Multivariate public-key cryptography (MPKC) is widely viewed as a promising candidate for post-quantum cryptography, whose security is based on the difficulty of solving systems of multivariate quadratic equations over finite fields. Within MPKC, the Unbalanced Oil and Vinegar (UOV) and its variants have attracted substantial attention. Key-recovery attacks against UOV and its variants can be broadly divided into attacks that compute singular points and attacks that do not. The rectangular MinRank attack, a key-recovery attack, was not originally designed to compute singular points; however, we show that singular points can be computed by modifying its target rank. Moreover, we further optimize this attack by varying the number m^′ of polynomials used in the MinRank condition, analyze how the resulting solution-space dimension depends on m^′, and apply this optimized setting to evaluate the security of UOV and its variants. Finally, we verify that the parameter sets currently proposed for UOV, MAYO, QR-UOV, and SNOVA withstand this attack.

Li et al. (IEEE Transactions on Dependable and Secure Computing 2026) proposed proxy-free public key authenticated encryption with ciphertext update and keyword search (proxy-free PAUKS). In this short note, we demonstrate that keyword information is leaked from updated ciphertexts. We also demonstrate that our attack is effective against the PAUKS scheme proposed by Li et al. (IEEE Transactions on Information Forensics and Security 2023).

Conventional pan-sharpening methods directly fuse panchromatic (PAN) and multispectral (MS) images, often causing artifacts and spectral distortion. In this work, we tackle the problem from an intrinsic image perspective, decomposing the MS image into reflectance (high-frequency textures and structures) and illumination (low-frequency lighting and spatial distribution) components. The PAN image is then used to enhance both components separately through specially designed networks, enabling detail enhancement and accurate color preservation. Finally, the enhanced reflectance and illumination components are recombined to reconstruct the high-resolution MS image. Extensive experiments on multiple datasets demonstrate that our method delivers state-of-the-art performance in both visual and quantitative evaluations.

Zamani, Safkhani, and Daneshpour proposed a Secure Channel Free Verifiable Public key Encryption with Keyword Search (SCF-VPEKS) scheme (Wireless Networks, 2025). Their scheme claims resistance against keyword guessing attacks, file-injection attacks, and malicious servers. In this paper, we conduct a comprehensive security analysis of their construction. We demonstrate that the verifiability mechanism of the scheme is not effective, as a malicious server can return incorrect or manipulated search results that pass the verification procedure performed by the data user (receiver). We further show that the scheme does not satisfy the standard indistinguishability notions for searchable encryption, namely Ciphertext-Keyword Indistinguishability against Chosen Keyword Attack (CW-IND-CKA) and Trapdoor-Keyword Indistinguishability against Chosen Keyword Attack (TW-IND-CKA). Specifically, by exploiting the deterministic and algebraically separable structure of the trapdoor, an adversary can extract a keyword-independent secret component from a single trapdoor and subsequently forge valid trapdoors for arbitrary keywords. As a consequence, both challenge ciphertexts and challenge trapdoors can be distinguished with probability one in polynomial time. Our analysis indicates that these vulnerabilities arise from the core construction methodology of the scheme.

The Forward-Forward (FF) algorithm replaces backpropagation with two forward passes—positive for real data and negative for synthetic inputs—optimizing local ”goodness” measures per layer. By avoiding large-scale gradient storage and repeated backward sweeps, FF significantly lowers computational overhead and energy consumption, making it attractive for low-power or edge devices. On the other hand, in the inference process of the FF algorithm, inference is attempted for each output class, and the one with the maximum output is selected. Therefore, the inference time increases proportionally with the number of output classes. Thus, we propose a binary inference method for the FF algorithm, implemented through a parallel classifier architecture. By using our proposed method, this formulation reduces the number of required forward evaluations from O(N) to O(log N) for an N-class problem. Under sufficient hardware parallelism, this property enables inference latency that is independent of the number of classes. However, this approach may fail to achieve sufficient performance as the number of classes increases. One possible cause is that the information between individual bits is not sufficiently exploited. In this paper, we further demonstrate that inference accuracy can be improved by introducing the concept of hierarchical structure, in which the first stage performs inference using multiple units. The effectiveness of the proposed method is demonstrated on the MNIST database and the EMNIST database. It should be emphasized that the proposed hierarchical inference does not rely on statistical bootstrapping or data resampling. Instead, multiple inference units are trained independently and their goodness outputs are integrated in a deterministic hierarchical manner to enforce consistency among bit-wise predictions.

This paper develops feedback-type stabilization boundary control methods for the Burgers' cellular automaton and the difference Burgers' equation, and provides stability analyses based on ultradiscretization. First, we propose a feedback-type stabilization boundary control law for the Burgers' cellular automaton with one control input term, and prove finite-time stabilization with an explicit upper bound on the stabilization time. Next, by using inverse ultradiscretization, we derive a feedback-type stabilization boundary control law for the difference Burgers' equation with one control input term, and present a method for calculating the convergence value of the closed-loop system. Numerical simulation results demonstrate that the proposed feedback-type control laws stabilize the systems and improve the stabilization performance compared with the previous constant-type control method.

Tensor-computable belief propagation (BP) decoding of low-density parity-check (LDPC) codes, designed to exploit the massive parallelism of general-purpose graphics processing units (GPGPUs), relies on hyperbolic tangent and inverse hyperbolic tangent computations that become numerically unstable under low-precision arithmetic, preventing full exploitation of the throughput advantages offered by GPGPU tensor cores. To overcome this bottleneck, we propose soft-minsum decoding, a tensor-computable algorithm that replaces the minimum operation in the min-sum algorithm with the soft-minimum function derived from the log-sum-exp technique. The resulting check-node update relies solely on exponential and logarithmic operations, which are numerically stable under low-precision arithmetic and naturally expressible as tensor operations. Simulation results demonstrate that soft-minsum decoding achieves BER performance comparable to standard BP, while the float16 implementation attains more than 2 times speedup over tensor-computable BP in float32, confirming the practical benefit of combining numerical stability with low-precision tensor computation.

Maximum likelihood (ML) decoding is the optimal error-correction method but is NP-hard in general, implying that there would be no polynomial-time ML decoding algorithm that capable of decoding arbitrary linear codes. Consequently, researchers have developed a variety of ML decoding algorithms with different characteristics so that they can be used in a complementary manner. Although many modern codes such as Turbo, LDPC, and Polar codes, can be efficiently decoded without ML algorithms, there remains continual demand for developing new ML decoding algorithms and evaluating ML decoding performance in various contexts. In this study, we propose an ML decoding algorithm applicable to general non-binary linear codes, formulated as a generalization of the adaptive and recursive ML decoding algorithm previously developed for binary codes. The proposed algorithm is implemented and applied to Generalized Reed-Muller codes, and its computational complexity of soft-decision ML decoding is discussed and compared with the complexity of the Viterbi decoder, in terms of the number of addition-equivalent operations, to demonstrate the efficiency of the approach. The substantial reduction in decoding complexity enables us performing practical decoding simulation of some classes of Generalized Reed-Muller codes. This paper also shows numerical evaluation results of the error-correcting capabilities of the codes, providing insights into their performance under optimal decoding conditions.

This paper presents an error diagnosis technique based on location variable (LV) simulation, which shortens the processing time for screening error location sets using BDD-based implicit representation. An LV is a Boolean variable which indicates whether the function of the location is modified or not. For each signal value with each signal line, the LV-simulation computes a signal value function in terms of LVs which represents necessary conditions for taking that value. Based on the rectification condition to modify every incorrect primary output value to the correct one, the results of screening error location sets are obtained in a BDD-based implicit representation. Experimental results have shown that the proposed technique reduces the processing time by 60.8% on average.

This paper presents a long-term characterization and modeling of Bias Temperature Instability (BTI) aging in ring oscillators (ROs) fabricated in a 65-nm technology. A dedicated measurement platform was used to apply constant stress and to continuously monitor RO frequencies for 300 days under 125 °C ambient temperature and 1.2V supply voltage. Aging degradation measurements on 840 ROs show that the mean degradation follows a power-law dependence throughout the entire observation period, in agreement with trends reported in shorter-duration measurements. In contrast, the RO-to-RO frequency distribution exhibits a pronounced time-dependent broadening. While the mean frequency decreases only marginally (≈0.55% from 1×10³ s to 1×10⁷ s), the standard deviation increases substantially (≈ 43%) and the temporal fluctuation of individual ROs increases by approximately 3×. Since environmental variations primarily induce a global shift of the distribution rather than a change in its shape, the observed variance expansion is attributed to the enhanced temporal variability of individual devices, consistent with the impact of random telegraph noise (RTN). To capture both the average BTI trend and the evolution of dispersion, we further conduct Monte Carlo simulations based on a hole-trapping framework that jointly accounts for BTI and RTN. By incorporating (i) biased clusters of traps with time constants near the onset of enhanced fluctuations, (ii) a threshold-voltage shift contribution proportional to trap time constants, and (iii) initially pre-occupied traps to represent prior stress history, the proposed model qualitatively reproduces the experimentally observed variance expansion under long-term stress.

Single Source Shortest Path (SSSP) is a fundamental primitive in graph analytics, yet efficient hardware acceleration remains challenging due to irregular graph structures, skewed vertex degree distributions, and frequent update conflicts during parallel edge relaxation. This paper presents a parameterized High-Level Synthesis (HLS) library for dynamically adaptive SSSP acceleration based on the Delta-Stepping scheduling framework. Instead of designing dedicated hardware for individual shortest-path algorithms, the proposed library abstracts representative SSSP algorithms into a unified four-stage processing pipeline with reusable hardware modules and configurable scheduling policies, enabling customized accelerator generation while preserving a common hardware architecture. The proposed design extends the conventional bucket-based Delta-Stepping method by introducing a hierarchical priority queue that preserves local ordering within buckets while supporting flexible scheduling. Based on this framework, a hybrid node-edge parallel execution model dynamically switches between multi-vertex parallelism for sparse regions and edge-partition parallelism for high-degree vertices to improve hardware utilization for irregular graph structures. A triangular systolic-array-based conflict detection module is further integrated into the streaming pipeline to efficiently resolve concurrent updates generated by parallel edge relaxation. Implemented on a Xilinx UltraScale+ FPGA, the proposed architecture achieves up to 1042 MTEPS at 200 MHz with 32 Edge Relax functional units, providing 1.76× higher throughput than our previous FPGA implementation based on conventional Delta-Stepping scheduling, 4.10× improvement over a CPU-based implementation, and up to 22.2× higher performance than representative earlier accelerator systems. The measured peak memory bandwidth utilization reaches 62% under off-chip DRAM execution.

In recent years, IoT has been widely adopted in several fields and is gaining importance in the smart home. The rising need for secure data transmission requires efficient encryption techniques, particularly when applied to real-time data transmission in home automation systems. In this paper, we introduce a hybrid Advanced Encryption Standard-Galois/Counter Mode (AES-GCM) cryptographic mechanism with Authenticated Associated Data (AAD). Symmetric key encipherment methods like AES hold great potential for incorporation into smart home systems. As investigated, the motivated GEM demonstrates significant promise for establishing IoT-aligned real-time data communication. The suggested hybrid model primarily functions to provide confidentiality, authentication, and data-integrity across the sensor, processor, and actuator layers in a home automation environment. To enhance computational efficiency and minimize energy consumption, the hybrid algorithm is deployed using a hardware-software co-design approach. Experimental analysis demonstrates that the proposed co-designed AES-GCM implementation achieves significant improvements in encryption /decryption time, as well as energy efficiency, for different key lengths. The results demonstrate that the suggested hybrid approach outperforms other state-of-the-art models, offering robust data security and energy-aware performance suitable for a large-scale, real-time home automation.

This paper proposes MoE-RD^* Lite, a risk-aware dynamic path planning algorithm that extends D^* Lite with a deterministic mixture-of-experts (MoE) heuristic. The method adaptively balances path distance, graded hazard risk, and corner safety according to local map features. Hazard regions are represented by continuous risk fields, and corner-aware guidance discourages unsafe diagonal transitions. Simulation and ablation results show that MoE-RD^* Lite achieves a favorable distance-safety trade-off and reduces search load in the evaluated scenarios.

This study examines the adaptive consensus problem for nonlinear multi-agent systems under matrix weighted network topology. To establish the consensus problem for such networks, we derive a fundamental eigenvalue inequality for positive semi-definite matrices. We then introduce a distributed control protocol that relies exclusively on local neighbor interactions and utilizes an adaptation mechanism to monotonically increment the coupling weights. Analytical results confirm that this approach effectively achieves consensus of matrix weighted network, with the adaptive gains remaining bounded and eventually settling at fixed steady-state values. The efficacy of this control strategy is verified through numerical simulations of a matrix-weighted multi-agent system.

Lattice trapdoors can be used to construct advanced cryptographic applications, such as group signatures, attribute-based encryption, and fully homomorphic encryption. The current state-of-the-art implementation of lattice trapdoors is based on the gadget sampling algorithm (Genise and Miccianio, Eurocrypt 2018), which can be split into the online phase and the offline phase. However, for fixed-dimensional lattices, the integer discrete Gaussian sampler has become a performance bottleneck for lattice trapdoors. In this work, we propose a perturbed gadget sampling algorithm (PGSample) and introduce an implementation of a sampler utilizing a reversed cumulative distribution table (SampleI-RCDT). This approach optimizes sampling performance while facilitating efficient trapdoor implementations. For different parameter sets, our online phase uses desktop-level CPU execution times as fast as 0.25ms. Compared to previous work, our PGSample-based trapdoor outperforms previous methods by 43.3% to 46.8% overall and achieves a 5.5×-6.2× speedup in the online phase. In conclusion, this work provides a more practical and flexible solution for distributed cryptosystems and constrained devices.

Software supply chain security has become increasingly important, and Software Composition Analysis (SCA) tools are expected to serve as a primary means for identifying open-source software (OSS) components and generating Software Bills of Materials (SBOMs). However, prior research has mainly focused on technical performance, such as detection accuracy in identifying OSS components and vulnerabilities, and has not sufficiently examined the differences in expectations and perceptions between vendors who design and provide these tools and users who deploy them in practice. In this study, we conducted semi-structured interviews with SCA tool vendors and enterprise users to qualitatively analyze perception gaps among stakeholders. In addition, we presented SCA tool accuracy evaluation results to vendors and users. We then collected their interpretations and reactions to the concrete performance data. Our analysis revealed three gaps: (1) a gap between the high level of accuracy expected by users and the actual accuracy of current tools, (2) differences among vendors regarding the perceived role and responsibility of SCA tools, and (3) a mismatch between vendors' assumptions of continuous use during development and users' actual practices centered on acceptance testing. Furthermore, we found that regulatory compliance is often a primary motivation for adoption, suggesting that SBOM generation can become an end in itself and may lead to “compliance theater,” where the process does not necessarily result in meaningful risk reduction. This study highlights the need for vendors to clarify intended use cases and detection limitations, for users to define their usage requirements more explicitly, and for the broader ecosystem to establish third-party evaluation infrastructures and standardization. Our findings provide design and operational implications for improving the effective use of SCA tools in software supply chain security.

This paper studies the key control security of key derivation functions (KDFs) based on the lightweight pseudorandom function Light-MAC.We first analyze KDFs constructed from LightMAC in the CTR, FB, and DP modes specified in NIST SP 800-108r1. We show that when the output is a single block, all these constructions fail to achieve key control security. We then analyze the multi-block output setting, focusing on the double-block output case of CTR-LightMAC. We prove that key control security improves compared to the single-block case, and for specific parameter choices such as (n, s) = (64, 24), the security bound improves significantly, where n is the block length of the underlying block cipher and s is a length of the counter used in LightMAC. To improve the insecurity of the single-block case, we propose LightMAC-DM, which incorporates the Davies-Meyer construction into the final processing step. We prove that LightMAC-DM preserves the original PRF advantage while improving key control security.

In the past decade, the construction of quantum maximum-distance separable (MDS) codes has been extensively and deeply explored. For lengths of the form $n=\frac{q^2-1}{m}$, where m is an integer such that m|q+1 or m|q-1, comprehensive results are available. In this paper, we investigate the case where the length $n=\frac{q^2-1}{m}$, yet m is neither a divisor of q-1 nor q+1. Ultimately, this problem can be reduced to identifying the necessary and sufficient conditions for the existence of pairs (m₁, m₂), where $m=\frac{m_1\cdot m_2}{m_1+m_2-1}$ is an integer. In addition, the greatest common divisor of m with both m₁ and m₂, denoted as gcd(m, m₁) and gcd(m, m₂), must be greater than 1. The quantum MDS codes presented in this paper have not been previously investigated and their minimum distances exceed $\frac{q}{2}$.

The fractional tap-length algorithm provides an effective approach for adjusting the tap-length in adaptive algorithms. Tap-length adaptation step-size (TASS) plays an essential role that determines tap-length estimation precision. However, some current adjustment strategies for TASS have many problems, such as failing to eliminate the influence of system noise, being either too conservative or too aggressive and lacking reasonable analysis of the parameters used. As a result, the performance of these algorithms is generally average and their practicality is limited. Based on the theoretical foundation of previous algorithms, an improved adjustment method for TASS is presented. This method uses iterative values of the ratio between instantaneous and cumulative noise-free prior error as the core adjustment variable to reasonably adjust TASS. The proposed algorithm uses few parameters and provides a reasonable analysis on them. It does not employ complex adjustment strategies or restrictive mechanisms. The proposed algorithm ensures a balanced relationship among convergence performance, steady-state performance and robustness.

Frequency-hopping sequences with low Hamming correlation are widely employed in multiple-access communication systems. Among these, the one-coincidence frequency-hopping sequence (OC-FHS) set is specifically designed to minimize mutual interference between different users. In this letter, we construct a new class of OC-FHS sets based on circular Florentine rectangle. The proposed OC-FHS set is shown to be optimal with respect to the Peng-Fan bound. Moreover, the size of the constructed set attains the known theoretical upper bound when N is an odd prime.

This paper proposes a new enhancement method for color images that can preserve the mean intensity of the input image. By performing processing in the HSI color space, which has the same color gamut as the RGB color space, the proposed method preserves hue and enables effective enhancement of saturation. The proposed method introduces generalized histogram equalization (GHE), which has been shown to be effective for gray-scale images, to enhance the intensity and saturation components of color images. By introducing two parameters, the proposed method preserves the mean intensity of the input image and enhances both the contrast and colorfulness of the color image. The effectiveness of the proposed method is demonstrated by actual image processing results.

Knowledge distillation is a widely used technique for transferring knowledge from a large teacher model to a smaller student model. While this approach has achieved remarkable empirical success across various domains, theoretical understanding of the approximation error when the student model has strictly less representational capacity than the teacher remains limited. Existing theoretical analyses have primarily focused on self-distillation settings where teacher and student share the same architecture, leaving a fundamental gap in our understanding of capacity-mismatched distillation. In this paper, we address this gap by analyzing knowledge distillation in multivariate polynomial regression, where the teacher and student models are polynomials of different degrees. We derive a closed-form expression for the optimal student model under KL divergence minimization and prove that the mean squared error between teacher and optimal student can be exactly characterized using the Schur complement of the feature covariance matrix. This characterization reveals that the Schur complement, which represents the conditional covariance of high-order features given low-order features, serves as a distillation difficulty indicator that quantifies how input distribution geometry affects knowledge transfer. Our analysis provides the first exact closed-form expression for approximation error in capacity-mismatched knowledge distillation, connecting this machine learning problem to classical results in linear statistical models. We extend our analysis to polynomial logistic regression and demonstrate that under high-temperature approximation, the expected KL divergence exhibits a specific scaling relationship with temperature. Comprehensive numerical experiments validate our theoretical predictions across various settings.

Although current digital identity systems are centralized, decentralized systems based on Verifiable Credentials (VCs) are gaining attention and moving towards practical implementation. As one of VC systems, a VC system with selective disclosure has been proposed, where credentials are represented as directed graph based on the concept of Linked Data (LD). However, in the existing VC system, the verification time increases depending on the number of Resource Description Framework (RDF) terms that correspond to vertices and edges in the proved graph, due to the characteristics of the utilized signature scheme. Meanwhile, a zero-knowledge proof system for directed graphs using AHO signatures and a pairing-based accumulator has been proposed. This system is characterized by its verification time and proof data size being independent of the number of vertices and edges in the graph. In this paper, we propose a LD-based VC system with selective disclosure that leverages the zero-knowledge proof system on graph; the verification time and proof size are independent of the number of vertices and edges. Furthermore, we reduce the proof data size by modifying the signature scheme from AHO signatures to SPS-EQ signatures and from the pairing-based accumulator to a set commitment. We implement and evaluate the proposed system on a PC.

Laser microphone is a remote eavesdropping device that illuminates a laser beam on a target object and reconstructs acoustic vibration by analyzing reflected light. In 2022, Laser Meager Listener showed the advantages of a laser microphone using a laser Doppler vibrometer (LDV), but the need for a scientific-grade LDV increases the attack cost and limits the flexibility. This paper addresses the issues by using self-mixing interferometry (SMI), which can realize an LDV with a single laser diode and simple electronics. Our proof-of-concept SMI-based laser microphone is evaluated in an end-to-end attack scenario considering an attacker eavesdrops on sound from a target object 3 meters away through a glass window. We evaluate the intelligibility of sound measured from five target objects under three propagation conditions, and the proposed method achieves a performance comparable to the previous attack using a scientific-grade LDV. Furthermore, this paper also studies how an attacker maximize the attack efficiency by choosing a target object with desirable physical properties through characterization of the attack with various targets with different materials and dimensions, confirming that high-frequency components and high reflectance are the key physical parameters. The proposed method can be extended to an invisible attack using an infrared laser, and the attack cost can be further reduced with our open circuit designs. We finally discuss the physical parameter that limits the attack distance.

This paper proposes card-based physical zero-knowledge proof (ZKP) protocols for Sudoku that require only a single shuffle. A zero-knowledge proof allows a prover to convince a verifier that a solution to a computational problem exists without revealing any information about the solution. Physical ZKPs implement this concept using tangible objects such as playing cards. In card-based cryptography, protocol efficiency is typically measured by the number of cards and shuffle operations required. Previous card-based ZKP protocols for Sudoku required multiple shuffles to verify correctness. Earlier studies gradually reduced this number, and the best known protocol before this work required two shuffles. In this paper, we further improve the efficiency of these protocols by presenting the first Sudoku ZKP protocol that uses only one shuffle. The key idea is to prepare inputs in the setup phase using private permutations (PPs). A private permutation is a deterministic rearrangement known only to the player performing it. Unlike shuffles, private permutations do not increase shuffle complexity and are easy to performin practice. By applying multiple private permutations interactively between the prover and the verifier, the protocol restricts possible card arrangements before the main verification phase, allowing the verifier to check correctness using only one shuffle. Based on this idea, we construct three protocols. Protocol 1 is a base protocol that uses 2n private permutations in the setup phase. Protocol 2 introduces a two-color card system to reduce the number of private permutations. Protocol 3 further optimizes the setup using multiple card colors, reducing the number of private permutations. All three protocols satisfy perfect completeness, perfect soundness, and the zero-knowledge property. For a standard 9 × 9 Sudoku puzzle, the protocols require 162 cards and only one shuffle, allowing implementation using three standard playing-card decks. Our results show that interactive input preparation reduces shuffle complexity and may also improve physical ZKP protocols for other puzzles.

This letter presents an analytical framework for evaluating the average block-level error rate (BLER) of short-packet communications (SPC) operating over Fisher-Snedecor F composite fading channels. By effectively resolving the complex integrals required for performance evaluation, we introduce a highly efficient, low-complexity analytical formulation based on the Beta function. This approach yields an exact, closed-form expression for the average BLER, eliminating the need for computationally intensive numerical integrations or infinite series expansions. Extensive Monte-Carlo simulations corroborate our theoretical derivations, providing insights into the impacts of multi-path scattering and heavy-tailed shadowing on the reliability of latency-critical wireless networks.

This paper presents a disturbance-resistant initial alignment method for vehicle-mounted strapdown inertial navigation systems (SINS). The objective is to estimate the initial attitude of a vehicle-mounted SINS under quasi-static conditions in the presence of engine-induced and human-induced disturbances. To enhance alignment robustness, a vectorized K-matrix filtering framework is developed for recursive initial attitude estimation. The main contributions are threefold. First, a simplified ninedimensional state-space model is derived by exploiting the symmetry and zero-trace properties of the Davenport K-matrix. Second, a Kalman filter (KF)-based recursive estimation framework is established directly in the vectorized K-matrix domain to suppress inertial sensor noise. Third, the full K-matrix reconstruction and attitude recovery procedure are explicitly formulated to improve reproducibility. Experimental validation is conducted using a dedicated vehicle-mounted SINS platform. Results indicate that the proposed method effectively mitigates the influence of external disturbances and improves azimuth repeatability. Multi-orientation tests show that the yaw repeatability error is reduced by 56.6% on average compared with the conventional two-stage Kalman filter approach.

Swarm control of multiple unmanned surface vehicles (USVs) has drawn significant attention, particularly in target encirclement. Traditional methods often involve direct encirclement, which may provoke unpredictable reactions from a target sensitive to its surroundings. This study proposes an innovative adaptive stochastic priority stack method (ASPSM). Based on a reinforcement learning autonomous scoring mechanism and integrated with distributed control, ASPSM enables each hunter to make optimal decisions using only local information. Theoretical analysis confirms that hunters can covertly encircle the target without detection. Simulations further demonstrate that ASPSM achieves effective encirclement even against dynamically and randomly moving targets.

We extend automaton-based multi-agent reinforcement learning to continuous state spaces by integrating deep reinforcement learning, while preserving a single global linear temporal logic (LTL) specification through an automaton-driven distributed architecture. Specifically, we propose centralized and distributed multi-agent deep reinforcement learning formulations. The first is a centralized formulation based on a joint product Markov decision process (MDP), which preserves the global specification but suffers from combinatorial growth in the joint action space. The second is a distributed formulation in which each agent learns over a local product MDP, while an aggregator updates a shared automaton state from aggregated labels and assigns specification-aware rewards. Experiments on a surveillance task in a continuous state space show that the distributed formulation maintains global LTL consistency and improves learning efficiency while reaching higher performance more reliably than the centralized formulation, and that reward allocation through the aggregator further accelerates convergence.

Multi-view pedestrian tracking (MVPT) aims to track pedestrians in the form of a bird's eye view (BEV). In MVPT, re-identification-based (ReID-based) end-to-end methods are one of the most effective approaches. They detect pedestrians and extract ReID features representing their appearances within a single model. These methods generate a BEV feature by projecting image features from multiple views into the BEV space via a perspective transformation, and then perform pedestrian detection and ReID feature extraction using that BEV feature. One problem with these methods is that the pedestrian body structure collapses in the BEV feature due to distortion caused by the perspective transformation. This prevents the model from fully capturing appearance information on pedestrians, leading to insufficient ReID features. To overcome this problem, we propose a new ReID-based end-to-end MVPT method, named MRIS. MRIS captures pedestrian appearance information from each view image space, where the pedestrian body structure is preserved. To achieve this in end-to-end MVPT, MRIS leverages back-projection from the BEV space into each view image space, which is the inverse projection of the perspective transformation. First, MRIS back-projects the BEV feature and detected pedestrian locations in the BEV space to each view image space. Back-projecting pedestrian locations enables the model to extract appearance features, which represent the appearance information, from image features for each pedestrian and each view. Back-projecting the BEV feature complements the information of occluded parts in each view image, helping the model extract effective appearance features. Then, since pedestrian appearance varies from one view to another, MRIS aggregates appearance features from multiple views into a single ReID feature for each pedestrian. Extensive experiments demonstrate that MRIS outperforms previous ReID-based end-to-end methods and show that capturing appearance information from the image space leads to better tracking performance than doing so from the BEV space.

Stable matching is an important problem that receives attention from researchers in several fields. In the problem setting, there are two sets with the same number of members. Each member has its matching preference. The goal is to find a one-to-one matching between each member of the two sets such that no pairs want to change the matching result. Since an instance of the stable matching problem may have more than one possible stable matchings, Nakamura et al. proposed a multi-stakeholder environment with selectability property, and applied it to the stable matching problem as an example use case. In their setting, the computing server could freely choose to return any stable matching depending on the benefits of the clients and the computing server. Their protocol also offered verifiability, but only against a semi-honest verifying server. To address this issue, we propose a privacy-preserving verification protocol for stable matching against a malicious server. Our verification protocol is constructed from private CDS schemes for stable matching, which do not require any asymmetric-key cryptographic primitives. From the implementation results, our proposed protocol is 4 to 5 orders of magnitude faster than the previous work.

In private information retrieval (PIR), a client retrieves an entry from a database without letting the database server know which entry is retrieved. Recently, hint-based PIR schemes have been proposed, where the client receives some hints about the database in the setup phase before the actual query begins. As an advantage, these schemes achieve sublinear database server online computation. On the other hand, the client needs to preprocess the hints with the database server, which can be costly. Moreover, the client needs to store some amount of hints, even if the client plans to query only once or a few times.

In this paper, we propose the first hint-based PIR with commodity servers, which we call CHOO-PIR. The purpose of the commodity servers is to manage the hints, so that the client does not need to be involved in the preprocessing, while the scheme can maintain the sublinear online computation. In order to preserve the privacy of the client against the commodity servers, we apply fully homomorphic encryption (FHE) or secret sharing (SS) schemes. Our FHE-based scheme achieves both client computational cost of e Õ(1) and database server online computational cost of e Õ(\sqrt{n}), where n is the database size. While the FHE-based scheme requires publickey cryptography, we can avoid such operations in the SS-based scheme, with a trade-off of client computational cost and online communication cost. The actual performance of the SS-based scheme is demonstrated through implementation results. The results show that CHOO-PIR has an advantage when the database size is large and the number of queries issued by resource-constrained clients is small.

Acoustic scene classification (ASC) aims to automatically identify the surrounding acoustic environment from audio recordings. The core objective of ASC lies in recognizing and modeling key acoustic events embedded in complex and diverse sound scenes. However, existing ASC approaches often fail to fully capture the relationships among heterogeneous acoustic events occurring at different temporal scales, thereby limiting their ability to represent scene-specific characteristics. To address these challenges, this paper proposes a hierarchical feature aggregation framework based on a convolutional neural network (CNN) and a Transformer, termed HFA-CT, which is designed to effectively extract and model key acoustic events for ASC. First, log-mel spectrograms are extracted from raw audio signals. Next, the spectrograms are processed by a convolutional neural network to obtain feature maps at multiple depths. To explicitly capture local acoustic events, a Group-wise Local Average Pooling (LAP) layer is applied to each depth-wise feature map by evenly partitioning the temporal dimension, enabling the extraction of localized event representations at multiple temporal resolutions. The resulting multi-depth local features are then fused and fed into a Transformer encoder, which models interdependence among acoustic events at different levels of detail. Finally, a softmax function is employed to produce the scene prediction. Experimental evaluations conducted on the DCASE 2018 and DCASE 2019 benchmark datasets demonstrate that the proposed method achieves classification accuracies of 80.86% and 81.48%, respectively, validating the effectiveness of the proposed hierarchical feature aggregation strategy for acoustic scene classification.

Modern job recommendation systems are expected to provide not only accurate job matching but also long-term career development support under dynamically changing labor market conditions. However, many existing approaches mainly focus on short-term recommendation accuracy while insufficiently considering multiple conflicting objectives, recommendation diversity, and evolving user preferences. To address these limitations, this paper proposes a personalized job recommendation framework based on Multi-Objective Deep Reinforcement Learning (MODRL). The recommendation process is formulated as a Markov Decision Process (MDP), where user skill profiles, job requirements, and historical interaction information are jointly modeled as environment states. The proposed framework simultaneously optimizes three objectives: match quality, recommendation diversity, and career advancement potential. A multi-objective reward mechanism is designed to balance skill-based matching accuracy, reduction of redundant recommendations, and long-term skill development opportunities. Experiments were conducted on a job recommendation dataset containing more than 100,000 user-job interaction records and over 500 job postings across 10 technical skill domains. The proposed MODRL model was compared with several baseline methods, including Content-Based Filtering, Collaborative Filtering, Matrix Factorization, Deep Neural Networks, and single-objective Deep Q-Networks. Experimental results demonstrate that the proposed approach achieves superior recommendation performance with an F1-score of 0.994 and an AUC of 0.999. Compared with single-objective reinforcement learning, the proposed multi-objective framework improves the F1-score by 9.98%. Feature importance analysis further indicates that Match_Score is the most influential factor in recommendation decisions. In addition, the experimental results reveal a clear decision boundary around Match_Score = 0.80, which provides useful guidance for practical deployment in intelligent job recommendation systems.

Multi-designated verifier ring signatures (MDVRS) are an anonymous signature primitive proposed by Kolby et al. (IACR CIC 2024) in which only designated verifiers can verify signatures. They formalized MDVRS and constructed an MDVRS scheme using a specific intermediate primitive, named provably simulatable designated-verifier ring signatures (PSDVRS). However, PSDVRS is known only under the discrete logarithm assumption. In this paper, we propose a new generic construction of MDVRS based on well-known primitives: public-key encryption schemes, one-time signatures, ring signatures, and non-interactive zero-knowledge arguments. As these building blocks can be constructed from various computational assumptions, we can realize MDVRS from a wide range of computational assumptions. Namely, we obtain post-quantum MDVRS for the first time. Furthermore, our generic construction achieves stronger security than the original scheme.

In recent years, methods that enable robots to generate actions based on human instructions have attracted significant attention. In cooking tasks in particular, robots must determine whether to continue linguistically specified actions, such as “frying” or “mixing,” based on the visually observed state of the ingredients. This study focuses on the decline in learning performance caused by the characteristics of training data of cooking task monitoring with CLIP models, which are capable of training correspondences between images and texts. In conventional task monitoring datasets, all frames—from the beginning to the end of the cooking process—are paired with a single recipe text, such as “Fry the chicken until golden brown.” In other words, one recipe text is associated with multiple images representing different ingredient states, resulting in highly ambiguous data. For example, in the initial frames, the ingredients have not yet been cooked to the state described by the recipe text, resulting in a mismatch between the image and the text and therefore low-quality paired data. Because CLIP models are trained under the assumption that the image and text in each pair are semantically aligned, training with such mismatched data degrades learning quality and leads to poor task-monitoring performance. To address the problem of degraded learning caused by ambiguous training data, we propose CookClip, a training method that enables the model to accurately learn the relationships between recipe texts and temporally sequential cooking images. This approach enhances the CLIP model's ability to monitor cooking tasks more effectively. Furthermore, we constructed a dataset called Time-series Cooking Image, specialized for cooking task monitoring. It consists of sequences of cooking images and corresponding recipe texts, and is used for both training and evaluation. This dataset includes image sequences corresponding to cooking initiation, successful completion, and failed outcomes—features that were not present in previous datasets. We evaluated the task completion detection performance of the proposed method, based on the error between the correct completion frame range and the predicted frame. A smaller error indicates that the model more accurately captures the temporal progress of cooking and successfully trains the correspondence between the recipe texts and cooking images. The error between the correct frame range and the output frame was 3,077 frames with the original CLIP, while it was reduced to 51 frames with the proposed method, confirming the effectiveness of our training approach. The dataset and construction method are publicly available at: http://isl.sist.chukyo-u.ac.jp/archives/TSCI.

In a secret sharing scheme, a dealer generates n shares from a secret to share the secret among n participants. We focus on an (s, t, n)-ramp scheme satisfying (i) the secret is correctly recovered by the shares of t or more arbitrary participants, and (ii) no information on a secret is revealed from the shares of s or less arbitrary participants. We propose a novel construction of a $(2,p-1,\frac{3p-7}{2})$-ramp scheme for a one-bit secret, where p ≥ 7 is an arbitrary prime number. This scheme has a property that a secret is reconstructed by arbitrary p - 1 participants in an anonymous way, i.e., without revealing their identities, by multiplying their shares.

Card-based cryptography enables the execution of cryptographic protocols such as secure multiparty computation, by using a deck of physical cards. To facilitate statistical analysis with card-based cryptography, we have previously presented protocols for integer-based arithmetic operations. These protocols utilize integer commitments, which are face-down card sequences encoding integers, and consist of simple operations. However, in statistical applications, they suffer from poor scalability, as the required number of cards increases significantly as the number of data points increases. To address this issue, this paper introduces a new representation of commitments, named floating-point number (FPN) commitments, which are face-down card sequences encoding floating-point numbers. These FPN commitments allow us to represent a wide range of numbers, including negative and arbitrary-precision values, with fewer cards. We then propose protocols for converting a conventional integer commitment to an FPN commitment and for performing arithmetic operations using FPN commitments. We also demonstrate that our new representation and protocols enhance the applicability of card-based cryptography to statistical processing.

BINGO is a classic game where players compete to complete a line on a 5 × 5 grid, and it is fun and popular. However, since it typically requires a specialized bingo machine, spontaneous games of BINGO are challenging. To address this, we introduce Suken BINGO, a new board game that allows players to declare integers instead of using a bingo machine, making it more accessible. Specifically, players verbally call out integers in a manner similar to the Japanese hand game Suken. This removes the need for a dedicated bingo machine.

Furthermore, we propose Secret Suken BINGO by combining Suken BINGO with card-based cryptography. In this secure variant, players conceal the integers on their bingo cards throughout the game, which adds an element of psychological maneuvering and strategic depth. We then detail the implementation and discuss the security and efficiency of the proposed protocol.

QR-UOV is a multivariate signature scheme derived from UOV that achieves compact public keys by exploiting quotient-ring structure, making it a promising candidate for post-quantum digital signatures. In QR-UOV, most parts of the public map are derived from the public key seed using a PRG. This public key expansion for QR-UOV includes rejection sampling to generate coefficients uniformly over 𝔽_q, since QR-UOV uses a small odd-prime base field. However, this rejection sampling introduces extra data movement and irregular control flow. For the recommended parameter set, public-key expansion accounts for nearly 90% of the QR-UOV verification time.

In this paper, we propose No Rejection Sampling (NoRS) QR-UOV, a variant of QR-UOV with rejection-sampling-free public-key expansion, which leaves the generation of secret-dependent coefficients unchanged. Concretely, the rejected value q is deterministically mapped to 0, which simplifies coefficient generation but introduces a slight bias in the resulting coefficient distribution. We evaluate the security impact of this modification through both theoretical and concrete analyses. Our results suggest that, for the proposed parameter sets, NoRS QR-UOV maintains the claimed security levels against the considered attacks.

On the implementation side, we develop a high-speed implementation of NoRS QR-UOV for x86 processors with AES-NI and AVX2. Benchmark results on a Skylake platform show that NoRS consistently accelerates QR-UOV at all security levels, with the largest gain in signature verification. Under the standard Round-2 parameter settings, benchmark results show that NoRS QR-UOV reduces the verification cost from 0.43 to 0.30 Mcycles at security level I, with comparable improvements at security levels III and V, corresponding to about 1.4× speedup over standard QR-UOV. Overall, the results suggest that relaxing coefficient uniformity in public-key expansion is a practical and effective design choice for QR-UOV.

This paper presents the security evaluation of masking countermeasures against a deep-learning-based side-channel attack (DL-SCA) which targets many post-quantum key encapsulation mechanisms (KEMs). The attack proposed in 2022 exploits side-channel information leaked during the re-encryption process of the Fujisaki-Okamoto (FO) transformation in the decapsulation phase using deep learning techniques. The obtained sensitive information is then used in a chosen-ciphertext attack (CCA) to recover the secret key. Since the FO transformation is commonly used in most post-quantum KEM schemes, the effectiveness of countermeasures against the DL-SCA is critical for tamper-resistant post-quantum KEM implementation. We apply the attack to a set of implementations with masking countermeasures for the post-quantum KEM and experimentally analyze the impact of various factors on the attack success rate, including side-channel information acquisition methods (e.g., sampling rate, points of interest, and segmentation) and neural network architectures. Finally, we demonstrate that masking of at least the 3rd order is required to ensure sufficient security against the DL-SCA.

This study is devoted to development of a new data-driven control method for the automatic path following problem of mobile robots based on just-in-time modeling. Especially, this letter proposes a control method that utilizes a database for experimental data of driving and makes the mobile robot follow a sequence of target points on a given path in series. Then, from experimental results using a real mobile robot, it can be confirmed that automatic path following control for the mobile robot is realized with small error, hence the proposed method has the effectiveness and the applicability.

Chen et al. (IEEE Transactions on Cloud Computing 2025) introduced Public-Key Encryption with Similarity Test (PKEST) for the electronic medical records classification shared in medical consortia. They claimed that their PKEST scheme is secure against chosen-ciphertext attack (CCA). In this short note, we demonstrate that the PKEST scheme is not CCA secure.

The room impulse response (RIR) characterizes sound propagation in a room from a loudspeaker to a microphone under the linear time-invariant assumption. Estimating RIRs from a limited number of measurement points is crucial for sound propagation analysis and visualization. Physics-informed neural networks (PINNs) have recently been introduced for accurate RIR estimation by embedding governing physical laws into deep learning models; however, the role of network depth has not been systematically investigated. In this study, we developed a deeper PINN architecture with residual connections and analyzed how network depth affects estimation performance. We further compared activation functions, including tanh and sinusoidal activations. Our results indicate that the residual PINN with sinusoidal activations achieves the highest accuracy for both interpolation and extrapolation of RIRs. Moreover, the proposed architecture enables stable training as the depth increases and yields notable improvements in estimating reflection components. These results provide practical guidelines for designing deep and stable PINNs for acoustic-inverse problems.

A secret sharing scheme is an important cryptographic primitive. Recently, Bogdanov et al. [4] unveiled the condition under which the existence of a function with a certain property is equivalent to the existence of a secret sharing scheme for a one-bit secret. In this paper, we first give a new intuitive condition that is equivalent to the condition in [4]. Next, we give a new (t, p)-threshold scheme for a one-bit secret using a basis of the p-dimensional vector space of a finite field 𝔽_p, where p ≥ 3 is an arbitrary prime and t is an arbitrary integer satisfying 2 ≤ t ≤ p - 1. The security of the (t, p)-threshold scheme is established by using the discrete Fourier transform of a function f corresponding to the scheme. This scheme can be extended to the case where a secret takes values in 𝔽_p.

The information bottleneck (IB) method seeks a compressed representation of data that preserves information relevant to a target variable for prediction while discarding irrelevant information from the original data. In its classical formulation, the IB method employs mutual information to evaluate the compression between the original and compressed data, as well as the utility of the representation for the target variable. In this study, we investigate a generalized IB problem, where the evaluation of utility is based on the H-mutual information that satisfies the concave (CV) and averaging (AVG) conditions. This class of information measures admits a statistical decision-theoretic interpretation via its equivalence to the expected value of sample information. Based on this interpretation, we derive a block coordinate descent algorithm to assess the tradeoff between compression and utility in the generalized IB problem.

In this paper, we proposed an initialization method for the Forward-Forward algorithm using M-sequences as pseudo-random sequences and demonstrated its effectiveness using the MNIST database. Our results show that this approach can accelerate learning while maintaining diversity through randomization, not only in deep learning with backpropagation but also in the Forward-Forward algorithm.

To enhance feature extraction in speech signals and prevent critical scale information from being diluted by redundant noise, we propose an AdaBoosted Multi-scale & Multi-feature Model (AMSF). The AMSF model integrates multi-scale convolution with AdaBoost and introduces a global perception fusion module, which helps to emphasize informative emotional features while suppressing noise. The model extracts MFCC and Logfbank features from speech signals and uses Pearson's correlation coefficient for feature fusion, enhancing discriminative power. In addition, DiffGrad optimization is adopted to stabilize the training process, accelerating convergence and improving model performance. Experimental results on the improvised subset of the IEMOCAP dataset show that the AMSF model achieves Weighted Accuracy (WA) of 86.17%, Unweighted Accuracy (UA) of 83.03%, Macro F1 score of 83.69%, and Weighted F1 score of 85.80%.

With the increasing complexity of modern power systems and the growing variability of electricity demand, real-time load regulation has become a critical challenge. In particular, regional power networks with heterogeneous load characteristics require efficient and adaptive control strategies. Although existing studies have explored load grouping based on consumption patterns to improve grid management, most approaches rely on static grouping schemes and lack the capability for real-time hierarchical regulation. To address these limitations, this paper proposes a Greedy-Based Hierarchical Hopfield Neural Network (GBH-HNN) for adaptive load grouping and real-time regulation. In the proposed framework, controllable loads are first organized into priority-based groups, while a hierarchical control structure is introduced to reflect different operational levels of the power system. The GBH-HNN then employs a greedy neuron update mechanism, which selectively adjusts the most critical load groups at each iteration, thereby achieving efficient and lightweight real-time regulation. The effectiveness of the proposed method is evaluated using publicly available household electricity consumption data under Hubei-oriented operational scenarios. Experimental results demonstrate that the proposed approach reduces the maximum load imbalance from 28.70 to 15.89 and the steady-state error from 4.64 to 2.46. In addition, it achieves faster convergence and a regulation accuracy of 84.48%, indicating its superiority in both efficiency and accuracy for real-time load regulation.

Automatic knowledge extraction is one of the most important goals of Natural Language Processing. Especially after the emergence of COVID-19, the number of related literature is growing by about ten thousand per month, significantly challenging manual annotation and downstream tasks. In this paper, we describe a system for biomedical multi-label topic classification. Firstly, BERT is pre-trained on biomedical corpora which helps to capture deep semantic information. Furthermore, we fine-tune the pre-trained BERT on the COVID-19 literature from the Lit-Covid Database. Finally, automatic correction for Biomedical Multi-label Topic Classification method is introduced to our system to effectively take advantage of the domain expert experience. Our Act-BERT model achieves a micro F-score of 91.75% and macro F-score of 89.28% in the test set, and the macro F-score is 0.53% higher than the best system, which demonstrates the potential and effectiveness of the proposed framework.