graph neural networks

Learning on Edges: A Narrative Review of Graph Neural Networks from Recursive Networks to Geometric Deep Learning

Graph neural networks---learning over relational, irregular structure by passing messages between nodes---generalized deep learning's grids to the graph: molecules, social networks, knowledge bases, and the web. This article presents a narrative review of that arc's canonical line: Sperduti and Starita's 1997 structure classification, Gori, Monfardini, and Scarselli's 2005 graph-domain learning, Scarselli and colleagues' 2009 GNN model, Bruna and colleagues' 2014 spectral networks, Defferrard and colleagues' 2016 localized filtering, Kipf and Welling's 2017 graph convolutions, Gilmer and colleagues' 2017 message passing, Hamilton, Ying, and Leskovec's 2017 GraphSAGE, Velickovic and colleagues' 2018 attention, Ying and colleagues' 2019 GNNExplainer, Wu and colleagues' 2021 comprehensive survey, and Bronstein and colleagues' 2021 geometric deep learning. The synthesis is organized around three themes: recursion, in which state propagation over nodes founded learning on graphs; convolution, in which spectral theory and message passing gave the graph a deep architecture; and geometry, in which attention, pooling, explainability, and symmetry made the network general. It is concluded that the GNN is deep learning's relational settlement---convolution's invariance learned from graph geometry rather than grid regularity---and that its message-passing abstraction is one of machine learning's cleanest unifications.

Accelerating Materials Discovery: A Review of Machine Learning in X‐Ray Absorption Spectroscopy

X‐ray absorption spectroscopy (XAS) is a critical technique for probing the local structural and electronic properties of materials. Advanced synchrotron radiation facilities generate complex, high‐dimensional spectra, which pose significant challenges for traditional analysis methods while simultaneously offering unprecedented opportunities for machine learning (ML). This review systematically elaborates how ML models are driving the transformation of XAS data analysis. We not only cover supervised and unsupervised learning methods for spectral classification and clustering but also delve into cutting‐edge deep learning architectures. These include graph neural networks for precise “structure‐to‐spectra” mapping and diffusion models for generative tasks and structure prediction. We provide a comprehensive overview of the key challenges in data‐driven XAS, including the feature engineering of spectra and structures, strategies for solving the “spectra‐to‐structure” Inverse Problem, and Sim2Real methods for bridging the “Domain Gap” between simulated and experimental data. Furthermore, we emphasize the importance of model eXplainable artificial intelligence and uncertainty quantification for building trust in scientific research. Finally, this review looks ahead to the future of the field, driven by materials informatics and autonomous experiments. Active Learning techniques, represented by Bayesian optimization, are pioneering “self‐driving” smart XAS experiments, which will greatly accelerate the discovery and design of new materials.

pinn-serving: Derivative-Aware Serving of Physics-Informed Neural Networks

A serving harness and benchmark suite for physics-informed neural networks (PINNs). Introduces derivative-graph baking, which differentiates a tanh MLP symbolically offline and emits the closed-form derivative recursion as a forward-only program, making derivative-valued queries compatible with standard inference backends and quantisation. Includes evidence that output-space accuracy monitoring cannot detect physics degradation at serving time, and a matched comparison against a Crank-Nicolson solver.

THE EVOLUTIONARY E-SPORTS CIVILIZATION MODEL (EECM): A Grand Unified Interdisciplinary Theory of Digital-Cognitive Civilization, Cognitive Capital, Artificial Intelligence Augmentation, Neuroeconomic Transformation, Virtual Ontology, and Civilizational Evolution

This paper proposes the Evolutionary E-Sports Civilization Model (EECM), a grand interdisciplinary theoretical framework designed to explain the transformation of gaming, cognition, digital interaction, artificial intelligence, and virtual systems into a new civilizational paradigm. The theory argues that E-sports represents far more than organized competitive gaming; rather, it constitutes a symbolic and structural manifestation of the ongoing transition from industrial civilization toward a digital-cognitive civilization.The framework integrates mathematics, complexity science, neuroscience, psychology, philosophy, sociology, economics, political science, cybernetics, systems theory, information theory, artificial intelligence, and digital ontology into a unified explanatory architecture. It proposes that modern civilization increasingly transforms psychologically meaningful activities into measurable, monetizable, and algorithmically optimized systems of production and social organization.Within this framework, cognition itself becomes a form of capital. Human attention, strategic reasoning, adaptive intelligence, reflex optimization, emotional regulation, collaborative cognition, and digital interaction evolve into economically productive assets. The theory therefore introduces the concept of Cognitive Capital, a post-industrial expansion of classical economic production factors.The EECM framework further argues that artificial intelligence acts as a civilizational amplifier, accelerating the transformation of digital environments into self-organizing socio-economic ecosystems governed increasingly through algorithmic optimization. E-sports is analyzed as an emergent prototype of future virtual civilization structures, where entertainment, labor, economics, governance, identity, and AI converge into unified digital systems.Mathematically, the paper formalizes these transformations using nonlinear dynamical systems, complexity theory, graph theory, information entropy, game theory, network theory, chaos theory, and probabilistic scaling functions. Philosophically, the theory synthesizes concepts from Aristotle, Plato, Nietzsche, Heidegger, Marx, Foucault, Bourdieu, Sartre, Kant, and contemporary philosophy of technology. Psychologically and neuroscientifically, the model incorporates flow theory, predictive processing, dopamine reward systems, cognitive load theory, neural plasticity, and human-machine symbiosis.The framework also explores geopolitical implications, platform sovereignty, AI governance, digital identity formation, algorithmic social structures, metaverse civilization, and the future political economy of virtual systems. Ultimately, the theory proposes that E-sports is not merely a recreational phenomenon but an early-stage manifestation of a broader civilizational transition in which cognition, attention, and digital interaction become the dominant organizational principles of society.

pinn-serving: Derivative-Aware Serving of Physics-Informed Neural Networks

A serving harness and benchmark suite for physics-informed neural networks (PINNs). Introduces derivative-graph baking, which differentiates a tanh MLP symbolically offline and emits the closed-form derivative recursion as a forward-only program, making derivative-valued queries compatible with standard inference backends and quantisation. Includes evidence that output-space accuracy monitoring cannot detect physics degradation at serving time, and a matched comparison against a Crank-Nicolson solver.

Graph Neural Networks for Spatio-Temporal Intrusion Detection

Cross-Database Performance of Speech-Based Graph Neural Networks for Parkinson’s Disease Detection

Graph Neural Networks for Spatio-Temporal Intrusion Detection

Smartpath-Rpl: A Deep Reinforcement Learning-Based Optimized Routing Protocol For Mobile Low-Power And Lossy Networks

QGPINNs: A Physics-Informed Neural Network Framework for Nonlocal Differential Equations on Quantum Graphs

We propose QGPINNs, a physics-informed neural network framework developed in PyTorch for the numerical solution of nonlocal differential equations on quantum graphs. The framework is designed as a general computational implementation in which the solution on each edge of the graph is approximated by a neural network, while a unified graph-based loss function enforces the governing equations together with initial, boundary, and vertex transmission conditions. In particular, the formulation incorporates standard continuity and Kirchhoff-Neumann vertex conditions and Dirichlet boundary conditions into the learning process to couple the local edge-wise neural approximations into a global solution on the graph. The framework is developed for two representative classes of nonlinear models: multi-order fractional elliptic problems and time-fractional evolution equations on quantum graphs. To improve accuracy and training stability, QGPINNs integrates several graph-adapted learning strategies, including soft and hard constraint enforcement, dynamic loss balancing, Fourier feature embeddings, and a learnable singularity-capturing feature for weakly singular solutions arising in the considered problems. The framework also extends naturally to inverse problems, including the identification of the orders of fractional operators and physical parameters from noisy observational data. We validate the accuracy, computational efficiency, and physical consistency of the proposed framework through numerical experiments on benchmark graph structures and real-world networks, including the IEEE 14-bus system and an open-channel agricultural drainage network.

Node-wise Feature Encoding for Neural Performance Prediction

This work introduces FeatureFormer, a neural performance predictor that incorporates explicit node-wise encodings of FLOPs, parameter counts, and memory proxies within a gated graph attention architecture and presents NNEQ, a new large-scale energy consumption dataset that enables unified evaluation of latency and energy prediction.

Self-Explainable Multi-Label Graph Neural Network for Correlated Evidence Attribution

Experiments and comparisons show that SEMGNN achieves competitive or improved predictive performance while providing more faithful and compact label-conditioned explanations, compared to post-hoc methods.