Ai Graph Neural Networks

Graph Neural Network Theory

Doctoral work examines the mathematical foundations of learning over graph structured data. Theory establishes what these architectures can and cannot represent.

Message Passing Frameworks

Research examines the paradigm where nodes exchange information with their neighbours. Message passing underpins almost every current graph architecture.

Neighbourhood Aggregation Methods

Doctoral study examines how information from neighbours is combined. Aggregation choice determines the expressive capability of an architecture.

Permutation Invariance Research

Research examines models producing identical output regardless of node ordering. Invariance is the defining structural requirement for graph learning.

Equivariance In Graph Models

Doctoral work examines outputs transforming consistently with input structure. Equivariance encodes symmetry directly into the architecture itself.

Expressive Power Analysis

Research examines which functions a graph architecture can represent. Expressivity limits determine which structures a model can distinguish.

Weisfeiler Leman Hierarchy Research

Doctoral study relates architectures to a classical graph distinguishing procedure. This hierarchy provides the standard yardstick for expressive power.

Graph Isomorphism Testing

Research examines determining whether two graphs are structurally identical. Isomorphism testing bounds what any message passing model can achieve.

Universal Approximation On Graphs

Doctoral work examines whether architectures can approximate arbitrary graph functions. Universality results clarify the theoretical ceiling of these models.

Oversmoothing Analysis

Research examines node representations converging as network depth increases. Convergence prevents deep graph models from learning useful distinctions.

Oversquashing Research

Doctoral study examines information compression when passing through narrow structures. Compression prevents distant nodes from influencing one another effectively.

Bottleneck Analysis In Graphs

Research examines structural constrictions limiting information flow across a graph. Bottleneck structure explains failures on tasks requiring long range interaction.

Depth And Receptive Field Research

Doctoral work examines how many layers determine the reachable neighbourhood. Depth choice balances long range reach against representation collapse.

Residual Connection Design

Research examines skip connections enabling deeper graph architectures. Skip connections substantially mitigate degradation with increasing depth.

Normalisation In Graph Models

Doctoral study examines stabilising values flowing through graph architectures. Graph structure makes standard normalisation approaches behave differently.

Graph Convolutional Networks

Research examines convolution generalised from grids to arbitrary graphs. These architectures established the foundation of the modern field.

Spectral Graph Methods

Doctoral work examines learning through the eigenstructure of graph operators. Spectral approaches connect graph learning with classical signal processing.

Graph Fourier Analysis

Research examines frequency representations of signals defined on graphs. Frequency analysis explains smoothing behaviour observed in these models.

Polynomial Filter Design

Doctoral study examines localised filters expressed as polynomials of graph operators. Polynomial filters avoid the cost of full eigendecomposition.

Spatial Convolution Methods

Research examines convolution defined directly through neighbourhood structure. Spatial formulations generalise readily to unseen graph structures.

Graph Attention Networks

Doctoral work examines learning how much each neighbour should contribute. Learned weighting outperforms uniform aggregation across many graph tasks.

Attention Mechanism Design In Graphs

Research examines architectural choices within graph attention computation. Attention design governs both expressivity and computational cost.

Graph Transformer Research

Doctoral study examines transformer architectures adapted to graph structured input. Global attention addresses the long range limitations of message passing.

Positional Encoding On Graphs

Research examines representing node position where no natural ordering exists. Position representation is essential for graph transformer architectures.

Structural Encoding Methods

Doctoral work examines encoding local structure as additional node features. Structural features raise expressivity beyond plain message passing.

Random Walk Feature Research

Research examines features derived from random traversals of a graph. Walk statistics capture structure that local aggregation entirely misses.

Sampling Based Architectures

Doctoral study examines architectures learning from sampled neighbourhoods. Sampling makes learning feasible on very large graph structures.

Inductive Learning On Graphs

Research examines models generalising to nodes and graphs unseen in training. Inductive capability is essential for any practical deployment.

Transductive Learning Research

Doctoral work examines learning where all nodes are present during training. Transductive settings dominate academic benchmarks yet rarely match practice.

Graph Autoencoder Research

Research examines learning compressed representations of graph structure. Autoencoders support link prediction and unsupervised representation learning.

Variational Graph Models

Doctoral study examines probabilistic latent representations of graphs. Probabilistic formulations support both generation and uncertainty estimation.

Graph Generative Models

Research examines models producing entirely new graph structures directly. Graph generation underpins molecular design and network synthesis applications.

Diffusion Models On Graphs

Doctoral work adapts denoising generative approaches to discrete graph structures. Discrete structure makes this adaptation genuinely nontrivial.

Graph Recurrent Architectures

Research examines recurrent computation applied over graph structured data. Recurrence supports sequential processing of nodes, edges and whole sequences.

Gated Graph Networks

Doctoral study examines gating mechanisms controlling information retention. Gating allows selective retention across many rounds of propagation.

Higher Order Message Passing

Research examines exchanging information between groups rather than single nodes. Higher order schemes exceed the expressivity of standard message passing.

Simplicial And Complex Networks

Doctoral work examines learning over structures with higher dimensional cells. These structures represent relationships that plain edges cannot capture.

Hypergraph Neural Networks

Research examines learning where single connections join many nodes. Hypergraphs naturally represent group rather than pairwise relationships.

Heterogeneous Graph Networks

Doctoral study examines graphs containing several distinct node and edge types. Most real world graphs are heterogeneous rather than uniform.

Relational Graph Models

Research examines architectures handling many distinct relation types. Relational modelling underpins knowledge graph representation learning.

Knowledge Graph Embedding

Doctoral work examines vector representations of entities and their relations. Embeddings support reasoning and completion over structured knowledge.

Multiplex Network Models

Research examines networks with several coexisting layers of connection. Layered structure represents differing relationship types simultaneously.

Signed Graph Learning

Doctoral study examines graphs containing both positive and negative relations. Negative relations require aggregation schemes that differ fundamentally.

Directed Graph Architectures

Research examines learning where connections have an inherent direction. Direction carries essential meaning in citation and dependency structures.

Bipartite Graph Models

Doctoral work examines graphs containing two distinct classes of node. Bipartite structure underlies recommendation and a wide range of matching problems.

Dynamic Graph Networks

Research examines graphs whose structure changes over the course of time. Most real networks evolve continuously rather than remaining fixed structures.

Temporal Graph Learning

Doctoral study examines representing time within graph structured learning. Temporal ordering carries information that static snapshots discard.

Continuous Time Graph Models

Research examines graphs evolving in continuous rather than discrete time. Continuous formulations suit event streams with irregular timing.

Spatiotemporal Graph Networks

Doctoral work examines learning across both spatial structure and time. These models underpin traffic, weather and sensor network forecasting.

Geometric Deep Learning Foundations

Research examines the unifying symmetry principles behind structured architectures. This framework connects graph, grid and manifold learning coherently.

Non Euclidean Representation Learning

Doctoral study examines learning in spaces with curvature or manifold structure. Curved spaces represent hierarchical relationships far more compactly.

Hyperbolic Graph Networks

Research examines embedding graphs within hyperbolic geometric spaces. Hyperbolic geometry represents tree like hierarchies with very few dimensions.

Group Equivariant Networks

Doctoral work examines architectures respecting specified mathematical symmetries. Built in symmetry substantially reduces the data required for learning.

Physics Informed Graph Models

Research examines embedding physical laws within graph architectures. Physical constraints improve both accuracy and extrapolation beyond training data.

Graph Neural Differential Equations

Doctoral study examines continuous depth formulations of graph propagation. Continuous formulations offer a principled view of very deep propagation.

Graph Sampling Methods

Research examines selecting portions of large graphs for tractable learning. Sampling strategy determines both efficiency and estimation bias.

Neighbour Sampling Research

Doctoral work examines selecting subsets of neighbours during aggregation. Neighbour sampling bounds computation for nodes with very many connections.

Subgraph Sampling Techniques

Research examines extracting subgraphs to serve as training examples. Subgraph methods permit standard batched training over enormous graph structures.

Cluster Based Training

Doctoral study examines training on densely connected graph partitions. Clustering preserves local structure while bounding memory requirements.

Mini Batch Training On Graphs

Research examines batched optimisation where examples are interdependent. Interdependence makes graph batching fundamentally harder than usual.

Full Batch Training Analysis

Doctoral work examines optimisation using the entire graph at each step. Full graph training is exact but rapidly exhausts all available memory.

Distributed Graph Training

Research examines training spread across many separate computing devices. Graph interdependence makes distribution far harder than for independent data.

Graph Partitioning For Training

Doctoral study examines dividing graphs across machines with minimal cutting. Partition quality determines communication cost during distributed training.

Memory Efficient Training

Research examines reducing memory demand during training of graph models. Memory is typically the binding constraint on the achievable graph scale.

Graph Model Scalability

Doctoral work examines behaviour as graph size grows by orders of magnitude. Methods effective on small benchmarks frequently fail entirely at scale.

Very Large Graph Learning

Research examines learning over graphs with billions of connections. Industrial graphs routinely exceed anything used in academic benchmarks.

Hardware Acceleration For Graphs

Doctoral study examines specialised hardware for irregular graph computation. Graph workloads suit conventional accelerators considerably less well.

Sparse Computation Research

Research examines efficient computation over sparse connectivity patterns. Sparsity is the defining computational characteristic of graph learning.

Graph Compiler Research

Doctoral work examines compiling graph computations into efficient programs. Compilation choices strongly affect achievable training throughput.

Inference Optimisation On Graphs

Research examines efficient prediction from already trained graph models. Inference cost dominates total expenditure in deployed graph based systems.

Model Compression For Graphs

Doctoral study reduces graph model size while preserving predictive quality. Compression determines which hardware can host a deployed model.

Quantisation Of Graph Models

Research examines reduced numerical precision within graph architectures. Precision reduction interacts unusually with sparse irregular computation.

Knowledge Distillation On Graphs

Doctoral work transfers capability from large graph models into smaller ones. Distillation on graphs must transfer structural as well as feature knowledge.

Regularisation In Graph Learning

Research examines techniques limiting overfitting within graph architectures. Graph structure substantially changes how conventional regularisation behaves.

Data Augmentation For Graphs

Doctoral study examines generating variants of graphs to improve learning. Augmentation is harder than for images because structure carries meaning.

Edge Perturbation Methods

Research examines modifying connectivity during training as regularisation. Connectivity perturbation reduces overfitting to specific graph structure.

Feature Masking Techniques

Doctoral work examines concealing node attributes during graph model training. Masking underpins many self supervised graph learning objectives.

Graph Contrastive Learning

Research examines learning by distinguishing related from unrelated graph views. Contrastive objectives learn useful representations without any labels.

Self Supervised Graph Learning

Doctoral study examines learning from graph structure without supervision. Labels are scarce while graph structure itself is abundantly available.

Pretraining On Graphs

Research examines general purpose training preceding task specific adaptation. Pretraining transfers less readily across graphs than across text.

Graph Foundation Model Research

Doctoral work examines broadly capable models spanning many graph domains. Domain heterogeneity makes general graph models genuinely difficult.

Transfer Learning Across Graphs

Research examines reusing knowledge between differing graph datasets. Structural differences frequently prevent successful knowledge transfer.

Domain Adaptation On Graphs

Doctoral study examines adapting models between graphs of differing character. Adaptation must address both feature and structural distribution differences.

Few Shot Graph Learning

Research examines learning graph tasks from very few labelled examples. Label scarcity is the normal condition across most graph applications.

Zero Shot Generalisation Research

Doctoral work examines model performance on entirely unseen graph tasks. Zero shot capability would transform practical usefulness very considerably.

Meta Learning On Graphs

Research examines learning to adapt quickly across many related graph tasks. Meta learning suits settings with many small related graph problems.

Continual Learning On Graphs

Doctoral study examines models absorbing new graph data without forgetting. Real graphs grow continuously and cannot be retrained from scratch.

Curriculum Learning For Graphs

Research examines ordering training examples from simpler toward harder. Ordering can improve both convergence speed and eventual final performance.

Active Learning On Graphs

Doctoral work examines choosing which nodes are most valuable to label. Selection exploits structure to maximise information from few labels.

Semi Supervised Node Classification

Research examines classifying nodes when only a few carry known labels. This setting is the canonical benchmark task within graph learning research.

Label Scarcity Research

Doctoral study examines learning where labelled examples are very limited. Labelling graph data frequently requires costly specialist expertise.

Label Noise On Graphs

Research examines learning where some of the assigned labels are incorrect. Noisy labels propagate through neighbourhoods and amplify their harm.

Class Imbalance In Graph Learning

Doctoral work examines highly uneven distribution of node categories. Imbalance is extreme in fraud detection and anomaly related applications.

Hyperparameter Optimisation

Research examines tuning configuration choices within graph architectures. Reported gains frequently vanish once tuning effort is properly equalised.

Neural Architecture Search For Graphs

Doctoral study automates discovery of effective graph architectures. Automated search explores design spaces too large for manual exploration.

Automated Graph Learning

Research examines automating the full graph modelling workflow end to end. Automation lowers the expertise barrier for applying these methods.

Benchmark Dataset Design

Doctoral work examines constructing meaningful graph learning benchmarks. Long used small benchmarks have distorted the direction of the field.

Evaluation Protocol Research

Research examines how graph methods should be fairly compared. Protocol differences explain many apparently contradictory published results.

Reproducibility In Graph Learning

Doctoral study examines whether reported graph results can be repeated. Reproduction studies have overturned many claimed methodological advances.

Leakage In Graph Benchmarks

Research examines information improperly crossing training and test partitions. Graph connectivity makes clean separation unusually difficult to achieve.

Baseline Comparison Research

Doctoral work examines whether simple methods match complex graph models. Well tuned simple baselines frequently match elaborate architectures.

Statistical Testing In Evaluation

Research examines sound statistical comparison between graph methods. Reported differences frequently fall within ordinary random variation.

Graph Library And Framework Research

Doctoral study examines software frameworks implementing graph learning. Framework capability determines what research can practically be attempted.

Data Loading And Pipeline Research

Research examines moving graph data efficiently into the computation itself. Data movement frequently dominates total graph model training time.

Computational Cost Analysis

Doctoral work examines resources consumed by differing graph learning methods. Cost analysis is frequently omitted from reported method comparisons.

Node Classification Research

Research examines assigning categories to individual nodes within a graph. Node classification is the most widely studied graph learning task.

Node Regression Methods

Doctoral study examines predicting continuous quantities for graph nodes. Regression tasks receive far less attention than classification does.

Link Prediction Research

Research examines predicting connections not currently present in a graph. Link prediction underpins recommendation and knowledge graph completion.

Edge Classification Methods

Doctoral work examines assigning categories to connections between nodes. Edge level tasks arise in transaction and interaction network analysis.

Graph Classification Research

Research examines assigning categories to entire graph structures. Graph level prediction underpins molecular property and material applications.

Graph Regression Methods

Doctoral study examines predicting continuous properties of whole graphs. Graph regression is central to scientific property prediction tasks.

Graph Matching Research

Research examines finding correspondence between nodes of separate graphs. Matching supports entity alignment and structural comparison tasks.

Subgraph Matching Methods

Doctoral work examines locating specified patterns within much larger graphs. Pattern search underpins query answering over structured relational data.

Community Detection Research

Research examines identifying densely connected groups within networks. Community structure reveals organisation invisible in raw connectivity.

Graph Clustering Methods

Doctoral study examines grouping nodes by structural and feature similarity. Clustering supports both analysis and computational partitioning.

Graph Partitioning Research

Research examines dividing graphs while minimising the connections severed. Partitioning underpins distributed processing of very large graph structures.

Anomaly Detection On Graphs

Doctoral work examines identifying unusual nodes, edges or substructures. Structural context reveals anomalies that isolated inspection misses.

Fraud Detection Research

Research examines detecting coordinated fraudulent activity within networks. Fraudsters form structural patterns that individual analysis cannot reveal.

Influence Maximisation Research

Doctoral study examines selecting nodes to maximise spread through a network. Influence problems arise in marketing, vaccination and communication.

Information Diffusion Modelling

Research examines how information propagates across network structure. Diffusion modelling explains both beneficial and harmful spreading processes.

Graph Ranking Methods

Doctoral work examines ordering nodes by their importance or relevance. Ranking on graphs underpins web search and citation analysis systems.

Recommendation On Graphs

Research examines suggesting items using interaction graph structure. Graph methods now underpin most large scale recommendation systems in use.

Collaborative Filtering Research

Doctoral study examines recommendation from patterns of shared preference. Graph formulations capture higher order preference relationships.

Graph Reasoning Research

Research examines drawing inferences by traversing graph structure. Reasoning capability distinguishes retrieval from genuine relational inference.

Multi Hop Reasoning Methods

Doctoral work examines inference requiring several successive relational steps. Multi hop tasks expose limitations of shallow propagation schemes.

Logical Reasoning On Graphs

Research examines combining formal logic with learned graph representations. Logical integration supports verifiable rather than merely plausible inference.

Combinatorial Optimisation Learning

Doctoral study examines learned approaches to hard discrete optimisation. Learned heuristics may outperform hand designed ones on specific instances.

Neural Algorithmic Reasoning

Research examines networks learning to execute classical algorithms. Algorithmic learning tests whether models capture genuine procedural structure.

Graph Algorithm Learning

Doctoral work examines learning classical graph algorithms from examples. Learned algorithms may generalise to problem sizes never seen in training.

Constraint Satisfaction Research

Research examines learned solvers for constraint structured problems. Many practically important problems reduce naturally to constraint satisfaction.

Routing Problem Learning

Doctoral study examines learned approaches to vehicle and network routing. Routing problems are economically significant and computationally hard.

Scheduling Problem Research

Research examines learned scheduling over dependency structured tasks. Scheduling problems arise across manufacturing, computing and logistics.

Graph Reinforcement Learning

Doctoral work examines sequential decision making over graph structured states. Graph representations suit problems with relational state structure.

Multi Agent Graph Learning

Research examines many interacting agents represented as a graph. Graph structure naturally represents which agents can influence one another.

Graph Structure Learning

Doctoral study examines inferring connectivity rather than assuming it given. Many problems supply features without any observed graph structure.

Latent Graph Inference

Research examines discovering hidden relational structure within data. Inferred structure can improve prediction and reveal system organisation.

Causal Discovery On Graphs

Doctoral work examines inferring causal structure from observational data. Causal discovery requires assumptions that observational data cannot verify.

Causal Inference With Graphs

Research examines estimating effects using graphical causal representations. Graphical formalism makes causal assumptions explicit and checkable.

Counterfactual Reasoning On Graphs

Doctoral study examines reasoning about outcomes under changed structure. Counterfactual reasoning supports both explanation and intervention design.

Uncertainty Quantification

Research examines honest confidence estimates from graph model predictions. Interdependent data complicates conventional uncertainty estimation.

Calibration On Graphs

Doctoral work examines whether predicted probabilities match observed frequencies. Graph models are frequently poorly calibrated in practice.

Out Of Distribution Detection

Research examines recognising graph inputs unlike the training data. Detection prevents confident prediction on genuinely unfamiliar structures.

Distribution Shift On Graphs

Doctoral study examines performance when graph properties change over time. Real graphs evolve and violate the assumptions training relied upon.

Robustness To Structure Shift

Research examines stability when connectivity patterns differ from training. Structural shift causes failures that feature shift analysis misses.

Generalisation Theory For Graphs

Doctoral work examines why graph models generalise beyond training data. Standard learning theory assumes independence that graphs plainly violate.

Molecular Property Prediction

Research examines predicting chemical properties from molecular structure. Molecules are naturally graphs and drove much early progress in the field.

Molecular Representation Learning

Doctoral study examines learned encodings of chemical structure. Representation quality determines performance across every downstream task.

Drug Discovery Applications

Research examines graph methods within therapeutic molecule discovery. Graph models now screen candidate molecules at very large industrial scale.

Protein Structure Learning

Doctoral work examines graph representations of protein three dimensional structure. Structural representation underpins recent advances in protein modelling.

Protein Interaction Networks

Research examines networks of interactions between biological molecules. Interaction networks reveal functional organisation within living cells.

Biological Network Analysis

Doctoral study examines graph learning across biological network data. Biological systems are relational and suit graph representation naturally.

Gene Regulatory Network Learning

Research examines inferring how genes regulate one another within cells. Regulatory inference is central to understanding cellular behaviour.

Single Cell Graph Analysis

Doctoral work examines graph methods applied to single cell measurements. Cell similarity graphs underpin most single cell analysis pipelines.

Materials Discovery Applications

Research examines graph learning for identifying materials with target properties. Materials search spaces are far too large for exhaustive exploration.

Crystal Structure Learning

Doctoral study examines graph representations of periodic crystalline solids. Periodicity requires representations that ordinary graphs do not provide.

Catalyst Design Research

Research examines learned prediction of catalytic surface behaviour. Catalyst discovery is central to industrial energy and chemical processes.

Molecular Dynamics Learning

Doctoral work examines learned models of atomic motion over time steps. Learned potentials accelerate simulation by many orders of magnitude.

Quantum Chemistry Applications

Research examines graph models predicting quantum mechanical properties. Learned prediction avoids the extremely costly quantum calculations.

Physics Simulation On Graphs

Doctoral study examines learned simulation of physical interacting systems. Graph models simulate systems where entities interact locally with neighbours.

Fluid And Particle Simulation

Research examines learned simulation of particle and fluid dynamics. Learned simulators run far faster than conventional numerical solvers do.

Mesh Based Simulation Learning

Doctoral work examines learning over computational meshes as graph structures. Mesh learning supports engineering simulation across many domains.

Weather And Climate Modelling

Research examines graph architectures applied to atmospheric forecasting. Graph models now match conventional forecasting at very much lower cost.

Traffic Forecasting Research

Doctoral study examines predicting flow across road network structure. Road networks are naturally graphs with strong spatial dependence patterns.

Transport Network Analysis

Research examines learning over transport and mobility network data. Network analysis supports planning, resilience and travel demand forecasting.

Power Grid Applications

Doctoral work examines graph learning across electrical transmission networks. Grid applications demand reliability that current models rarely guarantee.

Telecommunications Network Learning

Research examines graph methods for communication network management. Network operators hold enormous and naturally graph structured datasets.

Supply Chain Network Analysis

Doctoral study examines graph learning across production and distribution networks. Network analysis reveals vulnerabilities that firm level analysis misses.

Financial Network Research

Research examines graph learning over transaction and exposure networks. Financial contagion propagates through network structure very directly.

Social Network Analysis

Doctoral work examines graph learning over human social connection data. Social applications raise privacy concerns that technical work often neglects.

Misinformation Spread Modelling

Research examines propagation of false information through social networks. Spread modelling informs both detection and intervention design.

Epidemic Modelling On Networks

Doctoral study examines disease transmission across contact network structure. Network structure determines outbreak dynamics more than average contact rates.

Healthcare Graph Applications

Research examines graph learning across clinical and health system data. Health data is deeply relational across patients, treatments and outcomes.

Electronic Record Graph Learning

Doctoral work examines graph representations of clinical record data. Graph structure captures relationships that flat representations discard.

Medical Knowledge Graph Research

Research examines structured representations of biomedical knowledge. Knowledge graphs support reasoning across fragmented medical literature.

Brain Network Analysis

Doctoral study examines graph representations of neural connectivity. Brain connectivity is inherently a network and suits graph analysis naturally.

Neuroimaging Graph Methods

Research examines graph learning applied to brain imaging datasets. Graph methods relate imaging findings to underlying connectivity structure.

Code And Program Graph Learning

Doctoral work examines graph representations of computer program structure. Programs have rich graph structure that text representations discard.

Software Analysis Applications

Research examines graph learning for analysing and improving software. Dependency and call structure form naturally graph based representations.

Natural Language Graph Methods

Doctoral study examines graph structures derived from written language. Syntactic and semantic structures form graphs over any textual content.

Scene Graph Research

Research examines graph representations of objects and their relations in images. Scene graphs bridge visual perception and relational reasoning.

Point Cloud Learning

Doctoral work examines learning over unordered sets of spatial points. Point clouds are commonly processed through graph based architectures.

Computer Vision Graph Methods

Research examines graph architectures within visual understanding tasks. Graph structure captures relationships between detected visual elements.

Robotics Graph Applications

Doctoral study examines graph representations within robotic perception and control. Robot bodies and environments both have natural graph structure.

Recommender System Deployment

Research examines operating graph recommendation systems at production scale. Deployment demands latency and reliability that research rarely addresses.

Cybersecurity Graph Applications

Doctoral work examines graph learning for detecting malicious network activity. Attack patterns are structural and evade individual event inspection.

Explainability For Graph Models

Research examines making graph model predictions interpretable to people. Explanation must identify influential structure as well as features.

Attribution Methods On Graphs

Doctoral study examines identifying which nodes and edges drove a prediction. Structural attribution differs fundamentally from feature attribution.

Counterfactual Explanation Research

Research examines explaining predictions through minimal structural changes. Counterfactual explanations indicate what would change an outcome.

Concept Based Explanation

Doctoral work examines explanation using human meaningful structural concepts. Concept explanations are more usable than node level attributions.

Adversarial Attacks On Graphs

Research examines deliberate manipulation causing graph models to fail. Very small structural changes can entirely reverse model predictions.

Structural Attack Defence

Doctoral study examines defending graph models against connectivity manipulation. Defence is essential wherever adversaries can influence graph structure.

Poisoning Attack Research

Research examines corruption of graph data before or during training. Poisoning is realistic wherever graphs are built from open contributions.

Certified Robustness Methods

Doctoral work examines provable guarantees against structural manipulation. Certification provides assurance that empirical testing cannot offer.

Privacy In Graph Learning

Research examines exposure of sensitive information through graph models. Connectivity itself reveals sensitive information about individuals.

Membership Inference On Graphs

Doctoral study examines determining whether records were used in training. Graph connectivity makes membership disclosure unusually easy to achieve.

Differential Privacy For Graphs

Research applies formal privacy guarantees to graph structured data. Protecting connections is substantially harder than protecting isolated records.

Federated Graph Learning

Doctoral work examines training graph models without centralising the data. Graphs frequently span organisations that cannot share their portions.

Fairness In Graph Models

Research examines whether graph models treat groups of nodes equitably. Homophily within networks can concentrate disadvantage very systematically.

Bias Amplification In Networks

Doctoral study examines models reinforcing existing structural inequalities. Recommendation on graphs can entrench segregation within social networks.

Deployment And Adoption Research

Research examines why graph methods are or are not adopted in practice. Adoption lags research because deployment demands are rarely addressed.