Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Graph Self-Supervised Learning
Learning paradigm where graph models learn representations without explicit labels by automatically generating supervision tasks from the graph structure itself.
Contrastive Graph Learning
Self-supervised learning approach that learns graph representations by maximizing agreement between different augmented views of the same node or graph while minimizing agreement between different instances.
Graph Masking
Proxy task consisting of randomly masking nodes, edges, or attributes in a graph and training the model to reconstruct the missing elements, thereby forcing the learning of robust representations.
Graph Augmentation
Technique for creating alternative views of a graph through structural or attributive transformations that preserve semantics, used to generate positive pairs in contrastive learning.
Node Representation Learning
Process of learning dense vectors (embeddings) that capture the structural and semantic characteristics of nodes in a low-dimensional vector space.
Edge Representation Learning
Learning embeddings for graph edges that encode relationships and interactions between nodes, crucial for link prediction and edge classification.
Graph Pretext Tasks
Artificially constructed tasks derived from the graph structure itself, serving as training signals for self-supervised learning before transfer to downstream tasks.
Graph Context Prediction
Self-supervised task consisting of predicting the structural context of a node (such as its k-hop neighbors) or global properties of the graph from local representations.
Graph Clustering
Unsupervised proxy task used for representation learning where nodes are grouped into clusters based on their structural or attributional similarity in the embedding space.
Graph Attention Mechanisms
Mechanisms that dynamically compute attention weights for neighbors during information aggregation, allowing the model to focus on the most relevant parts of the neighborhood.
Graph Autoencoders
Encoding-decoding models that learn compressed representations of graphs by reconstructing the original graph structure or attributes, often used for self-supervised learning.
Graph Message Passing
Fundamental paradigm where nodes exchange and aggregate information with their neighbors through multiple iterations, enabling information propagation across the graph structure.