AI 용어집
인공지능 완전 사전
Message Passing Neural Networks
Fundamental paradigm of graph neural networks where nodes exchange information with their neighbors through successive iterations to learn representations.
Node Embedding
Dense, low-dimensional vector representation of a node in a graph, capturing its structural and semantic features.
Edge Features
Attributes or features associated with edges in a graph, used to enrich the message passing process between nodes.
Graph Convolution
Information propagation operation on graphs that generalizes traditional convolution to irregular data structures.
Graph Attention Networks
GNN architecture using attention mechanisms to dynamically learn edge weights during message passing.
Heterogeneous Graph Neural Networks
Extension of GNNs to handle graphs containing multiple types of nodes and edges with different relationships.
Temporal Graph Networks
GNN models designed to capture the dynamic evolution of graphs over time with continuous updates.
Readout Function
Function that aggregates final node representations to produce a graph-level prediction.
Graph Generation
Process of synthetic graph creation using generative models based on neural networks.
Graph Isomorphism Networks
Powerful GNN architecture approximating the Weisfeiler-Lehman isomorphism test for maximal graph discrimination.
Message Function
Function that computes the messages to send from a node to its neighbors using their latent states and edge features.
Update Function
Function that combines the current state of a node with messages received from its neighbors to produce the new latent state.
Aggregate Function
Mathematical operation that combines multiple messages into a single representation, typically sum, mean or max.