Supply Chain Network Topology

A graph model representing the multi-tiered structure of a supply chain, connecting suppliers, central warehouses, regional hubs, and retail nodes. Unlike simple linear chains, multi-echelon graphs capture cross-tier dependencies where a disruption at a Tier-2 supplier can cascade through multiple downstream echelons. These models are essential for inventory optimization and risk propagation analysis.

• Nodes represent distinct echelons: suppliers, factories, DCs, retailers
• Edges encode material flow, lead times, and capacity constraints
• Enables holistic what-if simulation across the entire network

The process of jointly learning an optimal graph topology and node representations from data when the underlying graph is noisy, incomplete, or absent. In supply chains, explicit connections may be undocumented or outdated. Structure learning infers latent dependencies—such as hidden supplier-sub-supplier relationships—from shipment histories, lead time correlations, and quality incident patterns.

• Discovers missing links in incomplete supply chain maps
• Refines noisy or outdated BOM relationships
• Uses techniques like iterative refinement and bilevel optimization

A process that propagates information across a graph using a diffusion kernel, often modeled via Personalized PageRank to capture long-range dependencies. In supply chain topology, diffusion models how a disruption signal—such as a factory shutdown—attenuates or amplifies as it travels through multiple echelons. This is critical for cascading failure analysis.

• Captures influence beyond immediate 1-hop neighbors
• Personalized PageRank biases propagation toward specific node types
• Models the temporal spread of inventory shortages or quality defects

The principle that connected nodes in a graph tend to share similar features or class labels. In supply chain networks, homophily manifests when geographically proximate suppliers share similar lead time profiles, regulatory environments, or disruption risks. High homophily improves GNN performance; low homophily (heterophily) requires specialized architectures.

• Assumption: connected suppliers exhibit correlated risk profiles
• Violated when a critical supplier is in a dissimilar region
• Influences choice between GCN, GAT, or heterophily-specific models

A vector representation that captures the absolute or relative position of a node within a graph structure. In supply chain topology, positional encoding distinguishes between a node's functional role (e.g., Tier-1 vs. Tier-3 supplier) and its structural role (e.g., bottleneck node vs. leaf node). Essential for Transformer-based graph models that lack inherent structural awareness.

• Laplacian eigenvectors encode global node position
• Random walk encodings capture local structural roles
• Enables differentiation between structurally similar but functionally distinct nodes

The resilience of graph neural network predictions against adversarial perturbations, including structural attacks that add or delete edges to cause misclassification. In supply chain contexts, an attacker might manipulate the topology by falsifying supplier relationships, leading to erroneous risk assessments or suboptimal routing decisions. Robustness evaluation is critical for production deployment.

• Edge perturbation attacks simulate falsified supplier connections
• Node injection attacks introduce phantom facilities into the graph
• Defenses include adversarial training and graph structure denoising