graph neural networks

Quantum-inspired scenario-adaptive differential privacy with hybrid attention–residual deep learning for utility-aware urban location analytics

A Quantum-Inspired Scenario-Adaptive Differential Privacy framework that combines k-nearest-neighbor (kNN)-based adaptive privacy budgeting with structured quantum-inspired noise generation and a Hybrid Attention–Residual Deep Neural Network for mechanism selection demonstrates that structured quantum-inspired perturbations can substantially improve privacy–utility trade-offs while maintaining robustness across independent real-world mobility datasets.

Wafer-scale SOT-MRAM for analog crossbar array applications

Analog crossbar arrays consisting of emerging memory devices can alleviate the computational strain required by vector matrix multiplications for neural network applications. The ability to produce spin orbit torque-magnetic random-access memory (SOT-MRAM) at wafer-scale positions SOT-MRAM as a strong memory candidate. In this work, we fabricate and measure 300 mm-compatible SOT-MRAM with 150% tunnel magnetoresistance (TMR) ratio, fast (2 ns) and low voltage (<1 V) operation, low energy dissipation (2 pJ), low write noise (0.1%), and low device-to-device variation of 10%. SOT-MRAM characteristics were shown to be effective for inference on calibrated models. The bi-stable anisotropy and stochastic switching of SOT-MRAM was leveraged for binary neural network training, able to reach ideal accuracy for a single device. Lastly, the devices were evaluated on probabilistic graph modeling and the interplay of TMR ratio and probability distribution is analyzed. Through these results, SOT-MRAM is shown to be a uniquely effective candidate for implementation of crossbar accelerators in memory- and energy-limited applications.

QUBO-Compatible Active Learning for Inverse Design of High-Entropy Alloys

A QUBO-compatible active-learning framework for inverse design of high-entropy alloys using a pretrained graph-neural-network predictor as a fixed property oracle and the learned quadratic surrogate can be exported directly as a QUBO.

A multimodal EEG-based psychological risk representation method using spatiotemporal graph attention and adaptive gated fusion

A cross-modal temporal representation method that integrates spatiotemporal graph attention with an adaptive gating mechanism that significantly improves the accuracy and robustness of risk prediction on the AMIGOS dataset, effectively overcoming the subjective interference of questionnaire-based assessments.

Modeling of a packed pebble cell on coupled conduction–radiation–convection heat transfer: A physics-embedded graph neural network

Physics-Aware Recurrent and Graphical Learning for Robust Distribution System State Estimation Across Unseen Networks

Robust distribution system state estimation (DSSE) is increasingly vital in modern, complex networks. However, conventional optimization-based DSSE approaches often face challenges with real-time complexity. Consequently, recent advancements have introduced learning-based, physics-aware DSSE methods that leverage network structural information to enhance model effectiveness. Among these methods, graph neural networks (GNNs) have gained prominence as an effective DSSE solution. However, GNN-based models encounter significant challenges in practical DSSE applications, due to their reliance on aggregating and propagating features only from adjacent nodes. To address this limitation, this paper presents a novel physics-aware DSSE framework. It integrates a structure-aware recurrent neural network and GNN with convolutional autoregressive moving average layers for improved feature aggregation and propagation. This model significantly enhances robustness and transferability across diverse unseen network topologies. The scalability of the proposed framework is demonstrated through extensive testing on real distribution systems. The proposed method demonstrates superior performance compared to existing methods by an average of 0.625% MAPE and 0.0093 rad. MAE in normal scenario under unseen networks. Additional ablation and measurement availability studies further verify the robustness and effectiveness of the proposed framework. The framework offers a pragmatic solution for DSSE, ensuring precise state estimation even in challenging environments.

Joint dynamic pruning method for graph neural networks under traffic prediction scenarios

Graph Theory, Graph Neural Networks, and Reinforcement Learning in marine ecology: A systematic review of methodologies, theoretical rigour, and future directions

Orbital-geometric atomic line graph neural network-VoltaDelta framework for assisted screening of zinc-ion battery cathode materials

MF-HGNN: A multi-view fusion heterogeneous graph neural network for psychiatric disorder diagnosis

A dynamic spatio-temporal graph neural network adaptive to operating conditions: Learning evolutionary correlations in multi-sensor data for aero-engine

Topology-aware graph neural network for fusing sparse sensor data with finite element predictions in structural random vibration assessment