This paper presents a comparative analysis of two lattice-based post-quantum digital signature schemes: FALCON and SOLMAE. FALCON which was finallty selected by NIST for PQC standardization, represents an efficient realization of the GPV framework over NTRU lattices. SOLMAE, inspired by FALCON, Mitaka, and Antrag, aims to improve implementation simplicity and performance while preserving strong security guarantees. Adopting a pedagogical approach, we provide algorithmic insights into both schemes and conduct practical evaluations using their Python implementations, focusing on key generation, signing, and verification procedures. Performance comparisons at two NIST security levels (512 and 1024 bits) highlight SOLMAE’s potential advantages in simplicity and execution time, suggesting its suitability for deployment in resource-constrained environments.
As cyberattacks become increasingly sophisticated, organizations face an urgent need for timely and accurate incident response to reduce their impact on critical systems. Automating the analysis of network traffic logs has become essential for supporting security analysts and specialists. Although many previous studies have applied machine learning to address this task, they often encounter challenges such as dependence on large-scale analytics platforms, limited exploration of machine learning algorithms, and difficulties in deploying distributed systems due to high costs, complexity, and privacy concerns. To tackle these limitations, we propose a lightweight and accurate machine learning-based framework for the automatic analysis of network traffic logs. Our approach transforms log data into feature vectors using a document-based feature representation method. Experimental results on benchmark datasets demonstrate that our method enables efficient and effective traffic log analysis suitable for practical deployment.
The Ring-LWE problem is a fundamental component of lattice-based cryptography, and evaluating its security is a crucial challenge. The algorithms for solving the Ring-LWE problem can be classified into four categories: lattice basis reduction algorithms, algebraic methods, combinatorial methods and exhaustive search algorithms. However, the combinatorial approach, the Ring-BKW algorithm, remains insufficiently analyzed. The Ring-BKW algorithm primarily consists of two steps, with the Reduction step being the bottleneck because many samples are required for decryption. In existing implementations of the Ring-BKW Reduction step, the block size remains fixed, preventing it from adapting to the sample reduction process and efficiently inducing collisions. In this study, we introduce a method that allows the block size in the Reduction step of the Ring-BKW algorithm to be variable. We propose two approaches: a static decision method, where users manually specify the block size for each reduction step, and a dynamic decision method, where the algorithm autonomously adjusts the block size. The proposed method increases the number of collisions compared to existing methods, resulting in approximately 55-fold and 425-fold more reduced samples for static and dynamic block size selection, respectively, in the Ring-LWE setting with q=17, n=2^4.
In recent years, the adoption rate of virtual reality (VR) technology has been on the rise, and the metaverse is attracting attention as a next-generation form of internet usage. VR offers a variety of applications and content, such as education, gaming, and tourism, where users can remain anonymous and behave as fictional characters. However, there is a possibility that individuals in the real world can be identified from motion data that records VR users’ head-and-hands movements in detail, represented in time-series data. Nair and Lieber demonstrated that publicly available replay data from the VR rhythm game "Beat Saber" can identify individuals with over 90% accuracy. Nevertheless, the features that had the greatest impact on identification accuracy were static attributes such as height and arm length. Therefore, in this study, we attempt to identify individuals in VR domain based on dynamic features such as users’ distinctive ways of moving their arms by employing the DTW (Dynamic Time Warping) distance derived from motion data recorded during VR experiences.
Fast and efficient computation of the Number Theoretic Transform (NTT) is essential for accelerating cryptographic algorithms and large integer multiplications. However, conventional NTT implementations rely on fixed computation paths that fail to adapt to varying input distributions and hardware-specific conditions. In this paper, we propose a HybridNTT that is a hybrid neural network architecture that combines 1D convolutional layers with Transformer encoders to dynamically predict the optimal NTT execution path. The model is trained on a large-scale synthetic dataset comprising various input distributions—uniform, normal, sparse, sorted, bursty, and patterned—alongside multiple NTT parameter settings, including different moduli and transform sizes. Experimental results demonstrate that HybridNTT achieves classification accuracy over 92% and significantly reduces NTT execution time by an average of 34.7%, while maintaining low inference overhead (<5%). These findings highlight the feasibility of machine learning-based optimization for fundamental mathematical operations like the NTT. The proposed method can be seamlessly integrated into cryptographic libraries or hardware accelerators to enable adaptive, context-aware performance tuning, offering a practical step toward intelligent mathematical computing.
