The Thirteenth International Symposium on Computing and Networking (CANDAR 2025) was held in Yamagata, Japan, from November 25 to 28, 2025. Following the symposium, the authors of highly evaluated papers were invited to submit extended versions for consideration in this special issue. After a rigorous peer-review process, nine papers were accepted from 14 submissions and are published here as extended versions. We would like to express our sincere gratitude to the anonymous reviewers, who carefully evaluated the submissions and provided insightful comments and suggestions. Their dedicated efforts and constructive feedback were instrumental in enhancing the quality of the accepted papers and shaping the content of this issue. Finally, we extend our heartfelt thanks to all those who contributed to the success of this special issue, including the organizers, guest editors, and reviewers. Their collective contributions made it possible to present this collection of extended papers, which reflects recent advances in networking and computing. We hope that this issue will serve as a valuable resource for researchers and practitioners and inspire further innovation in the field.

抄録全体を表示

Accurate positioning in mountainous and forested environments remains challenging for tourist assistance applications. This study proposes a hybrid proximity detection method that integrates Global Navigation Satellite System (GNSS) gating with Bluetooth Low Energy (BLE) refinement. The system first detects the user’s approach using GNSS and then estimates the point of closest approach based on BLE RSSI variations or packet reception counts. Experiments were conducted at the Senjogahara Nature Exploration Trail in Nikko, Japan, which includes both forested sections and flat open areas. In Senjogahara, GNSS achieved the highest positioning accuracy under normal conditions; however, its performance degraded under dense forest canopy. Even in such conditions, the hybrid method incorporating BLE-based refinement maintained more stable performance than the GNSS-only approach. Additional experiments were also conducted in a flat urban area, specifically at a university rugby field. In these experiments, the proposed method using a BLE reception-count trigger achieved the smallest positioning error, outperforming both the BLE-only and GNSS-only methods.

抄録全体を表示

In this study, we implement a machine learning–based NIPS on an FPGA to detect and block DDoS attacks, and propose a false-positive reduction method that considers server load. The proposed method periodically measures traffic volume and server response time to estimate the server load. By adopting the larger classification ratio determined from these two metrics, the system performs stepwise blocking control that suppresses unnecessary blocking under low-load conditions while ensuring detection performance under high-load conditions. In the accuracy evaluation, Precision, Recall, F1-score, FPR, and FNR were calculated for each classification ratio, and we quantitatively demonstrated the trade-off whereby Recall and F1-score improve as the classification ratio increases, while FPR also increases. Compared with the previous approach that relies solely on traffic volume as an indicator of server load, the introduction of response time provides a more flexible basis for load-dependent control and suggests the potential to better address low-traffic DDoS attacks under the evaluated conditions. However, in this study, the classification ratio update interval and threshold values are configured as representative examples for system validation. In addition, the response time measurement mechanism is applicable only to TCP traffic. Therefore, the reported results demonstrate the feasibility and effectiveness of the proposed approach within this scope, rather than providing a generalized validation across all attack types and protocols. In the implementation evaluation, a minimum latency of approximately 730 ns was achieved, and it was confirmed that throughput equivalent to approximately 10 Gbps can be processed without packet loss under multiple packetsize conditions. Furthermore, even when the classification ratio update period was changed to 0.5, 1, and 2 s, the latency variation remained below 0.5 %, and 10 Gbps traffic was processed stably, demonstrating the processing stability against changes in the update period.

抄録全体を表示

As digital identity systems continue to evolve, ensuring the secure handling of personal data and authentication tokens on the service provider side has become increasingly critical. Japan's My Number Card is a national identification card that contains an IC chip storing personal information such as name, address, and date of birth, and is increasingly used for digital services. The Digital Authentication App provided by Japan's Digital Agency enables online identity verification using the My Number Card and provides APIs that allow service providers to access user information with the user’s consent. Although the Digital Agency mandates the secure management of such data, its guidelines do not specify concrete technical countermeasures for server-side environments that process and store tokens and personal data. To address this issue, we propose a practical server-side architecture that integrates AWS Nitro Enclaves, a Trusted Execution Environment (TEE), with Japan's Digital Authentication App APIs. Our system confines OpenID Connect (OIDC) token validation and user information access entirely within a TEE, thereby resisting insider threats and compromised operating systems. Furthermore, audit logs are protected with digital signatures and hash chaining, enabling verifiable integrity and non-repudiation. To validate the feasibility and performance of the proposed architecture, we implemented a prototype demo system and conducted measurement experiments. Preliminary measurement results indicate that the integration of TEE introduces limited performance overhead while enhancing security guarantees. This architecture demonstrates a practical and scalable approach to strengthening trust in digital authentication services from the service provider's perspective.

抄録全体を表示

Semigroup theory is a branch of abstract algebra, and it provides mathematical tools for the theory of computation. Finite semigroups can describe state transition systems and thus they model physically realizable computers. Engineering questions like What is the minimal number of states to realize a particular computation? and Which type of computation is more capable? translate into the algebraic tasks of constructing isomorphisms and embeddings between semigroups of different representations. The underlying problem is (sub)graph isomorphism, which is computationally difficult in general. We describe variations of backtrack search algorithms that exploit the algebraic properties of semigroups, various improvements and optimizations. We carry out computational experiments to extend our algebraic knowledge. In particular, we report new computational results on transformation semigroups and on the more general family of diagram semigroups. We study the minimal degree representation problem, count distinct embeddings for three types of diagram semigroups and work on an open problem of embedding into 2-generated subsemigroups.

抄録全体を表示

Spin rate is a critical performance indicator in baseball pitching. However, professional-grade systems are expensive, and even more affordable alternatives, such as sensor-equipped balls, still pose barriers to widespread implementation. This study aims to develop and validate a method for easily measuring baseball spin rate using a standard smartphone camera and machine learning techniques. The proposed method utilizes YOLOv8 for ball detection and cropping, followed by a CNN model to classify the ball’s orientation into four discrete angles (0◦, 45◦, 90◦, 135◦) by leveraging the seam-pattern symmetry. To address the quantization noise and errors inherent in discrete classification, we introduce two key techniques: 1) an outlier correction algorithm based on temporal continuity, and 2) a cumulative linear regression method that estimates rotation speed from phase-unwrapped angles. Experimental results on 15 test videos demonstrated that the proposed method significantly outperformed the baseline angular difference method. Using a success criterion of error ≤ 8.9%, the method achieved a success rate of 90% with an average error of 5.7%, verifying that our approach effectively mitigates discretization artifacts and provides accessible, high-precision measurement.

抄録全体を表示

Existing sequential recommendation models primarily rely on item ID sequences, which ignore semantic information within user behaviors and diverse, dynamic interests. To address these challenges, this paper proposes a unified framework to model users' base and temporary interests with lower resource consumption. Our core method converts item titles, item genres, and user ratings in the order of interaction history into semi-structured natural-language prompts, thereby leveraging the semantic understanding capabilities of a pre-trained BERT model. Building on this, we design a multi-module fusion architecture: (1) a Mixture-of-Experts (MoE) network captures diverse user interests; (2) dedicated temporary interest and base preference modules model users' dynamic and stable tastes, respectively; (3) an adaptive fusion layer integrates these signals for precise prediction. Experimental results on benchmark datasets show that the proposed model outperforms all baseline models on Recall, and most models on NDCG. This study demonstrates that combining pre-trained language models with prompt engineering and multi-interest fusion modeling is a practical pathway toward building recommender systems.

抄録全体を表示

Detecting anomalies in high-performance computing (HPC) systems is essential to keeping them reliable. However, this task is difficult because modern systems are very complex and there is little labeled failure data. Because of this, we need unsupervised methods that can learn only from normal operating data. Many existing methods mainly look at static system states. They often ignore the dynamic behaviors that show how faults start to develop. As a result, they may miss the first signs of problems and only detect failures after they have grown. This paper introduces a Difference Analysis (DA) method to solve this problem. This unsupervised method measures how reconstructed data from an LSTM-based autoencoder (RUAD) changes over time. Unlike methods that use only reconstruction error, DA directly models the instability of system behavior. We evaluated our method on monitoring data from the Marconi100 supercomputer. DA achieved an ROC AUC of 0.881, while the conventional reconstruction error method scored 0.770. These results show that studying dynamic state changes is a more effective way than relying only on static reconstruction error to improve the reliability of large-scale HPC systems. DA may also provide useful signals for anomaly anticipation, which we leave for future study.

抄録全体を表示

In recent years, the rapid advancement of network technologies has led to increasingly sophisticated cyberattacks, posing significant challenges to traditional rule-based anomaly detection systems. This research proposes and verifies a dual-stage network anomaly detection pipeline that integrates machine learning models to identify both known and unknown threats effectively. The proposed system utilizes a Random Forest (RF) model for the classification of established attack patterns and an Autoencoder (AE) for the detection of unknown anomalies through a reconstruction-based approach. To ensure a highly reproducible and realistic experimental environment, the virtual network simulator GNS3 was employed to construct a complex topology including attacker, client, and victim nodes. Network traffic data, comprising normal communications and DDoS attacks (SYN Flood and Slowloris), were collected and processed using the NFStream library to extract key statistical features. The experimental results demonstrate the high performance of the proposed model, with the Random Forest achieving an accuracy of 1.00 in classifying known attacks even within complex traffic scenarios. Furthermore, the Autoencoder successfully identified anomalies with high precision (0.96 accuracy in GNS3-based experiments) by learning normal traffic baselines and detecting deviations through Mean Squared Error (MSE) analysis.

抄録全体を表示

Player authentication is crucial for preserving the integrity of esports competitions, yet current measures cannot entirely prevent impersonation via collusion. While behavioral biometrics offer a promising solution, their application has been largely limited to continuous input devices and homogeneous gameplay states. Taking fighting games as an ideal domain for addressing these gaps, we propose a method that applies keystroke-authentication principles to controller operation dynamics and introduce density-based segmentation to capture context-dependent operation patterns. Experimental results using data from 307 matches by 60 players in Street Fighter 6 demonstrate the effectiveness of our approach, achieving a PR-AUC of 55.7% and an EER of 12.5%. Further analysis revealed that, aggregating decisions using 30 seconds of controller operations, our method can achieve a PR-AUC higher than 90% and EER lower than 1%. These findings validate the applicability of keystroke-dynamics principles to controller dynamics and establish an interpretable baseline.

抄録全体を表示