This work presents a comparative statistical investigation into the enhanced ionizing radiation hardness between the Octagonal (OnM) and Conventional (CnM) MOSFETs under X-ray environment. The primary objective is to quantify the robustness of OnM across varying Total Ionizing Dose (TID) levels and biasing conditions. A comprehensive statistical framework was applied, integrating Z-score normalization, K-means clustering, Cramer’s V, and association analysis to evaluate device variability. Experimental results demonstrate that OnM devices exhibit significantly superior electrical consistency, evidenced by a markedly lower coefficient of variation (C.V.) for the on-state drain current (0.1020 vs. 0.5689) and higher cluster cohesion compared to CnM counterparts. Furthermore, a Bootstrap validation confirmed the high stability of OnM clusters (Jaccard coefficient minimum value of 81.2%), reinforcing their deterministic behavior. These findings statistically validate the OnM intrinsic resilience, supporting its suitability for reliable operation in mission-critical space, nuclear, and medical applications.

Modulation recognition in lowsignal-to-noise ratio (SNR) environments poses significant challenges for existing deep learning models, particularly those based on the Vision Transformer (ViT). Although ViT architectures demonstrate strong performance under quasi-static noise conditions, their robustness tends to decline in more complex, time-varying noise scenarios due to inherent limitations in self-attention and normalization mechanisms. This paper identifies and analyzes the root causes of ViT’s performance degradation under such non-stationary conditions and proposes ResGatNet, a hybrid neural architecture designed to address these limitations. ResGatNet integrates a large-kernel ResNet backbone to enhance local receptive fields, a Gated Convolutional Neural Network (CNN) module to adaptively suppress noise, and a Kolmogorov–Arnold Network (KAN) module to improve global representation with reduced parameter complexity. Experimental results across both quasi-static and time-varying noise settings demonstrate that ResGatNet achieves superior classification accuracy and generalization, significantly outperforming CNN- and ViT-based baselines under low-SNR conditions. These findings confirm that targeted architectural innovations can substantially improve robustness in non-stationary wireless environments.

The rapid advancement of Large Language Models (LLMs) and Large Vision Models (LVMs) has achieved breakthrough performance but has also created significant deployment challenges due to their massive scale, often exceeding 100 billion parameters. Parameter-Efficient Fine-tuning (PEFT) methods, particularly Low-Rank Adaptation (LoRA), address this issue by approximating weight updates as products of low-rank matrices, dramatically reducing the number of trainable parameters while maintaining competitive performance. However, LoRA’s effectiveness is fundamentally constrained by the limited effective rank of its update matrices, limiting its ability to capture the rich, high-dimensional changes required for complex adaptation tasks. We propose GroupLoRA, which overcomes this limitation through a divide-and-conquer strategy that partitions weight matrices into g groups, with independent LoRA modules for each group. GroupLoRA introduces two key innovations: (1) an Inter-Group Bridge module that enables information exchange between groups through learnable interaction matrices, and (2) Learnable Scaling Factors that dynamically adjust each group’s contribution to the final output. This design enables group specialization while maintaining cross-group coordination, maximizing model expressiveness within limited parameter budgets. Extensive experiments on LLMs and VLMs demonstrate that GroupLoRA consistently achieves higher performance with fewer parameters than conventional LoRA. Our approach successfully applies group-wise processing principles to parameter-efficient fine-tuning, offering a practical solution for efficient large-model adaptation across diverse tasks.