The core problem the researchers tackled is semantic, not computational. Grid monitoring systems rely on event rule bases written in a domain-customized grammar that mixes custom symbols with natural language. Feature variables are denoted by the Greek letter Omega, equipment parameter templates by dollar signs, state transitions by arrows, and redundancy checks by bracketed expressions specifying that at least x out of n features must fire within a time window of y seconds. A typical rule such as (Omega1|Omega2|Omega3)&Omega4&Omega5 means an event triggers when any one of three features occurs together with two others. Traditional string matching simply cannot parse this reliably, because the rules combine symbolic ambiguity, deeply nested AND/OR logic, and heavy reliance on expert power system knowledge such as protection operation logic and equipment topology.

To bridge this gap, the team built a three-layer standardized preprocessing pipeline. The symbol mapping layer converts domain symbols into natural language, using a dictionary of 75 attribute quantities and 16 operators that covers every symbolic expression in the rule base. The structured conversion layer decomposes nested logical expressions into explicit if-then-else conditional trees, with indentation marking nesting depth. The semantic enhancement layer then injects context such as equipment type, voltage level, and grid topology, turning a generic protection action into a fully specified statement like a 220 kV line distance protection Zone I action at a named substation. Combined with instruction tuning and retrieval-augmented generation, this pipeline cut the large language model’s hallucination rate from 15.8 percent to 3.1 percent.

The language model at the heart of the system is Qwen3.5-32B, fine-tuned using Low-Rank Adaptation with rank 16 applied to query and value projection matrices. Training drew on four purpose-built datasets totaling more than 9,000 samples, covering symbol understanding, operation logic, redundancy rules, and notification rules, constructed through a semi-automatic pipeline in which experts annotated 20 percent of samples manually and verified the rest, achieving an inter-annotator agreement of 94.6 percent by Cohen’s kappa. A triple verification mechanism checks grammatical correctness, semantic consistency against the original rule intent using a cross-encoder with a 0.85 similarity threshold, and practical performance through historical alarm playback. At inference time, a few-shot strategy dynamically retrieves three to five similar parsed examples, weighted 40 percent on structural Jaccard similarity and 60 percent on Sentence-BERT semantic similarity.

Retrieval-augmented generation supplies the external knowledge the model’s parameters lack. The researchers constructed an event rule knowledge base containing more than 900 rule templates, over 12,000 historical parsing cases, and a domain dictionary of more than 3,500 terms, indexed with both inverted indices for BM25 keyword retrieval and dense FAISS vector indices for semantic search. Documents are chunked into 512-token segments with 64-token overlap. The hybrid retrieval adds an average of only 45 milliseconds per query, comfortably within the 200-millisecond real-time budget for event generation. For rare rule patterns, the RAG-enhanced method reached an F1-score of 89.6 percent, an improvement of 18.3 percentage points over instruction tuning alone.

Alongside the language model, the team converted the event rule base into a knowledge graph with 42 entity types and 69 relationship types spanning equipment, signals, rules, events, and preplans. A single substation graph holds about 5,000 nodes, and a regional grid graph reaches tens of thousands. The system embeds this graph using GraphSAGE, an inductive graph neural network that generates 128-dimensional node vectors by sampling and aggregating features from local neighborhoods, 25 neighbors in the first layer and 10 in the second. Unlike transductive methods, GraphSAGE can compute embeddings for newly added nodes on the fly without retraining, which matters when grid equipment changes constantly. Monthly updates involving 20 to 50 modified rules complete within 15 to 30 minutes using a blue-green deployment strategy that swaps new subgraph versions into the live store with zero downtime.

Recognition proceeds in two layers. The first performs precise matching: when an incoming alarm combination fully satisfies a rule definition, the event is generated with 100 percent confidence. The second activates a dual-drive fusion model when precise matching fails due to signal loss or temporal disorder. Its knowledge-driven branch feeds GraphSAGE rule embeddings through a fully connected layer to produce a 64-dimensional knowledge vector, while its data-driven branch uses a two-layer bidirectional LSTM to encode the alarm sequence into a 128-dimensional temporal vector. A four-head attention fusion layer adaptively weights the two branches, balancing knowledge priors against temporal patterns. In case studies, this architecture recognized a 500 kV line trip event in 22 milliseconds, and recovered a 220 kV transformer internal fault with two of four expected signals missing, cross-checking SCADA status registers to upgrade confidence from 0.89 to 0.97 in 156 milliseconds.

Orchestrating everything is a hierarchical multi-agent architecture with four specialized agents. A Perception Agent on edge nodes parses alarm streams at 10,000 messages per second via MQTT and Kafka, filters duplicates within a 500-millisecond window, and outputs structured alarm vectors. An Analysis Agent on the central server runs rule matching, invokes the fusion model, and handles complex semantic parsing through the RAG pipeline. A Decision Agent cross-validates outputs from multiple analysis instances, resolves conflicts through a three-level mechanism, and routes low-confidence events to human operators. A Scheduling Agent ranks events by voltage level, equipment criticality, and cascading risk, then dispatches notifications. Agents communicate through a unified JSON protocol with Kafka pipelines, a shared Redis blackboard, and broadcast event buses, publishing heartbeats every two seconds so that failures trigger automatic failover within six seconds.

Conflicts between agents are resolved automatically in three stages. When candidate results differ in confidence by more than 0.15, the higher-confidence result wins, resolving about 72 percent of conflicts. When scores are close, the Decision Agent compares evidence chains using a weighted score combining matched rule conditions, temporal coherence, and topological consistency, resolving another 23 percent. The remaining 5 percent go to LLM arbitration with structured prompts, and any judgment below 0.8 confidence is escalated to humans. Crucially, the system includes a SEEN algorithm, short for Semantic Event Exception Notifier, that fuses structural anomaly detection with semantic deviation analysis to flag previously undefined event patterns. In six months of operation it detected 23 new event patterns; in one case study it flagged an unusual combination of capacitor bank overvoltage, unbalance current, and bus voltage abnormality that engineers confirmed as a new failure mode, with a new rule added to the base within 48 hours.

The authors are candid about limitations. GraphSAGE embeddings degrade between monthly retraining cycles as grid topology drifts, motivating continual graph learning. The system was trained on a single regional grid, so transferring it elsewhere requires substantial re-annotation. It recognizes individual events but cannot yet reason about causal chains that turn a line trip into transformer overload and load shedding. And the LLM’s decision process remains a black box in a safety-critical setting, which the team hopes to address with neuro-symbolic approaches that use the rule knowledge graph as a formal constraint layer. Still, the ablation studies confirm that every module earns its place: removing either the knowledge graph or BiLSTM branch costs more than 6 percent F1, the preprocessing pipeline contributes 5.6 percent, and attention fusion beats naive concatenation by 4.1 percent. For an industry drowning in alarm noise, the message is clear: grids can now be watched by machines that understand what they are reading.

Subject of Research: Artificial Intelligence

Article Title: Multi-agent collaborative semantic recognition and real-time event generation for power grid alarm monitoring

Article References: Yunhui, Z., Huijia, Y., Binling, X., Yongkui, W., & Xiaokai, L. (2026). Multi-agent collaborative semantic recognition and real-time event generation for power grid alarm monitoring. Discover Artificial Intelligence, 6(1), Article 1538. https://doi.org/10.1007/s44163-026-02516-1

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02516-1

Keywords: power grid, large language models, knowledge graph, multi-agent systems, alarm monitoring, GraphSAGE, Multi-agent, collaborative, semantic, recognition, real-time, event

The core insight behind the work is architectural. Convolutional neural networks such as ResNet and DenseNet are superb at detecting local patterns, textures and edges within small patches of an image, but they struggle to model the intricate topological structures and long-range spatial correlations that often distinguish a diseased retina from a healthy one. Diabetic retinopathy is precisely such a disease: microaneurysms, hemorrhages, exudates and abnormal vessel growth do not appear in isolation, and their distribution across the fundus carries diagnostic meaning. Graph neural networks, by contrast, operate on data represented as nodes connected by edges, and they learn by passing messages between neighboring nodes, iteratively aggregating information so that each node’s representation reflects both its own features and the context of its surroundings. That makes them naturally suited to capturing both local lesion detail and global structural organization in a way that fixed-grid convolutions cannot.

The researchers’ pipeline begins conventionally enough. They used the publicly available APTOS 2019 Blindness Detection dataset from Kaggle, consisting of fundus photographs resized to 224 by 224 pixels and Gaussian-filtered to reduce noise. The five original severity grades, ranging from no retinopathy to proliferative disease, were collapsed into a binary problem: any form of diabetic retinopathy versus none. Each image was then passed through a pre-trained ResNet50 backbone, with ImageNet weights providing a rich foundation of hierarchical visual features. Crucially, rather than collapsing the network’s output into a single global vector, the authors tapped an intermediate feature map of dimensions 512 by 28 by 28. Each of the 784 spatial positions in that map became a node in a graph, carrying a 512-dimensional feature vector that preserved the spatial structure of the underlying retina.

Edges between nodes were then formed using a k-nearest-neighbor strategy in feature space. For every node, the team computed Euclidean distances to all other nodes and connected it to its eight most similar counterparts, producing a graph in which proximity reflects feature similarity rather than mere pixel adjacency. The resulting graphs were split into training, validation and test sets of 70, 15 and 15 percent respectively, using stratified sampling to preserve class balance, and trained in mini-batches of 32 graphs on an NVIDIA Tesla P100 GPU. Two graph neural network architectures were trained and compared: GraphSAGE, which aggregates neighborhood information and concatenates it with the node’s own representation, and the Graph Isomorphism Network, or GIN, which applies a multilayer perceptron to the sum of a node’s features and those of its neighbors, giving it expressive power comparable to the Weisfeiler-Lehman graph isomorphism test.

Both models were trained for up to 100 epochs with the Adam optimizer at a learning rate of 0.001, using negative log likelihood loss and saving the weights that achieved the best validation accuracy. The results were striking. GIN reached an overall accuracy of 96.86 percent, with a recall of 98.66 percent and an F1 score of 96.96 percent, indicating an exceptional ability to catch true cases of the disease. GraphSAGE was close behind at 96.45 percent accuracy but led on precision at 95.53 percent, specificity at 95.29 percent and, most notably, an ROC-AUC of 0.992319 against GIN’s 0.986976, a near-perfect measure of the model’s confidence in separating the two classes. Both architectures comfortably outperformed state-of-the-art CNN baselines such as SE-ResNeXt50, EfficientNet and VGG-16, as well as a prior graph adversarial transfer learning approach that had reported 94.3 percent accuracy on the same dataset.

Perhaps the most scientifically interesting evidence came from the visualizations. When the researchers plotted the distributions of edge weights, the Euclidean distances between connected node features, they found that graphs from diseased retinas showed a sharp peak at low distance values, meaning pathological regions produce tightly clustered, locally coherent feature spaces. Healthy retinas, by contrast, yielded broader, more uniform distributions, reflecting greater textural and intensity variability. A principal component analysis of average node features reinforced the point: diseased and healthy graphs separated cleanly into distinct clusters in two-dimensional principal component space, demonstrating that the graph representations encode discriminative signals even before classification. In other words, the disease leaves a measurable structural fingerprint in the graph itself, something a standard CNN scanning local patches would not explicitly capture.

Interpretability, a persistent obstacle to clinical adoption of artificial intelligence, received dedicated attention. Because standard GIN and GraphSAGE models do not produce native attention coefficients, the authors computed gradient-based node saliency maps, scoring each node by the magnitude of the gradient of the predicted class output with respect to that node’s features. Nodes whose features most influence the prediction, and therefore correspond to regions the model relies on most heavily, light up in these maps, allowing clinicians to see which parts of the retina drove a diagnosis and to compare the structural patterns emphasized by each architecture. The team argues that such visualizations align with clinical observations and can build trust among medical professionals, a prerequisite for any diagnostic tool hoping to enter real workflows.

The study also probed robustness and statistical rigor. An ablation study trained the models on 10, 25, 50, 75 and 100 percent of the labeled data, with stratified sampling and multiple random seeds, to measure how performance degrades under scarce supervision, a common reality in medical imaging where expert annotations are expensive. A stratified five-fold cross-validation, with class weighting computed independently within each training fold and quadratic weighted kappa emphasized as the primary metric given the ordinal nature of the APTOS labels, confirmed the robustness of the results. A Wilcoxon signed-rank test on the paired performance differences between the two models produced a statistic of 7.0 and a p-value of 0.5625, meaning the observed differences between GIN and GraphSAGE were not statistically significant, so the choice between them can legitimately be guided by diagnostic priorities rather than raw superiority.

That trade-off has practical consequences. In a screening context, recall matters most, because a missed case of retinopathy, a false negative, can cost a patient their sight; GIN’s 98.66 percent recall makes it the natural choice when sensitivity is paramount. In a setting where false alarms trigger costly referrals and unnecessary anxiety, GraphSAGE’s higher precision and specificity become more valuable. The authors also analyzed computational complexity, showing that both architectures scale as O(L(Ed + Nd squared)) for L layers, E edges, N nodes and hidden dimension d, though GraphSAGE carries roughly twice the trainable parameters of GIN because its concatenation operation doubles the input dimensionality of each transformation layer. Notably, the entire pipeline, from graph construction through training and testing, ran in roughly sixteen minutes on a single cloud GPU, demonstrating that such models are feasible even in resource-constrained environments.

The researchers are candid that this is a step, not a destination. Future work, they write, should focus on optimizing hybrid CNN-GNN architectures that exploit the complementary strengths of both paradigms, expanding datasets with more diverse and longitudinal samples to improve generalizability, and streamlining the models for real-time clinical deployment. Integrating attention-driven and explainable artificial intelligence techniques, they argue, will be essential to earning clinician trust and embedding graph-based diagnostics into ophthalmology workflows. If the near-perfect discrimination reported here holds up in prospective, multi-center studies on unfiltered clinical data, graph neural networks could reshape automated retinal screening, catching the disease earlier, easing the burden on overworked specialists and, ultimately, preserving sight for millions of people living with diabetes.

Subject of Research: Graph neural network-based classification of diabetic retinopathy from retinal fundus images

Article Title: Graph hybrid neural network integration for enhanced diabetic retinopathy detection in retinal image

Article References: Afzal, S., Rakshit, B., & Somasundaram, K. (2026). Graph hybrid neural network integration for enhanced diabetic retinopathy detection in retinal image. Discover Artificial Intelligence, 6(1), Article 1427. https://doi.org/10.1007/s44163-026-02375-w

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02375-w

Keywords: diabetic retinopathy, graph neural networks, GraphSAGE, graph isomorphism network, deep learning, retinal images, medical image classification, APTOS 2019 dataset, ResNet50, computer vision, ophthalmology, explainable AI

The problem with existing approaches, according to the research team led by Yuhan Zhang and Jing Zhao, is that conventional image generators operate at the pixel level. They learn what parks look like but not how parks work. Real park design is a decision-making process in which functional spaces act as nodes and spatial relationships act as edges in a network. Designers must balance where to place plazas, sports areas, and ecological zones, how those zones connect to one another, and how pathways, water bodies, and vegetation knit everything together. Standard generative models have no built-in mechanism to enforce that spatial logic, which is why their outputs often contain inconsistencies, such as activity areas stranded far from any path or circulation routes that make no functional sense.

The researchers’ framework works in two sequential stages. In the first, a GraphSAGE model learns the semantic relationships and spatial connections among park functional nodes. Users provide a site condition image and a set of functional requirements, and the model predicts which pairs of functional zones should be connected, generating an explicit relationship graph. Crucially, the team defined park relationships more broadly than simple geometric contact: two zones count as connected if they share a boundary, if they are linked through transitional elements such as roads, plazas, or water bodies, or if they form part of a continuous circulation route from entrances to core public spaces to secondary activity areas. This composite definition captures the layered, dynamic nature of park spatial organization that wall-based architectural models cannot.

In the second stage, the predicted relationship graph is transformed into a spatially structured functional zoning diagram using park design standards and a Voronoi post-processing algorithm, which allocates pixel areas within the irregular site boundary according to each node’s target size. A CycleGAN model then maps that zoning diagram onto a concrete park design scheme, refining visual details and spatial morphology. The team added two custom loss terms, a region semantic consistency loss and an instance semantic consistency loss, to the standard CycleGAN objective, explicitly forcing the generator to preserve regional correspondence among roads, green spaces, and water bodies and to maintain the boundaries of independent functional blocks.

Building the training data was a substantial undertaking in itself. The researchers collected 9,286 urban park design scheme drawings from professional design repositories and online platforms worldwide, focusing on medium-scale comprehensive parks of 10 to 20 hectares, a range chosen because such parks serve district-level populations with complete composite functions while exhibiting similar design logic in spatial granularity and composition. From these, 198 schemes with clearly delineated functional zoning were manually screened and annotated using LabelMe, then augmented to 1,188 graph samples. Functional spaces were consolidated under a blue-green-grey infrastructure classification framework into a three-tier system spanning plazas, sports and fitness areas, ecological green spaces, waterfront spaces, and other major types.

The experiments showed that the graph generation stage performs robustly. As the training set expanded from roughly 100 to 800 samples, the model’s test AUC climbed from 0.8207 to 0.8991, with the steepest gains occurring below 400 samples and performance plateauing beyond 600. When compared head-to-head against GCN, GAT, and GIN architectures on identical data, GraphSAGE achieved the best results, with a validation AUC of 0.909 and an accuracy of 0.846, a finding the researchers attribute to its inductive neighborhood sampling, which suits park graphs where node types and quantities vary and adjacency relationships are sparse. The generated graphs also reflected genuine design principles: open lawns, water bodies, and plazas emerged as highly connected central hubs, while dense forest areas sat at the periphery as ecological buffers, exactly as human designers would organize them.

The full hybrid model’s most striking advantage appeared in planning utility metrics. Compared with a CycleGAN baseline and Stable Diffusion, the full model achieved a circulation connectivity index of 70.2 percent versus 46.7 percent for the baseline and 38.2 percent for Stable Diffusion, along with a functional adjacency compliance rate of 78.4 percent and a node accessibility ratio of 82.5 percent. It also recorded the lowest Fréchet Inception Distance at 0.3190, indicating the closest match to the real distribution of park layouts. Interestingly, the baseline CycleGAN actually scored higher on pixel-level similarity metrics such as PSNR and SSIM, suggesting that the graph constraints deliberately shift the model away from pixel-perfect replication and toward diverse layouts that obey functional logic.

Ablation experiments confirmed that the graph reasoning component carries most of the weight. Removing GraphSAGE caused the largest performance degradation, with FID rising from 0.3190 to 0.3634, while removing the rule-based coordinate assignment or Voronoi allocation produced smaller declines, indicating those components mainly handle geometric translation rather than functional reasoning. Semantic-category analysis showed the largest FID improvements for roads, water bodies, and green space, the elements with strong spatial organization characteristics, while buildings, which occupy small and sparse areas, saw little benefit. A panel of 12 landscape professionals scored the hybrid model’s outputs above 7 out of 10 on all six evaluation dimensions, including spatial structure, park roads, and usability, consistently higher than the baseline, though still below manually designed ground truth.

The researchers are careful to position the tool as an assistant rather than a replacement. The framework is designed for the early conceptual phase of design, where it can rapidly generate layout drafts with basic spatial logic in about 2.12 seconds per 512-by-512 image, allowing designers to explore and compare schemes before committing to detailed development. It embeds the generative model within the knowledge-driven workflow that designers already follow, from requirement analysis through functional zoning to spatial composition, rather than attempting to bypass that reasoning with a purely image-to-image translation.

Limitations remain. The training data concentrates on medium-scale comprehensive parks, so the model is not yet suited to pocket parks or very large green space systems, and the unsupervised CycleGAN stage still produces limited detail in small-scale elements and textures. The team suggests future work could expand the dataset across park types and styles and enrich the graph inputs with terrain elevation, surrounding context, and existing vegetation. But the core insight stands: teaching generative AI to reason about space as a graph before it renders an image produces park designs that are more rational, more usable, and far closer to what professional landscape architects would actually build.

Subject of Research: Graph-constrained generative AI for automated urban park layout design

Article Title: Graph-constrained generative adversarial network for automated urban green space layout generation

Article References: Zhang, Y., Liu, Y., Ni, B., Chen, R., & Zhao, J. (2026). Graph-constrained generative adversarial network for automated urban green space layout generation. Machine Learning with Applications, 26, Article 101020. https://doi.org/10.1016/j.mlwa.2026.101020

Image Credits: AI Generated

DOI: 10.1016/j.mlwa.2026.101020

Keywords: generative adversarial networks, graph neural networks, urban parks, landscape architecture, CycleGAN, GraphSAGE, generative design, spatial layout generation, urban green space, artificial intelligence, Voronoi diagram, human-AI collaboration

The researchers focused on Cyprus, a country that maintained detailed surveillance data across the first four waves of the COVID-19 pandemic. Each confirmed case in their dataset carried a set of epidemiological attributes, including age, gender, date of infection, residency, and a classification of the case according to NACE, the European statistical classification of economic activities that encodes occupational context. Cases that had been linked by contact tracing formed the observed edges of an infection network, while the unlinked cases represented the missing structure the team hoped to recover. Framing the problem this way turns source attribution into a link prediction task: given two cases, can a model estimate how likely it is that one infected the other, even when no tracer ever documented that connection?

The first of the two approaches evaluated in the study relied on classical machine learning classifiers trained on engineered edge-level features. For every candidate pair of cases, the researchers combined the node-level attributes of the two individuals using a variety of operators, producing a feature vector that described the pair rather than either case alone. This design allowed the models to learn patterns such as whether infections were more likely between people of similar ages, overlapping infection dates, the same region of residence, or compatible occupational categories. Because the raw feature space constructed from all these combinations was large, the team applied Minimum Redundancy Maximum Relevance, or mRMR, to select features that were strongly associated with true transmission links while avoiding redundant information, and then refined the selection with greedy forward feature selection. Permutation importance was subsequently used to quantify how much each retained feature contributed to the classifiers’ decisions.

Among the classifiers tested, Random Forest and Gradient Boosting consistently delivered the strongest performance. During the first two pandemic waves, these models achieved F1-scores ranging from approximately 0.82 to 0.90, indicating a strong balance between precision and recall in distinguishing plausible transmission pairs from implausible ones. Mean Reciprocal Rank values, which measure how highly the true infector appears in a model’s ranked list of candidate sources, ranged from 0.55 to 0.64 for those early waves. In practical terms, this means that for a substantial share of unlinked cases, the correct or nearly correct infector sat at or near the top of the model’s shortlist, exactly the kind of ranking an epidemiologist would need to prioritize follow-up investigations.

Performance declined in the later waves, and the decline is itself informative. For the third and fourth waves, F1-scores ranged from 0.68 to 0.75 and MRR values dropped to between 0.23 and 0.33. The authors’ results suggest that the structure of transmission changed as the pandemic evolved: later waves in Cyprus were shaped by widespread community transmission, variants of concern, and vaccination, all of which blur the simple epidemiological signatures, such as tight date proximity or shared workplace, that made early-wave links easier to predict. When many candidate infectors share similar attributes with a case, the ranking task becomes fundamentally harder, and the numbers reflect that added ambiguity rather than any weakness in the method itself.

The second approach moved beyond hand-engineered features and turned to graph representation learning. The team employed a GraphSAGE model, a neural architecture that generates node embeddings by aggregating information from a node’s attributes and from the topology of the observed network around it. Each case is thus represented as a dense vector that encodes both what the case looks like epidemiologically and where it sits in the web of known connections. To score a candidate link between two cases, the researchers evaluated several strategies for combining the two embeddings, including concatenation, absolute difference, squared difference, the Hadamard product, and the dot product, each of which captures a different notion of similarity or compatibility between the pair.

Link prediction with graph representation learning achieved F1-scores ranging from roughly 0.70 to 0.79 and MRR values from 0.23 to 0.42 across all four pandemic waves. Compared with the classical classifiers, the GraphSAGE approach was somewhat less accurate in the early waves, where the engineered features captured the epidemiological signal very effectively, but it offered a more uniform level of performance across the entire pandemic period. This consistency matters for public health practice, because a tool that degrades sharply between epidemic phases is harder to integrate into routine surveillance workflows than one whose reliability is more stable. The embedding-based method also has the advantage of learning representations directly from the network, potentially capturing relational patterns that a fixed feature-engineering pipeline would miss.

The most consequential step of the study came after validation. Having confirmed that the best-performing classifier could reliably recover known links held out from the training data, the researchers applied it to the genuinely unlinked cases in the Cypriot dataset, asking the model to infer the most likely infector for each isolated node. The resulting reconstructed networks showed markedly fewer isolated components and larger connected structures than the original traced networks. Importantly, the reconstruction preserved epidemiologically relevant characteristics of the network, most notably the outdegree distribution, which reflects how many secondary infections are attributed to each case. Maintaining this property indicates that the inferred links did not distort the underlying transmission dynamics, such as the presence of superspreading individuals, but instead extended the observed network in a plausible direction.

The implications of this work extend well beyond one country or one pathogen. Incomplete infection networks are a chronic feature of outbreak response, from foodborne disease investigations to emerging zoonotic threats, and the quality of public health decisions depends heavily on how accurately those networks reflect reality. A reconstructed network with fewer isolated cases gives epidemiologists a more complete picture of transmission chains, enabling more targeted interventions: contact investigations can be prioritized toward the candidate sources ranked highest by the model, and containment resources can be directed at the connection points that matter most for interrupting spread. The authors note that such tools are particularly valuable during small epidemic waves, when case numbers are limited and every unexplained infection carries a proportionally larger informational cost.

The study also illustrates a broader lesson in applied machine learning for public health: classical feature-based models and modern graph neural networks are complementary rather than competing tools. Engineered epidemiological features, when chosen carefully with methods like mRMR and validated with permutation importance, can deliver excellent performance when the transmission signal is strong and interpretable. Graph representation learning, by contrast, offers robustness across changing epidemic conditions and the flexibility to incorporate network structure directly. As surveillance systems grow richer and more digitized, hybrid pipelines that combine both strategies, validated rigorously on known links before being trusted with unknown ones, are likely to become a standard component of the computational epidemiology toolkit, turning the frustrating gaps in contact tracing data from dead ends into testable hypotheses about the hidden architecture of disease spread.

Subject of Research: Link prediction in COVID-19 infection networks using machine learning and graph representation learning

Article Title: Predicting missing links in COVID-19 infection networks

Article References: Dimitriou, P. A., Silvestros, V., Constantinou, E., Pitris, C., & Kolios, P. (2026). Predicting missing links in COVID-19 infection networks. PLOS Complex Systems, 3(8), e0000124. https://doi.org/10.1371/journal.pcsy.0000124

Image Credits: AI Generated

DOI: 10.1371/journal.pcsy.0000124

Keywords: COVID-19, link prediction, machine learning, graph representation learning, GraphSAGE, contact tracing, epidemiology, Cyprus, Random Forest, Gradient Boosting, network reconstruction, public health

The heart of the new system is a carefully engineered convolutional neural network built on the U-Net architecture, a design that has become the workhorse of medical image segmentation since its introduction. U-Net’s signature shape, an encoder that compresses an image into abstract features followed by a decoder that expands those features back into a pixel-level mask, is well suited to outlining structures in three-dimensional brain scans. But standard U-Net models often struggle with the peculiar challenges of multiple sclerosis lesions, which are small, irregularly shaped, scattered throughout the white matter, and easily confused with other bright spots on magnetic resonance images. The research team addressed this by bolting two powerful attention mechanisms onto the backbone. The first is the Convolutional Block Attention Module, known as CBAM, which refines feature maps along two complementary axes: it learns which channels of information matter most and which spatial regions deserve focus, effectively teaching the network where to look and what to weigh.

The second enhancement is an Extended Atrous Spatial Pyramid Pooling module, a technique borrowed from semantic segmentation research that solves a different problem: scale. Lesions in the brain come in wildly different sizes, from a few voxels to sprawling patches of demyelination, and a network that only looks at one resolution tends to miss the extremes. Atrous convolution, sometimes called dilated convolution, widens the receptive field of the filters without adding parameters or losing resolution, allowing the model to sample context at multiple scales simultaneously. By stacking these dilated convolutions in a pyramid arrangement, the module captures both the fine texture of a tiny lesion and the broader anatomical context of a large one. The combination of channel-wise attention, spatial attention, and multi-scale context extraction is what the authors credit for the model’s standout performance on the Brain Magnetic Resonance Dataset of Multiple Sclerosis, where it achieved its near-perfect Dice score, a metric that measures the overlap between the machine’s segmentation and expert annotations.

Impressive as those numbers are, the team did not stop at a single dataset. To test whether the architecture had genuinely learned the anatomy of abnormal tissue rather than memorizing the quirks of one collection of scans, they evaluated MedSegNet-AXU on the widely used Brain Tumor Segmentation challenges from 2019, 2020, and 2021. Across different imaging modalities and tumor sub-regions, the model delivered mean Dice scores between 92 and 96 percent and Jaccard indices, a stricter measure of overlap, between 85 and 91 percent. This cross-domain validation matters because a segmentation tool that only works on one dataset is a laboratory curiosity, while one that generalizes across diseases and imaging protocols could become a genuine clinical instrument. The results suggest that the attention-driven design captures features of pathological tissue that transcend the specific appearance of multiple sclerosis lesions.

But segmentation, however accurate, is only the first half of the story. Once the network had traced the boundaries of each lesion, the researchers extracted 21 radiomic features from the segmented tissue, quantitative descriptors that capture properties such as shape, intensity distribution, and texture that the human eye cannot reliably grade. Crucially, they organized this analysis around three functional systems that multiple sclerosis attacks: the sensory, motor, and visual pathways. The idea is elegant in its simplicity. If the mathematical fingerprint of a lesion correlates with the symptoms a patient experiences, then a scan alone could hint at how the disease is manifesting, potentially flagging damage before a clinical exam reveals it. The radiomic features were then fused with clinical metadata, information about the patients themselves, to create a richer description of each case than imaging or demographics could provide alone.

With dozens of candidate features in hand, the team faced the classic machine learning dilemma of separating signal from noise. They applied Chi-square feature selection, a statistical test that scores each feature by how strongly its distribution relates to the outcome of interest, and retained the top 20 for classification. Then came the most novel step. Rather than feeding these features into conventional classifiers alone, the researchers represented each patient as a node in a graph, connected by edges that encode relationships in the data, and trained graph neural networks to make predictions. Among the models tested, GraphSAGE stood out dramatically. This algorithm works by allowing each node to aggregate information from its neighbors, layer by layer, so that a patient’s classification is informed not just by their own features but by those of similar patients in the learned graph structure. It achieved test accuracies of 86.21 percent for the sensory system, 96.55 percent for the motor system, and 75.86 percent for the visual system, significantly outperforming the traditional machine learning baselines.

High accuracy alone can be misleading in medical machine learning, where a model may latch onto spurious correlations that vanish in the clinic. The researchers therefore subjected GraphSAGE to two independent reliability checks. The first, Threshold Analysis with Anomaly Edges, probes how the model behaves when unusual or out-of-distribution connections appear in the graph, testing whether its predictions degrade gracefully or collapse when faced with atypical patients. The second, Correlation Range Clustering, examines the structure of the relationships the model relies on, verifying that the graph edges reflect meaningful similarity rather than artifacts of construction. The fact that the model survived these stress tests lends credibility to the headline numbers and addresses one of the most persistent criticisms of deep learning in medicine: that its predictions are often opaque and fragile in the face of real-world variability.

The clinical implications of this pipeline are considerable. Today, diagnosing and monitoring multiple sclerosis depends on radiologists manually counting and measuring lesions across time points, a laborious process prone to inter-observer disagreement, and on clinical scales that quantify disability through examination. A framework that automatically segments lesions, extracts quantitative features, and links them to functional systems could standardize this workflow, reduce the variability that plagues longitudinal monitoring, and give neurologists an earlier, more objective read on disease activity. The fusion of imaging with clinical metadata also points toward personalized medicine: if a patient’s lesion profile predicts motor decline, for example, treatment intensity might be adjusted before irreversible damage accumulates. The authors frame the approach as a comprehensive tool for lesion analysis aimed at precise diagnosis and personalized treatment strategies, and the architecture’s demonstrated portability to tumor segmentation hints at applications well beyond one disease.

Caveats remain, as they always do at this stage of translation. The model was trained and validated on research datasets rather than deployed in a live hospital setting, and the classification accuracies, while strong, vary across the three functional systems, with the visual system proving hardest to predict. Prospective studies in diverse patient populations will be needed to confirm that the performance holds under the messier conditions of routine clinical imaging, with different scanners, protocols, and artifact levels. Nevertheless, the study represents a notable convergence of several trends in medical artificial intelligence: attention mechanisms that make segmentation networks more discerning, radiomics that convert images into measurable biology, and graph learning that exploits the relationships among patients rather than treating each case in isolation. Published open access with a permanent DOI, the work invites other groups to build on it, and it offers a glimpse of a future in which a routine brain scan quietly tells a neurologist not just where the lesions are, but what they mean for the person carrying them.

Subject of Research: Deep learning segmentation of multiple sclerosis lesions and graph-based classification using MRI radiomics and clinical data

Article Title: MedSegNet-AXU: advanced segmentation and graph-based classification of multiple sclerosis lesions from 3D magnetic resonance imaging via radiomics and clinical data fusion

Article References: Karim, W., Zaman, S. B., Sutradhar, D., Debnath, R. K., Azam, S., Yeo, K. C., Zhang, Y., & Jonkman, M. (2026). MedSegNet-AXU: advanced segmentation and graph-based classification of multiple sclerosis lesions from 3D magnetic resonance imaging via radiomics and clinical data fusion. Complex & Intelligent Systems. https://doi.org/10.1007/s40747-026-02535-6

Image Credits: AI Generated

DOI: 10.1007/s40747-026-02535-6

Keywords: multiple sclerosis, lesion segmentation, 3D brain MRI, U-Net, attention mechanism, CBAM, atrous spatial pyramid pooling, radiomics, graph neural networks, GraphSAGE, feature selection, clinical data fusion

The problem the researchers set out to solve is one of the most stubborn in modern exploration geophysics. Geophysical signals from deep ore bodies arrive at the surface attenuated and weak, their low amplitudes easily drowned out by near-surface geological background. Conventional inversion techniques that translate raw electromagnetic, magnetic, and gravity measurements into three-dimensional models of the subsurface suffer from non-uniqueness: many different underground configurations can produce nearly identical surface readings. When the target is a subtle mineralized body rather than a large, conductive massive sulfide, the ambiguity becomes severe. Human interpreters and standard machine learning tools alike struggle to decide which faint blips in an inversion volume are genuine exploration targets and which are artifacts.

What makes the challenge doubly hard is the label problem. Machine learning classifiers typically need thousands of labeled examples to learn reliably, but in mineral exploration the only trustworthy labels come from expensive deep drilling. At Kalatongke, the research team had just 906 labeled samples drawn from boreholes to constrain a volume of 2,023,623 three-dimensional inversion voxels. That means less than half of one tenth of one percent of the data carried any ground truth. Worse, mineralized intervals are dramatically outnumbered by barren rock, producing a severe class imbalance that causes ordinary classifiers to simply predict the majority class everywhere and miss the ore entirely.

LSR-GAD, short for a label-scarce reliability-aware graph anomaly detection framework, attacks both problems at once. The core idea is to represent the three-dimensional inversion volume as a graph rather than a stack of independent pixels. Each inversion voxel becomes a node in the graph, carrying its electrical, magnetic, and gravity inversion properties as features. A graph sample and aggregate network, known as GraphSAGE, then encodes local spatial relationships by aggregating information from neighboring nodes. This neighborhood aggregation matters because ore bodies are spatially continuous objects, not random scatterings of anomalous voxels; a voxel surrounded by other weakly anomalous voxels is far more likely to be part of a real mineralized zone than an isolated outlier, and the graph structure lets the model learn that context directly.

To cope with the scarcity of labels, the framework combines the small number of labeled borehole samples with unlabeled anchor nodes, allowing the network to extract useful structure from the vast unlabeled portion of the data while still being anchored by what little ground truth exists. The class imbalance is handled through a weighted binary cross-entropy loss, which penalizes errors on the rare mineralized class more heavily than errors on the abundant barren class, effectively telling the model that missing an ore body is a far costlier mistake than a false alarm in barren rock. Synthetic experiments and ablation tests confirmed that each ingredient contributes: removing the graph structure, the sample weighting, or the regularization each degraded the continuity of predicted targets and increased missed detections under sparse-label conditions.

Perhaps the most consequential design choice is the explicit treatment of uncertainty. Rather than producing a single brittle prediction, the framework incorporates Monte Carlo dropout, a technique in which the network’s dropout layers remain active at inference time so that many slightly different versions of the model each contribute a prediction. From this ensemble of predictions, the method estimates predictive probability, standard deviation, and entropy for every voxel. High probability with low variance marks confident exploration targets; high probability with high variance flags regions where the model is guessing. This turns the output from a binary map into a risk-aware decision tool, letting exploration managers weigh where to spend scarce drilling budgets against quantified confidence rather than blind faith in a black box.

Scale presented one final engineering hurdle. A million-scale voxel volume cannot be pushed through a graph network in one piece, so the team developed an anchor-graph-based block-wise inference strategy that partitions the volume into manageable blocks for prediction while preserving the benefits of the graph representation. This makes the approach practical for real exploration datasets, which routinely span kilometers of depth and lateral extent at fine spatial resolution. The combination of graph-based spatial reasoning, semi-supervised learning, class-rebalancing, uncertainty estimation, and scalable inference is what allows LSR-GAD to function reliably in exactly the regime where conventional supervised methods collapse: deep, subtle targets with almost no labels.

The Kalatongke case study demonstrates the framework on a genuinely difficult target. Kalatongke is a magmatic copper-nickel sulfide district hosted in the Central Asian Orogenic Belt, a vast accretionary collage where Permian mafic intrusions carried sulfide melts rich in copper and nickel. Known ore bodies have been extensively drilled, but the next generation of discoveries lies deeper and subtler, in concealed intrusions whose geophysical signatures are faint and entangled with background geology. The team integrated electrical, magnetic, and gravity inversion models, letting the classifier exploit the fact that a genuine mineralized body should leave coherent, mutually consistent traces across multiple physical properties rather than an artifact confined to a single dataset.

Trained on the 906 labeled samples, the model generated a full probability and uncertainty map of the survey volume. The stand-out result is the high-probability anomaly near line 805 at approximately 1.0 kilometer depth, which corresponds spatially to the gabbroic interval encountered in borehole ZK2017-6. Because that agreement was achieved with a tiny training set, it suggests the framework can generalize from limited ground truth in ways that matter for greenfield exploration, where deep drilling data simply does not yet exist. The authors position the output explicitly as a risk-management tool: probability and uncertainty layers guide drill targeting, and each new borehole feeds back into the training set, progressively sharpening the model as a district matures.

Beyond one deposit, the study points toward a broader shift in how geoscience handles sparse, expensive labels. Deep learning is spreading rapidly through geophysics, but most successful applications sit in data-rich domains; mineral exploration, with its handful of boreholes against millions of unlabeled voxels, has remained a hostile environment for standard supervised learning. By fusing multi-source geophysical data on a graph, rebalancing the loss, and quantifying its own uncertainty, LSR-GAD offers a reusable template for intelligent exploration of complex concealed ore bodies, whether the target is nickel-copper sulfide, rare earths, or energy-critical metals. As demand for critical minerals accelerates worldwide, tools that can squeeze reliable guidance from minimal ground truth may become as essential to explorers as the magnetometer itself.

Subject of Research: A label-scarce reliability-aware graph anomaly detection framework for identifying deep subtle mineralization anomalies from multi-source geophysical inversion data, applied to the Kalatongke Cu-Ni deposit in the Central Asian Orogenic Belt.

Article Title: LSR-GAD for Reliable Detection of Deep Subtle Mineralization Anomalies under Label Scarcity: A Case Study of the Kalatongke Deposit in the Central Asian Orogenic Belt

Article References: Lv, P., Zhou, N., Chen, W., & Han, S. (2026). LSR-GAD for Reliable Detection of Deep Subtle Mineralization Anomalies under Label Scarcity: A Case Study of the Kalatongke Deposit in the Central Asian Orogenic Belt. Natural Resources Research. https://doi.org/10.1007/s11053-026-10765-1

Image Credits: AI Generated

DOI: 10.1007/s11053-026-10765-1

Keywords: graph neural network, mineral exploration, label scarcity, geophysical inversion, Kalatongke deposit, uncertainty quantification, Monte Carlo dropout, GraphSAGE, multi-source data fusion, copper-nickel sulfide, Central Asian Orogenic Belt, machine learning

The physical mechanism most widely invoked to explain particle adhesion is adiabatic shear instability. During impact, the periphery of the particle experiences extreme strain rates, often exceeding ten to the seventh per second. When thermal softening locally outpaces strain hardening, a narrow shear band forms, concentrating deformation and heat into a thin region at the particle-substrate interface. This localized softening promotes the formation of material jets and enables metallurgical bonding between surfaces that would otherwise rebound elastically. The Johnson-Cook constitutive model, which couples strain, strain-rate, and temperature effects into a single flow stress description, is the standard framework for capturing this behavior in finite element simulations, and its parameters directly shape how the response surface varies across the process parameter space.

Single-particle finite element models, such as the spherical particle on a cylindrical substrate configuration used in this study, have become the canonical tool for interrogating these mechanisms. The geometry chosen here, a forty-micrometer-radius aluminum particle striking an aluminum substrate of two hundred fifty micrometer radius and depth, reflects the typical scale of cold spray powder feedstock and allows the impact event to be resolved with sufficient mesh refinement to capture the deformation gradients near the contact zone. Normal incidence is assumed, which is a reasonable first approximation since most particles in a cold spray jet strike the substrate at or near perpendicular orientation, although oblique impacts and particle-particle interactions in real deposits introduce additional complexity that single-particle studies deliberately set aside.

The five output targets selected for prediction capture complementary aspects of the impact response. Maximum equivalent plastic strain quantifies the severity of deformation, which correlates with the extent of interfacial contact area development and hence with bonding likelihood. Average contact plastic strain provides a more spatially averaged measure of deformation at the interface. Maximum temperature indicates whether adiabatic heating approaches the softening regime necessary for jetting. Maximum von Mises stress characterizes the mechanical loading experienced by the material, relevant to residual stress development and potential substrate damage. The deformation ratio, finally, describes the geometric flattening of the particle, a quantity experimenters can measure directly in cross-sectioned deposits, making it a useful bridge between simulation and experimental validation.

The choice of particle velocity, particle temperature, and friction coefficient as the three input parameters reflects their dominant roles in governing impact outcomes. Velocity controls the kinetic energy available for plastic work and is widely regarded as the single most influential cold spray parameter, with critical velocities below which particles rebound and above which they bond. Particle temperature, set by the gas temperature in the spray nozzle, pre-softens the material and lowers the energy barrier for deformation. The friction coefficient at the interface, though harder to control experimentally, governs tangential restraint and energy dissipation during sliding contact, influencing jet formation and the distribution of plastic strain around the contact periphery.

The surrogate modeling strategy adopted in this work addresses a persistent bottleneck in process simulation. A single resolved finite element impact simulation can require substantial computational resources, and exploring a three-dimensional parameter space at useful resolution demands hundreds or thousands of such runs. Once trained, a surrogate model evaluates new process conditions in milliseconds, enabling optimization studies, sensitivity analyses, and inverse design tasks that would be intractable with direct simulation. The practical value of the surrogate, however, depends entirely on its accuracy across the operating envelope, which is precisely where the choice of machine learning architecture becomes consequential.

The central representational innovation of the study is the treatment of each simulation sample as a node in a k-nearest-neighbour graph constructed in feature space, rather than as an isolated feature vector. In conventional feedforward networks, each training example is processed independently, and any information about the similarity between neighboring process conditions is implicit only in the aggregate statistics of gradient descent. By contrast, graph neural networks explicitly pass messages along edges connecting similar samples, allowing each node’s prediction to be conditioned on the responses of its parametric neighbors. For a physical system like cold spray impact, where the response surface is smooth and continuous in the vicinity of any given operating point, this inductive bias aligns naturally with the structure of the underlying data-generating process.

The comparative results carry a clear message about which architectural assumptions suit this problem. GraphSAGE-style spatial aggregation and the geometric attention network both achieved coefficients of determination above 0.93 for most targets, with the attention model reaching 0.97 for maximum plastic strain. These architectures share a common principle: they aggregate information from spatially proximate neighbors in feature space, weighting that information either uniformly or through learned attention coefficients. The attention mechanism’s slight edge is physically sensible, since the velocity-dominated nature of the response means that neighbors at different velocities carry unequal informational value, and attention weights can adaptively emphasize the most relevant ones.

The comparatively poor performance of the Chebyshev spectral graph convolution network and the topologically augmented multilayer perceptron, including negative R-squared values on several targets, is instructive rather than merely negative. Spectral methods operate through polynomial approximations of the graph Laplacian, which excel at multi-scale feature extraction when the graph structure itself carries meaningful community or frequency information. For a k-nearest-neighbour graph built from a relatively smooth parametric dataset, the spectral structure may be too weak or too sensitive to the choice of graph construction to provide a useful signal. Similarly, persistent homology descriptors encode global topological features of the point cloud, such as connected components and loops across scales, which may be largely uninformative for a response surface governed by local parametric gradients rather than by global shape features.

Negative R-squared values deserve particular emphasis for readers less familiar with regression diagnostics. A negative coefficient of determination indicates that the model’s predictions are worse than simply predicting the mean of the training data for every input. This is not a marginal failure but a categorical one, signaling that the architecture has failed to extract any generalizable input-output relationship from the training set. In surrogate modeling applications, such failures typically arise from an inductive bias mismatched to the data structure, or from architectures whose capacity is poorly matched to the available training set size, rather than from noise in the underlying simulation data, which is deterministic in this case.

The velocity-dominated character of the input-output relationships, confirmed by the three-dimensional feature space visualizations and two-dimensional contour projections, is consistent with decades of experimental cold spray research. Deposition efficiency, critical velocity, and coating quality all vary steeply with particle velocity, while temperature and friction act as secondary modulators. A surrogate model that respects this hierarchy, as the attention-based spatial aggregation evidently does, can allocate its representational capacity where the response varies most sharply. This interpretability of model performance in terms of constitutive physics is one of the study’s more valuable contributions, since it transforms an empirical architecture comparison into a statement about the physics of the process itself.

Several limitations frame the scope of these findings. The dataset derives from a single material system, aluminum on aluminum, with a single particle size and normal impact geometry, so extrapolation to dissimilar material pairs, oblique impacts, or multi-particle interactions remains untested. The friction coefficient is treated as a fixed input parameter, whereas in reality interfacial friction evolves with temperature, pressure, and surface state during the impact itself. Furthermore, the surrogate learns from simulation data and inherits any idealizations embedded in the finite element model, including the constitutive parameters of the Johnson-Cook model and the assumed contact behavior. Experimental validation against measured deformation ratios or deposition efficiencies would strengthen confidence in the surrogate’s predictions beyond the simulation domain.

Nevertheless, the framework points toward practical applications in cold spray process development. A validated surrogate could accelerate the identification of operating windows that maximize interfacial plastic strain while keeping substrate stresses within acceptable limits, or support real-time process control where nozzle gas conditions are adjusted in response to measured particle velocities and temperatures. The graph-based representation could also be extended to incorporate additional parameters, such as particle size distributions, substrate preheating, or nozzle standoff distance, as further node features, provided the training dataset is expanded accordingly through continued automated simulation campaigns.

More broadly, the study contributes to a growing recognition in computational materials science that the structure imposed on training data can matter as much as the choice of model family. Where physical responses vary smoothly across a parametric space, encoding that continuity directly into the learning architecture, as graph-based neighborhood aggregation does, provides a form of physics-informed bias that improves both accuracy and data efficiency. For solid-state deposition processes, and potentially for other impact-dominated manufacturing problems with expensive simulations and smooth response surfaces, this representational insight may prove as consequential as the specific performance numbers reported.

Subject of Research: Application of geometric and topological deep learning to predict thermo-mechanical performance in cold spray deposition process modeling

Article Title: Geometric and topological deep learning for predicting thermo-mechanical performance in cold spray deposition process modeling

Article References: Mishra, A. (2026). Geometric and topological deep learning for predicting thermo-mechanical performance in cold spray deposition process modeling. Discover Informatics, 1(1), Article 13. https://doi.org/10.1007/s44564-026-00012-3

Image Credits: AI Generated

DOI: 10.1007/s44564-026-00012-3

Keywords: cold spray deposition, geometric deep learning, graph neural network, GraphSAGE, finite element simulation, Johnson-Cook model, topological data analysis, graph attention network, thermal spray coating, additive manufacturing, surrogate modeling, supersonic particle impact