Anshul Verma

Multi-access edge computing (MEC) enhances the performance of resource-constrained devices by offloading computation to nearby edge servers, but simultaneously achieving energy efficiency and low latency under dynamic conditions remains challenging. This paper proposes Lyapunov Optimization with Virtual Queues (LOVQ), a task offloading framework based on the Lyapunov Drift Plus Penalty (LDPP) method that jointly optimizes energy consumption and delay while ensuring long-term system stability. The LOVQ framework integrates three complementary algorithms for adaptive task offloading, decay-aware updates to virtual energy queues, and battery-aware task processing under energy-constrained conditions. Simulations with 20 edge devices and 2000 tasks show that LOVQ reduces average energy consumption to 30.0 J, achieving a 33% reduction compared with local execution and a 6% reduction compared with classical Lyapunov-based offloading. LOVQ also achieves the lowest average delay of 0.92 s, reducing delay by 31% relative to local execution and exceeding random and edge-only offloading strategies. These results demonstrate that LOVQ significantly improves the energy–delay trade-off while maintaining system stability, making it a scalable solution for dynamic MEC environments.

The growing deployment of applications in real-world edge environments is driving an increasing demand for large language model (LLM) training in distributed systems. This demand creates significant challenges for resource management, particularly in resource-constrained edge cloud environments. Federated learning (FL) enables privacy-preserving LLM fine-tuning across multiple clients, but it also introduces dynamic computational and communication loads that can strain system resources. To address these challenges, we propose an integrated framework that combines LLM fine-tuning using federated adaptive local low-rank adaptation (FedALoRA) with forecasting-driven autoscaling policies. Differential privacy (DP) is incorporated to ensure secure aggregation, balancing privacy guarantees with system performance. Using the WikiText-2 dataset for FL-based LLM training and the Bitbrains dataset for CPU usage forecasting, we evaluate deep learning models (LSTM, GRU, BiLSTM, CNN, CNN-LSTM, Transformer). Results demonstrate that FedALoRA enables fast federated convergence, reducing perplexity from 199.3 to 85.2 without DP and from 211.6 to 92.2 with DP, incurring only a 7–13% privacy overhead. Among forecasting models, CNN-LSTM achieves the lowest MSE (0.25) and consistently requires the fewest pods (as low as 11 replicas), highlighting its superior accuracy and resource efficiency for proactive autoscaling. Comparative analysis demonstrates that proactive autoscaling consistently outperforms reactive autoscaling, with non-DP settings achieving slightly faster adaptation and DP settings offering stronger privacy guarantees. Overall, the proposed framework balances performance, adaptability, and privacy, making it well-suited for federated-LLM Autoscaling in dynamic edge environments.

Identification of skin lesions remains a challenging task because of low image contrast, large variations in disease manifestations, and strong visual similarities among dermatological conditions. Deep transfer learning (TL)-integrated artificial intelligence (AI) systems offer an effective solution for rapid and accurate computer-aided diagnosis of multiple infectious skin diseases. In this study, a TL-based framework is proposed as an automated human monkeypox (MP) skin lesion detection (HMSLD) for monkeypox (MP), chickenpox (CP), measles (MS), and normal skin lesion classification. A publicly available dataset comprising 654 RGB clinical images. The HMSLD performed image preprocessing, data augmentation, class imbalance handling through inverse class weighting, and comparative analysis of pre-trained TL models, namely Xception, ResNet152V2, NASNetLarge, EfficientNetB0, EfficientNetB7, MobileNetV2, and MobileNetV3. MobileNetV3 performed best among the models. Their overall performance with an accuracy = 91%, precision = 91%, recall = 89%, F1-score = 90%, and an AUROC of 0.99, indicating excellent discriminative capability. MobileNetV3 achieved 96% precision for measles and 100% recall for normal skin, while EfficientNetB0 performed the highest precision at 96% and an F1-score of 93%. MobileNetV3 provides the most balanced trade-off among the evaluated models between classification accuracy, robustness, and computational efficiency. These evaluated performances indicate the potential of the HMSLD framework. It may be a reliable and cost-effective computer-aided diagnostic tool for early MP screening and other clinically similar skin diseases.