Dakota State University

This paper addresses the data poisoning attack in smart agricultural systems using federated unlearning, where the contribution of compromised field devices must be removed from the global model under communication constraints. We propose FedSCAN, a three-phase framework that identifies critical layers based on parameter sensitivity, classifies active neurons within these layers through relative weight change analysis, and performs unlearning via sparse low-rank adaptation to active neurons while freezing base model parameters. FedSCAN limits computational overhead to a compact parameter subset, enabling devices to participate in a communication-efficient manner. Experimental results on four datasets demonstrate that FedSCAN achieves up to $426.3 \times$ communication cost reduction compared to retraining while maintaining remaining accuracy within 3% of the retraining baseline. Compared with state-of-the-art methods, FedAU, FedEraser, and FedOSD, on the agricultural datasets, FedSCAN reduces communication overhead up to $196.3 \times$ while achieving up to 16.4% higher remaining accuracy than FedAU and 13.1% than FedEraser. On the Fruit-360 dataset with 131 fine-grained categories, FedSCAN achieves up to $181.8 \times$ communication reduction while maintaining remaining accuracy within 0.07% of the retraining baseline.

In 2009, W.D. Banks and I.E. Shparlinski studied the average densities of primes p <= x for which the reductions of elliptic curves of small height modulo p satisfy certain algebraic properties, namely cyclicity and divisibility of the number of points by a fixed integer m. In this paper, we refine their results by restricting the primes p under consideration to lie in an arithmetic progression k mod n. Furthermore, for a fixed modulus n, we investigate statistical biases among the different congruence classes k mod n of primes satisfying the aforementioned properties. (c) 2025 Elsevier Inc. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

Land cover information is essential for understanding Earth's surface dynamics and how vegetation, water, soil, climate, and terrain interact. The National Land Cover Database (NLCD) has been the authoritative source for consistent U.S. land cover mapping. To extend NLCD's temporal resolution and reduce production latency, we developed the Land Cover Artificial Mapping System (LCAMS)-a prototype spatiotemporal deep learning framework piloted as the foundation for the new Annual NLCD. LCAMS builds on concepts from legacy NLCD and the U.S. Geological Survey Land Change Monitoring, Assessment, and Projection (LCMAP) initiatives. It employs a loosely coupled two-stage architecture consist ing of independent but functionally interdependent spatial and temporal models. Spatial models extract per-year information from Landsat data, while the temporal models refine the spatial outputs to enforce inter-annual consistency-critical for reliable land change monitoring. LCAMS produces annual 30 m resolution land cover and impervious surface outputs, with region-specific fine-tuning to generalize across diverse landscapes and temporal dynamics. Validation was conducted using an independent dataset of 1925 randomly sampled plots from five U.S. Landsat Analysis Ready Data (ARD) tiles spanning 1985-2021, selected for spatial and temporal variability. This dataset was used consistently to evaluate LCAMS, Legacy NLCD, and LCMAP. Using the NLCD legend, LCAMS achieved 72.1 +/- 1.60% overall agreement, compared to 71.1 +/- 1.7% agreement for Legacy NLCD. Using the LCMAP legend, LCAMS achieved 83.4 +/- 1.22% agreement, compared to 84.6 +/- 1.11% agreement for LCMAP. Overall, LCAMS delivers comparable accuracy while offering higher thematic resolution, longer temporal coverage, and automated production of annual 30 m CONUS land cover.

The current study investigates the factors affecting students' behavioral intentions to utilize "artificial intelligence" (AI) applications in learning by extending the "unified theory of acceptance and use of technology" (UTAUT) model with the inclusion of cognitive flexibility. Data were collected from 315 undergraduate students and analyzed employing a two-step approach involving "structural equation modeling" (SEM) with AMOS. The measurement model was first run to assess and confirm the reliability and validity of the variables, followed by the structural model to test the hypothesized relationships. The results revealed that "performance expectancy (PE), effort expectancy (EE), and social influence" (SI) exerted significant positive effects on students' behavioral intentions to use AI applications, while "facilitating conditions" (FC) exhibited no significant impact. Additionally, "cognitive flexibility" (CF) did not directly influence behavioral intention but exerted a significant indirect effect through its influence on PE, EE, and SI. These findings underscore the significance of students' cognitive adaptability in shaping their perceptions of technological usefulness and ease of use and highlight the value of integrating psychological variables into technology acceptance models in educational contexts. Implications for educational technology implementation and future research directions are discussed.

Accurate segmentation of cardiac structures in 2D echocardiography is essential for diagnosing cardiovascular disease and computing clinical metrics such as chamber volumes and ejection fraction. Conventional U-Net architectures excel at extracting local spatial features but struggle with long-range dependencies inherent in noisy ultrasound images, while pure Transformer-based models capture global context at the expense of fine boundary detail. To address these limitations, we propose KMS-Net, a novel hybrid segmentation architecture that integrates Kolmogorov-Arnold Networks (KANs), a class of learnable, spline-based function approximators that replace fixed activation functions with trainable nonlinear mappings, alongside multi-scale attention mechanisms. Specifically, spline-based KAN layers (grid size of 7) provide highly expressive nonlinear feature modeling. Residual convolutional blocks incorporate Efficient Multi-Scale Attention (EMA) and a KAN-based Atrous Spatial Pyramid Pooling with Squeeze-and-Excitation module (KASPPS) that aggregates rich multi-scale contextual information. Selective State-Space (SS2D) layers enable computationally efficient long-range spatial dependency modeling, and Multi-Scale Attention Gates (MSAG) refine skip connections for precise boundary delineation. Evaluated on the CAMUS dataset (500 patients, two-chamber/four-chamber views at end-diastole/end-systole), KMS-Net achieves mean Dice scores of 0.9417 and 0.9241 (2-CH view) and 0.9518 and 0.9328 (4-CH view), outperforming U-Net variants, TransUNet, Swin-UNet, U-Mamba, and recent KAN-based models. The proposed method further attains state-of-the-art performance on the ACDC cardiac MRI benchmark (mean Dice: 92.65%, HD95: 1.08 mm) and demonstrates strong zero-shot generalization to the EchoNet-Dynamic dataset, confirming its robustness across diverse imaging modalities and clinical settings.