In current online economic dispatch problems of power systems, the cost function of each generation unit typically only includes time-varying generation costs, which does not account for the costs associated with current carbon emission market transactions and high-frequency communications. These factors need to be considered. Additionally, considering the lag in obtaining information about cost coefficients, feedback delay should be considered as a factor for the online optimization process of economic dispatch. This research addresses the challenge of online economic dispatch in the presence of delayed feedback and carbon emission expenses, proposing a decentralized, event-initiated algorithm for online optimization. To address these challenges, this paper proposes a decentralized, event-triggered online optimization algorithm tailored for Agentic AI in smart energy systems operating over an edge–cloud continuum. In each local optimization iteration, each node in the power system can access its local objective function with a delayed time sequence and updates its local decision-making behavior online based on information from multiple neighboring nodes, controlled by an event-triggering mechanism, to minimize the global cumulative generation cost and carbon emission cost. The research results show that, under the assumption that the time-varying balanced undirected communication topology remains connected, the designed online optimization algorithm ensures that the upper bound of static network regret grows sub-linearly, fundamentally related to feedback delays and event-triggering thresholds and scaled as O ( T ) . This work provides a foundational paradigm for Agentic AI in smart energy systems, bridging distributed optimization theory with practical constraints of modern edge-cloud infrastructures.
On-device deployment often relies on lightweight models that require fine-tuning when deployment environments differ from development conditions or when model performance changes over time. Although recent LLM-enabled agents are capable of reasoning about system states and forming adaptive strategies, they often incur substantial computational and budgetary overhead, making them impractical for resource-constrained edge settings. This paper presents Budgeted Agentic AI for Adaptive Lightweight Model Fine-Tuning (BA2), a framework for budget-constrained lightweight model adaptation that combines bounded optimization steps with auxiliary evaluation, rollback, and compression actions. BA2 enables agents to dynamically adjust adaptation strategies according to real-time budget conditions while maximizing system performance. BA2 explicitly accounts for constrained resources, including limited fine-tuning steps, tool invocation quotas, and token consumption. An Engineer-Manager architecture is introduced, where an LLM-based Engineer generates parameterized candidate operations from system logs, and a lightweight contextual-bandit Manager selects actions conditioned on both environmental states and remaining budgets. Experimental results demonstrate that BA2 achieves superior cost-performance trade-offs compared with static heuristics and inference-time search agents under tight budget constraints, while remaining competitive with strong baselines in higher-budget regimes.
Open Multi-Agent Systems (OMAS) are societies composed of autonomous agents that operate with diverse or aligned objectives. To enhance cooperation among these independently functioning agents, researchers have adapted the concept of social norms from human societies as regulatory mechanisms. However, social norms often exist as unwritten, unspoken rules that can only be acquired through observation and inference, a significant challenge for autonomous agents. This paper proposes a novel Dual-task Transformer architecture that enables agents to learn social norms by observing interactions in their environment. Our approach leverages the transformer's capability to capture long-range dependencies between observed actions, performing two simultaneous tasks: contextually classifying action sequences to predict punishments and rewards, and generating structured representations of the inferred social norms. The model is trained on a modified version of the Moral-Stories dataset, transformed to represent sequences of agent actions and their normative outcomes. We integrate this inference capability with rule-based systems to assist agents in their decision-making processes. Experiments conducted within the MESA agent-based modeling framework demonstrate that agents equipped with our dual-task inference model rapidly internalize learned social norms and effectively regulate their behavior, showing a significant reduction in norm-violating actions and an increase in the adoption of norm-compliant behavior. Our approach offers a cognitively inspired mechanism for norm learning in OMAS that is both extensible through fine-tuning and capable of capturing complex dependencies in observed social interactions.
Alzheimer's disease is a neurodegenerative disease that affects brain tissues, causing memory loss. The most common and primary symptoms of Alzheimer's are the complexity of remembering more information, behavioral changes, and a deep confusion about events, locations, and time. The detection of Alzheimer's disease is desirable for therapy and patient care. Still, it is difficult to handle heterogeneous and high-dimensional biomedical data for attaining a reliable prediction. To solve these complexities, this paper devises the Multi-Axis-Multi-Head Capsule Network (MaxMHCapsuleNet)-based detection and classification of Alzheimer's disease. Initially, image sharpening is performed using a Kriging-Weighted Laplacian kernel, and the Spatial-Channel Mamba-UNet (SCM-UNet) segments the affected region. Furthermore, the techniques, such as 90 degrees rotation and random masking, augment the images. In the feature extraction process, GoogLeNet and Visual Geometry Group Network (VGGNet) features are extracted. The Alzheimer's disease detection is done by MaxMHCapsuleNet, which detects the normal (non-demented) and abnormal classes. If the detected result is abnormal, multiclass classification is done by MaxMHCapsuleNet. In the classification process, the abnormal classes are classified into mild, moderate, and very mild demented stages. Moreover, the devised MaxMHCapsuleNet attains the accuracy, precision, and recall of 91.709%, 90.388%, and 92.036%.
In this research work, a novel multimodal data-based recommendation system is introduced to ultimately enhance student's talents and broaden knowledge without any interference. Initially, the required multimodal data, such as video, images, and text data are gathered from standard sources and it is given to the feature extraction phase. Here, the Bidirectional Encoder Representations from Transformers (BERT), Visual Geometry Group 16-based Autoencoder (VGG16-AE) and 3D VGG16-AE mechanisms are utilized for accurately recognizing and differentiating the significant feature sets such as, first set of features (f1), second set of features (f2) and third set of features (f3) from the input data. Then, the extracted three set of features are fed into the learning recommendation phase here, the Multi-scale Feature Fusion-based Residual Bidirectional Gated recurrent unit with Sparse Attention (MFF-RBG-SA) mechanism is designed to efficiently analyze and generate better solution in an earlier stage. Moreover, the overall performance of the designed approach is estimated with various performance measures and compared with conventional methods to guarantee its reliable effectiveness in the educational sector. In the validation phase, the designed approach has attained higher accuracy value of 95.35%, 95.28%, 95.38% and 95.33% in terms of Linear, Tanh, softmax and sigmoid activation functions, which demonstrate its superior learning recommendation performance than other conventional methods.
