The widespread adoption of electric vehicles (EVs) is constrained by charg- ing infrastructure limitations, range anxiety, and temporal-spatial mismatches between energy demand and supply. Vehicle-to-vehicle (V2V) energy sharing presents a promising solution by leveraging EVs as distributed energy storage units, yet existing approaches lack predictive capabilities for managing net- work congestion and optimizing resource allocation at scale. This paper intro- duces a novel three-layer graph-based architecture for intelligent V2V energy sharing that integrates dynamic graph construction, Graph Attention Network (GAT)-based congestion prediction, and multi-objective routing optimization. The system models EVs and charging stations as nodes in a temporal bipartite graph, employs deep learning to forecast network bottlenecks, and implements an enhanced matching algorithm that simultaneously optimizes response time, energy efficiency, load distribution, and cost-effectiveness. Evaluated on 10,000 real-world DC fast-charging sessions transformed into V2V scenarios, the pro- posed framework achieves an 85.0% fulfillment rate with an average response time of 29.6 minutes, representing a 9.0% improvement in service success and a 30.0% reduction in wait times compared to greedy baselines. The GAT predictor demonstrates 87.4% accuracy in anticipating congestion, enabling proactive re- source allocation that reduces congestion-induced failures by 51.7%. This work extends existing urgent charge-sharing models by introducing predictive graph learning for large-scale V2V networks, offering a scalable solution for sustainable urban mobility infrastructure.

The explosive growth of artificial intelligence (AI) and machine learning (ML) workloads has driven the development of specialized hardware accelerators, including GPUs, TPUs, NPUs, and custom ASICs. Central to their performance and efficiency are memory and interconnect architectures specifically high bandwidth memory (HBM), on-chip networks (NoC), and 3D stacked compute. This paper presents a comprehensive survey of AI/ML hardware accelerators with a memory/interconnect-centric focus, analyzing technical evolution, state-of-the-art implementations, and performance-areapower trade-offs. Through a detailed case study of the Cerebras Wafer Scale Engine, we demonstrate how radical architectural approaches address scaling challenges. As our core contribution, we propose a novel three-dimensional taxonomy that classifies accelerators based on memory architecture, interconnect topology, and integration strategy. This taxonomy provides researchers and practitioners with a systematic framework for evaluating architectural choices, comparing designs, and guiding future innovations in emerging areas like chiplets and in-memory computing.

Accelerating electric vehicle (EV) deployment is constrained by uneven charging infrastructure, driver range anxiety, and persistent mismatches between where energy is demanded and where it is available. Vehicle-to-vehicle (V2V) energy sharing alleviates these pressures by treating every EV as a mobile storage asset, yet current coordination schemes react to congestion rather than anticipating it, and rarely optimize multiple service objectives simultaneously. This paper proposes a three-layer temporal graph architecture for intelligent V2V energy sharing. The first layer continuously constructs a bipartite graph over EVs and charging stations, encoding proximity, state of charge, queue occupancy, and dynamic pricing on each edge. The second layer applies a Graph Attention Network (GAT) trained on historical session data to predict congestion probabilities on candidate provider edges before service degradation occurs. The third layer resolves each energy request through a multi-criteria matching function that jointly minimizes response latency, maximizes energy efficiency, distributes provider load equitably, and controls service cost. Evaluated on 10,000 realworld DC fast-charging sessions from the DESL-EPFL dataset, which are transformed into realistic V2V scenarios through a principled simulation pipeline, the proposed framework achieves an 85.0% fulfillment rate with a mean response time of 29.6 minutes. These figures represent a 9.0% gain in service success and a 30.0% reduction in waiting time relative to the strongest greedy baseline. The GAT predictor reaches 87.4% overall accuracy and enables proactive resource allocation that cuts congestion driven failures by 51.7%. Scalability experiments confirm sub-250 ms processing for networks of up to 5,000 nodes, supporting city-scale real-time deployment.

Chest X-ray (CXR) imaging remains a widely used diagnostic modality for the detection of thoracic diseases, including pneumonia, tuberculosis, pleural effusion, lung cancer and cardiomegaly. Despite the extensive use of CXR examinations in clinical practice, the development of robust multi-label artificial intelligence (AI) systems remains constrained by limitations in dataset availability, annotation consistency and reproducible labeling methodologies. Existing public datasets frequently rely on heterogeneous labeling protocols, limited pathology coverage or annotations that inadequately account for linguistic negation and uncertainty within radiology reports. In this paper, we introduce MultiCaRe-Thorax, a curated thoracic benchmark dataset derived from the publicly available MultiCaRe repository, with a reproducible negation-aware natural language processing (NLP) framework for automated generation of multi-label thoracic disease annotations. Unlike existing CXR benchmarks that are distributed in their pre-curated form, MultiCaRe-Thorax provides a fully reproducible pipeline from raw case reports to curated dataset, establishing a new resource that is both scalable and transparent. Our pipeline performs thoracic case identification, image–report matching, duplicate removal, data cleaning, disease-specific label extraction and quality-control verification to produce a curated cohort comprising 5252 chest radiographs from 3415 unique patients. Sixteen clinically relevant thoracic pathologies were automatically annotated using a dictionary- and pattern-based NLP framework incorporating explicit negation detection and n-gram mining, achieving 95% accuracy against 500 manually reviewed reports (Cohen’s $\kappa =0.89$). Sixteen clinically relevant thoracic pathologies were automatically annotated using a dictionary- and pattern-based NLP framework incorporating explicit negation detection and n-gram mining, achieving 95% agreement against 500 manually reviewed reports. Five ImageNet-pretrained convolutional neural network (CNN) architectures (ResNet50, DenseNet121, EfficientNetB0, EfficientNetB7 and InceptionV3) with two hybrid ensemble configurations were evaluated under a patient-level data partitioning protocol. Among individual architectures, DenseNet121 attains the best AUC on tuberculosis (0.757) and pleural effusion (0.767) with roughly one-third the parameters of ResNet50, while ResNet50 attains the higher mean AUC across all sixteen pathologies (0.688 vs. 0.682) and on interstitial lung disease (0.769) and lung mass/cancer (0.777); we report this as an accuracy–efficiency trade-off rather than a single best architecture. The DenseNet121-InceptionV3 ensemble achieves the highest reported AUC of 0.821 for lung mass/cancer detection. We report 95% confidence intervals for the evaluated ensemble and remaining backbones and note substantial estimation uncertainty for the rarest classes (e.g., congenital heart disease, $n=8$ test cases). Grad-CAM visualizations were also employed to provide clinically interpretable explanations of model predictions.