey paper investigates the methods to measure the ground truth. First, we list the primary examples of ground truth measurement systems and summarize their characteristics, such as the scale, accuracy, and frequency when collecting ground truth. Then, we classify them into the following five categories: OMC, ToF/MFPS, SLAM, Manual, and Others. Their advantages and disadvantages are also discussed based on four indices: accuracy, coverage, device cost, and labor cost. Finally, we discuss the appropriate way to select the ground-truth measurement system for quantitative evaluation. Our survey will be the guideline for selecting an appropriate ground-truth measurement system for indoor positioning and indoor navigation.

Wadler demonstrated that several laws, called “free theorems”, about polymorphic functions can be derived from their types alone. Free theorems form the foundations of many program transformations, including function fusions particularly. Walder's approach requires systematic but cumbersome reasoning, which makes it difficult for non-experts to apply. To alleviate this problem, Voigtländer proposed a conjuring lemma. This lemma derives free theorems from an expression containing polymorphic functions but whose type does not contain polymorphic type variables. Unfortunately, it cannot derive function fusion laws. In this work, we generalize the conjuring lemma on the basis of Bird and de Moor's relational calculus. This calculus uses non-deterministic computations for uniformly describing and thereby relating specifications and implementations. To make the generalized conjuring lemma easier to use, we develop simpler corollaries by borrowing intuitions from the worker/wrapper transformations. The effectiveness of our approach is demonstrated through experiments with the derivation of three well-known fusion laws. While the original conjuring lemma cannot derive any of them, the generalization can systematically derive all.

In group discussions, each participant has his/her own opinion, and if necessary, tries to persuade other participants to reach an agreement. Therefore, persuasiveness is an important skill for communicating with others. Based on this motivation, this study aims to estimate the persuasiveness of group discussion participants. First, human annotators rated the level of persuasiveness of each participant in group discussions among four people. We then created multimodal and multiparty models for estimating persuasiveness of a participant using speech, language, and visual (head pose) features using gated recurrent unit (GRU)-based neural network. In our experiment, in estimating the highest persuasive participant among the group, the performance of the proposed method was 76% in accuracy. In the binary classification task for estimating high or low persuasiveness participants among the group, the performance of the best performing multimodal multiparty model achieved 80% in accuracy. The experimental results show that multimodal models are better than unimodal models, and multiparty features contribute to improving the model performance. As an application of the proposed method, we implemented a meeting browser with persuasiveness visualization. The level of persuasiveness is visually indicated using the background color of each participant's timeline. Finally, we conducted a user study for our meeting browser, and found that the persuasiveness visualization helps the subjects grasp the flow of the discussion in a shorter time compared to a browser without persuasiveness visualization.

Integrated broadcast-broadband (IBB) systems have been standardized for device collaboration between broadcast and broadband contents as well as between TVs and diverse devices such as Internet of Things or smart devices. However, in existing IBB standards, TVs are typically identified as the primary devices (i.e., servers), and the other devices are identified as the companion devices (i.e., clients), which makes communication between two TVs impossible. To this end, we propose a common receiver architecture as a distributed IBB system that realizes local device cooperation with equal roles (client and server, i.e., “servient”). As an equal-role device collaboration for presenting content, this study focuses on the use case of switching devices while playing the same content. The proposed architecture acts as middleware to realize local device communication between TVs and other devices using web-based communication, such as hypertext transfer protocol. Device-independent cooperation is achieved using our developed prototype for TVs and smartphones, and the prototype is verified via thorough user testing with significantly satisfactory feedback.

In this paper, we propose adaptive message prioritization for vehicular cooperative perception at target intervals. The purpose of this paper is to ensure that Connected and Autonomous Vehicles (CAVs) can perceive the surrounding traffic environment by exchanging wireless messages such as Cooperative Awareness Messages (CAMs) and Cooperative Perception Messages (CPMs). Especially, under challenging conditions such as the wireless congestion control or the mixed-traffic of CAVs and human-driven vehicles, the number of perceptible objects by these messages decreases due to the message transmission rate control under the congestion control or the less message transmission under the low CAV penetration rate. To perceive the surrounding objects under the challenging conditions, the proposed method assigns high transmission priority to CPMs that include the information of multiple objects perceived by in-vehicle sensors. Specifically, the method controls the transmission frequency of CAMs and CPMs by assigning transmission priority to them such that the maximum perception interval approaches the target interval. Simulation result shows that 75% CAVs perceive surrounding vehicles within the target interval even under both the wireless congestion control and 20% CAV penetration rate.

Rate Monotonic (RM) is one of the widely used real-time scheduling algorithms. In RM, tasks with shorter periods are given higher priorities and tend to have short response times and jitters. In contrast, those with longer periods are likely to suffer from increased response times and jitters due to their lower priorities. We proposed the Execution Right Delegation (ERD) method for uniprocessor systems based on RM where a high-priority server for a privileged (or important) task is introduced to shorten response times of the task. In our previous paper, we proposed an extended ERD method for two-core multiprocessor systems in which the system model is based on partitioned systems while only a privileged task can migrate. In this paper, we relax the restriction of the number of cores. In addition, the number of privileged tasks that was previously limited to one is expanded to two. In the evaluation, it is confirmed that response times of privileged tasks are reduced compared with partitioned Fixed-Task-Priority (FTP) and global FTP scheduling.

The use of Linux in the domain of embedded systems is growing very fast. Due to the complexity of hardware and software on Linux-based platforms, it is challenging to meet the performance requirements, especially for the mixed-criticality system with both real-time and best-effort tasks. We propose a reference design of Linux-based mixed-criticality system using PREEMPT_RT patch to improve real-time performance and QEMU/KVM virtual machine to reduce interferences. An evaluation environment of the proposed design is built with recent software and hardware to investigate the performance characteristics via experimental measurements. We measure and analyze the baseline kernel latency, the system throughput and the real-time performance under multiple conditions. In detailed analysis, we reveal novel insights on the real-time capabilities and limitations of Linux-based embedded systems. The results show that our design can meet the 100µs deadline goal of 1kHz real-time task with high probability under various extreme interferences, and can deliver high throughput for best-effort workload.

Device authentication is essential between network-connected Internet of Things (IoT) devices and the systems connected to them to prevent security threats, such as leakage and tampering of information. However, there are cases where IoT devices with limited manufacturing costs do not have secure key storages. As cryptographic communication and authentication using symmetric keys may suffer from various attack opportunities, such devices should be secured against the white-box adversary, which can fully access the executing environment and the memory.

In this study, we propose two secure encryption schemes in the white-box model. The proposed scheme utilizes a space-hard cipher and physically unclonable function (PUF) in a hybrid manner to prevent key extraction and code lifting attacks in the white-box model. In this scheme, the security is maintained at the same level similar to that of the Even-Mansour cipher by periodically changing the whitening key without updating the key of the space-hard cipher, even after the space-hard cipher is stolen by code lifting attacks. Therefore, this scheme can achieve a significant level of security against key extraction and code lifting by low-cost updating only the whitening keys.

The creation and spreading of new pathogens by pathogen-pathogen interaction were modeled and computer-simulated using the Susceptible Infected Susceptible (SIS) model on three-layer scale-free networks: the Barabási-Albert (BA) networks, the Dorogovtsev-Mendes-Samukhin (DMS) networks and the shuffled BA networks. The new pathogen is a model of a new virus, critical privacy information (old pathogens are uncritical data) etc. When the emergence of a new pathogen was low, bimodal metastable states were observed. They had high states and low states of the prevalence. The high states were metastable states with positive prevalence. The low states have the prevalence fluctuating near 0. On both shuffled and non-shuffled networks, the epidemic prevalence in the metastable state on the BA networks are well explained by the the heterogeneous mean-field (HMF) formulation of the SIS+g model. The model does not match well on the DMS networks; this may be because their connectivity distributions are convex or concave. By the three-layer simulation, the infected nodes in layer-0 and layer-1 were found to be not always well-mixed i.e. they are localized. The localization occurred when the epidemic prevalence was low. The nodes with medium connectivity contributed to it. The meaning of results for privacy is discussed.

Graphs are a generalized concept that encompasses more complex data structures than trees, such as difference lists, doubly-linked lists, skip lists, and leaf-linked trees. Normally, these structures are handled with destructive assignments to heaps, which is opposed to a purely functional programming style and makes verification difficult. We propose a new purely functional language, λ_GT, that handles graphs as immutable, first-class data structures with a pattern matching mechanism based on Graph Transformation and developed a new type system, F_GT, for the language. Our approach is in contrast with the analysis of pointer manipulation programs using separation logic, shape analysis, etc. in that (i) we do not consider destructive operations but pattern matchings over graphs provided by the new higher-level language that abstract pointers and heaps away and that (ii) we pursue what properties can be established automatically using a rather simple typing framework.

Identification of services constituting traffic from given IP network flows is important for many purposes such as management of quality of service, prevention of security problems, and providing a discounting service for customers only in accessing specified services like zero-rating service. The simplest methods for identifying these services are identifications based on IP addresses and port numbers. However, such methods are not sufficiently accurate and thus require improvement. Deep packet inspection (DPI) is an advanced method for improving the accuracy of identification. Many current IP flows are encrypted with the transport layer security (TLS) protocol. Therefore, an identification method cannot analyze almost all the data encrypted by TLS. In the cases of TLS 1.2 or less, some fields, e.g. server name indication (SNI), in the protocol header for the TLS session establishment are not encrypted and then can be analyzed. Thus, we can expect that the service can be identified from IP flows, which are composed of TLS sessions, by analyzing these fields. For achieving this, two challenges are mainly required. One is grouping TLS sessions by accesses from many TLS sessions that pass through a network element. The other is the identification of service from TLS sessions grouped in the first challenge. In our work, we mainly focus on the second theme, i.e., service identification from given TLS sessions. In our previous work, we proposed a method for identification by analyzing these non-encrypted data based on DPI and n-gram. However, there is room for improvement in identification accuracy because this method analyzed all the non-encrypted data including random values without protocol analysis. In this paper, we propose a new method for identifying the service from given TLS sessions based on SNI with protocol data unit (PDU) analysis. The proposed method clusters TLS sessions according to the value of SNI and identifies services from the occurrences of all groups. We evaluated the proposed method by identifying services on Google, Yahoo, and MSN sites, and the results showed that the proposed method could identify services more accurately than the existing method. The average ratios of inaccurate identifications were decreased by 65%, 72%, and 41% in our experiments of Google, Yahoo, and MSN services, respectively.

The functional testing specification is usually designed based on the IO entities' recognition from the embedded system's functional requirement sentences. However, it is hard for the software testing engineers to ably recognize the appropriate IO entities from the functional requirement sentences without clearly indicated entities and much experience with the domain knowledge. The conventional rule-based methods of extracting IO entities are inapplicable when the requirement sentences drafted by humans become too semantically complex. Even though all the sentences keep aligned with the structure, it is still infeasible to manually hard-code each rule when those rules change from time to time without any explicit writing standard. With the successful application of artificial intelligent techniques in natural language processing (NLP), we propose a method that intelligently solves the issue of the entities recognition by using BERT (Bidirectional Encoder Representations from Transformers) based named entity recognition (NER) which is the technique of NLP to recognize the phrases having similar attributes in semantics. In this paper, we specifically focus on the issue of IO entities' recognition in the embedded systems that implement the inverter control function such as elevator and hybrid hydraulic excavator systems. Our evaluation result demonstrates that the best model variant fine-tuned on 829 sentences achieves more than 80% F-measure in recognizing the IO entities, and the model can provide applicable information for the improvement of industrial productivity in the target industries. Our contribution of this paper is to provide insight into the case whether the IO entities in the target system manages to be interpreted well by exploiting the BERT model with the sole reliance on the small size of exemplary IO entities data and three existing model variants pre-trained on large corpus open datasets with general language knowledge.

Internet topology data is useful for evaluating new protocols, services, and vulnerabilities in network infrastructures. To infer the AS-level topology, some projects deploy BGP routers worldwide, mainly on IXPs, collect and publish BGP routing information from other ASes called Vantage Points. Deploying more routers on various IXPs will make the topology data precise, but it is hard to deploy a lot of routers because of administrative costs. So, we carefully choose locations where we can obtain new topological information with a small number of new routers. In this study, we investigate which ASes and IXPs should be used as Vantage Points with a small number of additional routers. This work consists of two stages. First, we analyze current BGP routing information available to infer currently unobserved but possibly existing connections, and find that such connections tend to be between two ASes that have a high percentage not including 100% of common adjacent ASes. Second, we calculate which IXPs should be more suitable for installing BGP routers for the collection using the results on the first stage and data of ASes on IXPs. We also discuss IXPs recommended to deploy new routers are in different countries where many ASes are allocated.

In recent years, malware-infected devices, such as Mirai, have been used to conduct impactful attacks like massive DDoS attacks. Internet Service Providers (ISPs) respond by sending security notifications to infected users, instructing them to remove the malware; however, there are no approaches to quantify or simulate the performance and effectiveness of the notification activities. In this paper, we propose a model of security notification by ISPs. In the proposed model, we simulate the security notification with composite parameters, indicating the nature of malware attacks such as persistence of malware, user response ratio, and notification efforts by ISPs, and then discuss their effectiveness. Moreover, we conduct a simulation based on the actual attack.

Many smart home devices, such as smart speakers, are connected to a home network. Conventional home networks do not control communications among devices, including malicious devices. As a result, malicious devices in a home network can steal private information and manipulate devices that have no or weak access control mechanisms. Some methods have been proposed to protect devices, but cannot control communications in some situations and do not discuss their authorization architecture or usability. Malicious devices can bypass access control, and users may make incorrect configurations due to their complexity. They can make access control meaningless. In this paper, we propose CREBAS, which enables network access control to protect devices from malicious or unintended communications. We propose a capability-based authorization architecture and discuss its validity according to the guidelines for access control in a home. Also, we propose methods that provide more fine-grained and stricter network access control according to authorized capabilities. We have implemented a prototype of CREBAS. The prototype has shown that the performance degradation is acceptable, and it is possible to perform access control with one click in a real protocol used in a home network.

Modern systems suffer system faults in both hardware and software. Requirement degradation is a widely applied redundant-design approach to deal with system fault by degrading or disabling (non-critical) requirements. However, due to the environment's unpredictability, it is unrealistic for engineers to prepare such redundant designs for unforeseen and unknown faults in advance. Requirement-aware self-adaptive systems have been investigated to address this challenge. Specifically, such a system can autonomously analyze which requirements are guaranteeable so that the requirement degradation can be determined automatically to satisfy the maximal requirements. The previous work has proposed an algorithm to recognize the guaranteeable requirements by analyzing a two-player game between the environment and the system. However, such an algorithm is not designed for the system fault, and the entire game must be re-analyzed from scratch, making it inefficient for runtime adaptation. In this paper, we propose an efficient difference analysis algorithm that reuses the previous analysis result to reduce the runtime computational overhead. Specifically, it carefully specifies which part of the result is reusable and only re-analyzes the part that may be affected. We evaluated the algorithm's effectiveness with four case studies, and the results indicate that an average of 46.5% computational time is reduced.

Capital expenditure (CAPEX) and operating expenses (OPEX) a key practical concerns for telecommunication companies, emphasizing the attraction of Network Function Virtualization (NFV) in recent years. However, the NFV Infrastructure (NFVI) includes a wide variety of resource types and stakeholders. While many attempts have been made to enhance the ability of existing resource scheduling approaches to support the complex and conflicting NFV ecosystem, they are no longer sufficient. In this paper, I propose the Multiple Objective Resource Scheduling Algorithm (MORSA) to optimize the NFVI resource usage. In developing MORSA, I designed a plug-in architecture to satisfy various requirements related to NFVI resources and stakeholder policies. MORSA allows the NFVI Resource Scheduler to optimize, simultaneously, any combination of possibly conflicting objectives with multifaceted constraints in complex real world situations. Evaluation results show that MORSA obtains good approximate solutions by resolving conflicting objectives in reasonable computation times.

Lightweight cryptography algorithms are increasing in value because they can enhance security under limited resources. National Institute of Standards and Technology is working on standardising lightweight authenticated encryption with associated data. Thirty-two candidates are included in the second round of the NIST selection process, and their specifications differ with respect to various points. Therefore, for each algorithm, the differences in specifications are expected to affect the algorithm's performance. This study aims to facilitate the selection and design of those algorithms according to the usage scenarios. For this purpose, we investigate and compare the 32 lightweight cryptography algorithm candidates using specifications and software implementations. The results indicate that latency and memory usage depend on parameters and nonlinear operations. In terms of memory usage, a difference exists in ROM usage, but not in the RAM usage from our experiments using ARM platform. We also discovered that the data size to be processed efficiently differs according to the padding scheme, mode of operation, and block size.

Synthesizers have become commonplace in music production due to their ability to create a diverse and unique timbre. However, many users find it difficult to control a synthesizer due to the complexity of the relationship between synthesis parameters and output sound. As such, the sound matching task focuses on the estimation of synthesis parameters that recreate the target sound as closely as possible within the capabilities of the synthesizer. In this paper, we introduce a quality-diversity approach to the problem of sound matching by introducing a novelty objective to genetic algorithm-based sound matching. We show that this QD method allows for the discovery of a collection of matches that are more diverse compared to matches found by standard genetic algorithms. These matches are spread across a space of interpretable spectral features, making it useful for the intuitive discovery of unique and useful matches.

The COVID-19 pandemic raised an unprecedented educational challenge that suddenly dropped the entire Higher Education world in a genuine paradigm shift. The 18 months spent since the beginning of this still ongoing crisis represent valuable hindsight for a first analysis and discussion of the moves and the trends the universities were involved in regarding the Teaching and Learning practices. They also represent an opportunity to highlight the dynamics that are already shaping their short and middle/long-term perspectives, especially related to ICT, through the angles of the faculty and stakeholders. This paper proposes a qualitative French/Japanese insight into these moves, trends, and dynamics, as the result of the first step of an ongoing international comparative study.

Cross-modal information retrieval based on subjective information aims to enable flexible media retrieval services, such as allowing users to specify, for example, an image to search audio clips. The resulting audio clips should have an impression similar to the specified image. Existing methods focus on building cross-media cross-modal relationships using objective information (such as the standard caption). However, such a relation can be built only between the pieces of media that are originally related, which limits the flexibility of cross-modal media retrieval. This research leverages subjective information in the media clips for similarity calculation to achieve more flexibility. We propose a novel cross-modal stochastic neighbor embedding technique called c-SNE. c-SNE can extract features of subjective information from pieces of media and map them in the common embedding space. It is a learning technique to bridge the heterogeneous gap between the modal distributions using label-weighted SNE. It allows users to find the media that share the same subjective information with a query medium. Our experimental results on the benchmark datasets demonstrate that the proposed method effectively performs in cross-modal distribution alignment and retrieval. Furthermore, our user study with ten users with 600 data points confirmed that c-SNE outperforms three related methods in the actual usage situation from the users' perspective.

Real-time flood prediction in urban areas is an important tool for city emergency planning. Earlier studies suggest two approaches to predict flooding: a supervised machine learning approach based on observed data and a modeling approach for urban environments based on hydraulics. However, the first approach can only be applied in areas where there is sufficient data on flooding and is not accurate enough for prediction. The second approach can provide accurate predictions even for cities that have never experienced flooding, but models of complex urban environments are not suitable for real-time prediction owing to significant computational complexity. Therefore, we propose a third approach, machine learning supervised by a program. This approach consists of training a lightweight neural network using an integrated flood analysis program composed of multiple hydrologically based models. We demonstrate that this trained neural network is 19.8 times faster and and has accuracy comparable to that of the previous modeling approaches.

With the rapid spread of social networking services (SNS) and social media, tourism has become a more accessible leisure activity. While many tourists visit well-known tourist destinations, the number of tourists visiting lesser-known ones in non-urban or rural areas has been in decline, and the decline has been exacerbating various problems such as economic disparities. Under these circumstances, it is becoming increasingly important to attract tourists to lesser-known areas with few tourism resources, because attracting tourists brings various positive effects to local residents and municipalities. Existing efforts to attract tourists include studies on tourist spot mining, which search a tourist destination for sightseeing spots that are worth visiting but relatively less well known, based on user-generated content about spots on the Web, such as posts on SNS and images on photo-sharing services. However, as they rely on the amount of user-generated content about potential sightseeing spots to be mined, these methods are limited in discovering a new sightseeing spot that is not recognized as a tourist attraction in advance. To address these challenges, we are inspired by an interesting phenomenon: a previously little-known spot that looks similar to a famous place becomes a hot topic on social media, attracting many people and suddenly becoming a popular tourist attraction. In this paper, we call such a spot with a similar appearance to a well-known one, as generic point-of-interest (POI), and we provide a novel concept of discovering new sightseeing spots by computationally finding generic POIs; we give a new perspective of discovering a new tourist attraction by characterizing a commonplace spot that has a similar appearance to a well-known one as a new tourist attraction. We also propose a pipeline implemented for mining generic POIs, where we calculate the similarity between candidate spots of generic POIs and famous sightseeing spots. Furthermore, we evaluate the feasibility of our approach both through human evaluation and through a photo contest on the Web we held with Tourism Shikoku. The results prove that our proposed method can discover new tourist destinations. The primal advantage of our approach is that it gifts new tourism value to a spot with at least a single place photo similar to a famous sightseeing spot, and thus it discovers a new tourist attraction even in areas with a minimal number of user-submitted images. We hope that our novel concept will be the first step in efforts to attract tourists in various regions around the world, enabling utilization of local spots that even the locals may not be aware of.

Cyber Threat Information (so-called indicator) sharing plays an important role in rapid incident response, and ISAC and MISP provide information sharing schemes. Companies and organizations receive indicators and check whether their employees' computers and servers are exposed to threats by referring to their proxy and access logs. We have already proposed an incident response scheme that automates this sequence of procedures. Meanwhile, the amount of logs generated per day in large companies and organizations is enormous. In addition, the number of indicators received is expected to increase. In that situation, a naive matching algorithm between indicators and logs can take a considerable amount of time and delay incident response seriously, even when using our proposed system. Therefore, we propose a more effective matching algorithm. We also implemented the algorithms and measured their elapsed time and memory consumption. In addition, we compare our algorithms with basic and naive matching algorithms from both a theoretical and practical perspective. As a result, our proposed algorithm ran faster with a little additional memory consumption. Combining this algorithm and our previously proposed incident response system, we achieve a more effective one than before.

Fully homomorphic encryption (FHE) is a promising tool for privacy-preserving applications, and it enables us to perform homomorphic addition and multiplication on FHE ciphertexts without decrypting them. FHE has two types: one supporting the exact computation and the other supporting the approximate computation. Further the FHE schemes supporting the exact computation have two types, bit-wise FHE, which encrypts a plaintext bit by bit, and integer-wise FHE, which encrypts a plaintext as an integer. Both types of FHE are important depending on the types of computation we need to execute securely. In this work, we focus on integer-wise FHE, and propose improved methods for integer-wise homomorphic comparison and division operations. For a comparison operation, we improve on the work of Iliashenko and Zucca (PoPETs'21) whose complexity is O(p) homomorphic multiplications, and achieve the complexity $O(\!\!\sqrt{p})$ where p is a plaintext modulus. For a division operation, as opposed to the work of Okada et al. (WISTP'18), we propose a simple method to reduce the processing time by introducing an equality function based on Fermat's little theorem without changing the multiplicative depth, and show the analysis of why this approach can achieve better efficiency in detail. In our homomorphic division, the number of interpolated polynomials is reduced by half, thus also achieving the reduction of the processing time of precomputations and the number of polynomials to be stored. We also implement our improved methods in HElib, which is one of popular FHE libraries using the BGV encryption. As a result, we show that, e.g., in the plaintext space ℤ₂₅₇, our homomorphic comparison with the Paterson-Stockmeyer method is faster by a factor of about 14.5 compared with Iliashenko and Zucca (PoPETs'21) and our homomorphic division is faster by a factor of about 1.45 compared with Okada et al. (WISTP'18).

Video-guided machine translation, as one type of multimodal machine translation, aims to engage video contents as auxiliary information to address the word sense ambiguity problem in machine translation. Previous studies only use features from pre-trained action detection models as motion representations of the video to solve the verb sense ambiguity and neglect the noun sense ambiguity problem. To address this, we propose a video-guided machine translation system using both spatial and motion representations. For the spatial part, we propose a hierarchical attention network to model the spatial information from object-level to video-level. We investigate and discuss spatial features extracted from objects with pre-trained convolutional neural network models and spatial concept features extracted from object labels and attributes with pre-trained language models. We further investigate spatial feature filtering by referring to corresponding source sentences. Experiments on the VATEX dataset show that our system achieves a 35.86 BLEU-4 score, which is 0.51 score higher than the single model of the SOTA method. Experiments on the How2 dataset further verify the generalization ability of our proposed system.

An ideal self-adjusting environment requires adapting human thermal comfort automatically and continuously measuring the changes in human thermal comfort. According to the PMV (Predicted Mean Vote) model, human thermal comfort could be evaluated by human biometric data and environmental data. In this paper, we proposed a method to estimate human thermal comfort through a small number of wearable and environmental sensors based on machine learning. There are two typical definitions of thermal comfort: subjective thermal comfort representing the subjective perception of heat and objective thermal comfort calculated by the PMV formula. We used a subjective questionnaire and PMV formula to obtain the correct label for two kinds of thermal comfort, respectively. Among the three machine learning models, the random forest has 0.73 in MAE, which is suitable for estimating 7-level subjective thermal comfort, and the neural network has 0.47 in MAE, which is suitable for estimating objective thermal comfort. We investigated the estimation accuracy by changing the sensors' combinations. As a result, a small number of sensors could still roughly estimate human thermal comfort.

Learning a classification model from a dataset with the presence of irrelevant variables makes such a task more difficult and less accurate. To address this issue, feature selection has been proposed. However, existing feature selection struggles to select accurate and relevant sub-features or subspaces. It has been demonstrated in the literature that learning the Markov boundary (MB) of features helps to extract appropriate subsets of relevant features from the full data space. However, due to the fact that almost all existing methods learn MBs with static data, a state-of-the-art algorithm for learning MBs with streaming data, called SDMB (streaming data-based MB), has recently been proposed. SDMB has successfully improved online feature selection based on learning Markov boundary (MB) from data stream for classification tasks. Nevertheless, we found that SDMB has computational redundancy and there is room for improvement in terms of runtime. Therefore, we propose an improvement to SDMB. The main drawback of SDMB is that it entirely recomputes the conditional independence/dependence tests (i.e., the dependency scores) between a target variable and other variables in each iteration of learning MBs from the full data space. This paper proposes an incremental SDMB, called FastSDMB, to address the above bottleneck by reducing the dependency score computation time of SDMB. FastSDMB avoids the computational redundancy in SDMB by using a novel incremental computing collective dependency scoring process. This incremental process retains and reuses, in current iterations, the calculated conditional independence test score of variables that have not changed from previous iterations. To demonstrate FastSDMB's effectiveness and capability of handling fast data arrival scenarios, comparisons are made first over the state-of-the-art SDMB, and then we apply FastSDMB to various classification models in order to showcase the resulting performance improvement when implemented over a wide variety of synthetic and real-world datasets. The experimental results show that FastSDMB can run up to 4.8 × faster on average than SDMB and it maintains the same accuracy when learning Markov boundary. Thereby proving its efficiency in fast relevant feature selection based on Markov boundary learning from data stream.

Logical Natural Language Generation, i.e., generating textual descriptions that can be logically entailed by a table, has been a challenge due to the low fidelity of the generation. Previous works have addressed this problem by annotating interim logical programs to control the generation contents and semantics, and presented the task of table-aware logical form to text (Logic2text) generation. However, although table instances are abundant in the real world, logical forms paired with textual descriptions require costly human labor, which limits the parallel data size. To mitigate this, we propose topic-conditioned data augmentation (TopicDA), which employs controlled sequence-to-sequence generation as auxiliary tasks to augment logical forms and textual descriptions directly from tables. We further introduce logical form generation (LG), a dual task of Logic2text that requires generating a valid logical form based on a text description of a table. We hence propose a semi-supervised learning approach to jointly train a Logic2text and an LG model with both labeled and augmented data. Experimental results on both Logic2text and LG demonstrate that our approach can effectively utilize the augmented data and outperform supervised baselines by a substantial margin.

Non-autoregressive (NAR) models can generate sentences with less computation than autoregressive models but sacrifice generation quality. Previous studies addressed this issue through iterative decoding. This study proposes using nearest neighbors as the initial state of an NAR decoder and editing them iteratively. We present a novel training strategy to learn the edit operations on neighbors to improve NAR text generation. Experimental results show that the proposed method (NeighborEdit) achieves higher translation quality (1.69 points higher than the vanilla Transformer) with fewer decoding iterations (one-eighteenth fewer iterations) on the JRC-Acquis En-De dataset, the common benchmark dataset for machine translation using nearest neighbors. We also confirm the effectiveness of the proposed method on a data-to-text task (WikiBio). In addition, the proposed method outperforms an NAR baseline on the WMT'14 En-De dataset. We also report analysis on neighbor examples used in the proposed method.

Human-in-the-loop topic models allow users to encode feedback to modify topic models without changing the core machinery of the topic models. Basic refinement functions have been proposed in prior works in which the main focus was to modify the top word lists of topics, e.g., add a single word in a topic having distribution over a large vocabulary set. In this work, we point out that such refinements have very little to no effect on document-topic associations, which are rather important in practical applications, and propose keyphrase-based refinement functions that are designed to improve document-topic associations efficiently. In the proposed method, these keyphrases are extracted by using a neural keyphrase generation model that summarizes a document in a few keyphrases which are human-interpretable representations of each given document. The proposed refinement functions are as simple as word-based refinements but directly modify the document-topic association in functionality by referring to the keyphrase representations of documents. To examine the capability of the refinement functions for revising topic models, we conducted experiments based on a simulated user that has a fixed preference for the document-topic association using 20Newsgroups dataset. Our results showed that the proposed keyphrase-based refinements outperform the basic word-based refinements in terms of the F1 score computed on the fixed preference.

Concurrency control ensures the correctness of databases when transactions are processed in parallel. Bamboo is a state-of-the-art concurrency control protocol extended from 2PL Wound-Wait. One problem of Bamboo is that it requires transactions to fetch timestamps from a single centralized atomic counter. To replace the concentrated access to it, each transaction should generate timestamps independently. This paper proposes three methods of decentralization to address the problem. The first is the thread-ID method (TID), which dismisses the process of fetching timestamps entirely by assigning an ID to each thread, and transactions use the thread IDs as their timestamps. In high-contention settings, the performance of TID plummets, but the second method FairTID, which is an extension of TID, sustains the performance even in such settings. The third method (RandID) allocates timestamps using random-number generators. Experiments were conducted to evaluate the performance of the proposed methods. The results indicated that there were up to 62% and 59% improvement in throughput from Bamboo with FairTID and RandID, respectively. In high-contention settings, TID underperformed Bamboo, but FairTID and RandID showed consistent improvement from Bamboo.

Collaborative learning fosters the ability to creatively solve problems in collaboration with other learners. Researchers in learning science have transcribed learners' speech to qualitatively analyze collaborative learning to reveal various patterns that increase learning performance. Although prior studies have identified speakers to support the process of transcription, those studies were limited in simultaneously identifying multiple speakers. We propose a novel speaker-identification algorithm that can simultaneously recognize multiple speakers using business-card-type sensors. The algorithm can remove ambient noise with low-cost sensors and still identify multiple simultaneous speakers. The experimental evaluations show that the algorithm accurately identifies simultaneous multiple speakers in a multi-person activity under conditions with varying numbers of users, environmental noise, and users' short utterances.

Blockchain is gaining attention as a technology to support cryptocurrency. However, one of the problems in using blockchain is the consumption of communication resources. It is caused by that the information received by a node can be delivered to the same node more than once through different neighbors. In this paper, we propose a method to reduce duplicate messages in a blockchain network by applying Plumtree algorithm. Through simulation experiments, we confirmed that the number of messages can be reduced by more than 70% when block propagation is simulated up to 50 blocks.

Point & teleport (P&T) is an artificial locomotion technique that enables users to travel in unlimited space in virtual reality. While recent P&T techniques assign orientation control to an additional axis, these techniques suffer from increased complexity in controls and limited performance. Researchers concluded that teleportation, followed by a self-orientation adjustment by physically turning around, is preferable and that P&T with orientation specification can be optional. However, P&T has not been tested under a seated condition, where the orientation control may advantageously perform. Therefore, in this paper, we reevaluate P&T with orientation specifications while the users are seated. Nonetheless, for consistent alignment with the results in preceding research, we evaluate the accuracy while the users are standing. Knowing that additional cognitive load may badly affect the performance, we present a new P&T design, points to teleport (P2T), with minimal complexity in mind (i.e., point twice sequentially to determine the future location and then orientation, which mimics classic P&T that requires users to turn around for orientation specification). Thus, we revisit P&T with orientation specification.

With the proliferation of the Internet of Things and smart devices, content presentation devices and content types are constantly increasing. However, there is little consideration given to selecting the appropriate devices for presenting a particular content. Therefore, we propose a web-based device selection method based on semantic data from the resource description framework. In this study, we use notifications as examples of content, and a network graph representing users, content, and devices is arranged in a two-dimensional space by our user-centric force-directed algorithm. By selecting device nodes near the user node from the placement results, devices are determined to present content. Based on the proposed algorithm, a device selection system and a notification presentation application were prototyped. The experimental results show that the proposed method has a higher probability of selecting the correct devices than the existing method of obtaining vectors from RDF data.

This paper investigates the Ricochet Robots game problem from a complexity standpoint. The problem consists of moving robots in a rectangular square-tiled board, from initial tiles to reach specific target tiles. A robot can only move vertically or horizontally and when it starts to move in a given direction, the robot follows this direction, until being blocked by a wall or another robot. This paper proves that the corresponding decision problem to Ricochet Robots is Turing-complete for endless board game and an infinite number of robots. A reduction from a universal Turing machine to Ricochet Robots is exhibited.

Context-free grammars with lookahead (CFGLAs) are context-free grammars with positive and negative lookahead. In this paper, we develop a recognition algorithm for CFGLA by combining derivatives of context-free grammars and derivatives of regular expressions with lookahead. The original semantics of CFGLA has some problems, and the recognition algorithm does not run in polynomial time. We introduce better semantics with the least fixed point of pairs of lower and upper bounds. Under the new semantics, the recognition algorithm runs in O(n³ |G|) time and O(n² |G|) space for an input string of the length n and a given CFGLA G.

In the context of the Curry-Howard correspondence, it is well-known that a typed calculus with control operator(s) corresponds to a formal system of classical logic, and various concrete examples have been discovered and studied since the early 90s. Among such work, Curien and Herbelin proposed the lambda-bar-mu-mu-tilde-calculus to exhibit the so-called duality of computation such as the (program/continuation)-duality and the (call-by-value/call-by-name)-duality. In their calculus, programs and continuations are first-class citizens formalized in a symmetric manner; and the call-by-value and call-by-name evaluation disciplines are characterized in some appropriate subcalculi, which are dual to each other. This work aims to introduce an extension of the lambda-bar-mu-mu-tilde-calculus with necessity and possibility modalities, which logically corresponds to classical S4 modal logic. In this paper, we show that the calculus is defined as a combination of existing calculi, and show that it satisfies desirable properties such as the subject reduction and the strong normalization. Two well-behaved subcalculi that realize the call-by-value and call-by-name disciplines are also discussed, following the work of Curien and Herbelin. As an application, we demonstrate that our calculus can be used to prove the strong normalization of the S4 modal lambda-mu-calculus of Kimura and Kakutani.

Our university offers a basic mathematics course in a “flipped classroom” format. During each class in this course, learners use a learning plan support system that we developed to register learning plans and learning records for out-of-class study. Previously, this system was unable to support the creation of highly feasible learning plans leading to the attainment of learning goals. In this study, we therefore developed a support system of learning plan creation that aims to promote reflection leading to the creation of a learning plan that enables learners to learn as planned and attain their learning goals. In addition, two types of reflection support functions were developed: a chatbot-based reflection function that asks questions emphasizing learning outcomes and a dashboard-based reflection function that asks questions emphasizing the implementation of the plan. In the 2020 and 2021 school years, we performed an experiment using this system in the University's basic mathematics course, and the results suggest that use of the dashboard-based reflection function may significantly improve plan implementation rates and learning completion rates, and may have enabled the creation of highly feasible learning plans. Moreover, the results of analysis of the reflection contents registered using the reflection support functions showed that more than 60% of the reflections were related to improving plan implementation rates and goal attainment rates.