Transductive learning

As explained in Knowledge Graphs and LLMs in Action

Transductive learning learns representations or predictions for the specific nodes and graph structure observed during training rather than a general rule for unseen nodes.

Transductive learning optimizes embeddings or predictions for a fixed set of nodes and the graph structure observed during training. It suits static graphs where new nodes are not expected, and can infer labels or edges among known nodes. section 11.1.1

Figure 11.1 The node embedding problem. The goal of the task is to learn an encoding function, encode(x), that maps nodes to a multidimensional embedding space. These embeddings are optimized so that distances in the embedding space reflect relevant aspects of the original graph, such as the relative positions of the nodes.

The node-embedding view makes the objective concrete: map observed nodes into a multidimensional space whose distances preserve relevant graph relationships. section 11.1.1

Use transductive learning when the graph’s membership is stable:

• Classify unlabeled nodes from patterns learned across partially labeled nodes. section 11.1.1
• Predict new edges between nodes already observed during training. section 11.1.1
• Recommend existing products to existing users from their past interactions. section 11.1.1

Fixed graph, changing relationships

A shallow embedding model can represent members of a known karate club and learn relationships among them. If a new member joins, the lookup table cannot automatically provide an embedding; retraining is required. section 11.3.2

The central trade-off is generalization. Transductive methods can fit the observed graph well but do not learn a reusable rule for unseen nodes. Inductive methods instead learn parametric mappings from node features and can apply them to nodes absent during training. section 11.1.1

How Knowledge Graphs and LLMs in Action covers it

How deep it goes

Defines fixed-node optimization and contrasts it with inductive mappings, but stops short of implementation details.

Examples

Uses partially labeled nodes, known-node edge prediction, e-commerce recommendations, and a karate club membership example.

What it assumes

Assumes graph nodes, edges, embeddings, and training data, while contrasting fixed-node learning with generalization to unseen nodes.

What you can do after

Evaluate whether a graph-learning problem can tolerate retraining when new nodes appear.

How it connects

Connects node embeddings, shallow embeddings, inductive learning, node features, and graph neural networks.

liveBookmanning.com

Read Knowledge Graphs and LLMs in Action on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Optimization Algorithms: AI techniques for design, planning, and control problemsTransductive embedding is a graph embedding technique that learns representations only for entities observed during training.
• 

  liveBookmanning.com
  Graph Neural Networks in ActionA transductive embedding method directly learns representations for nodes in a fixed graph and does not naturally generalize to unseen nodes or graphs.

Compare how 3 books explain transductive learning →

Related concepts

• inductive inferencecontrasts with · Inductive inference is the process of deriving general rules, hypotheses, mappings, or predictions from observed examples or data.
• inductive embeddingalternative to · Inductive embedding is a graph representation technique that learns a function for generating embeddings of previously unseen entities or graphs.
• Node2Vecused by · Node2Vec is a graph representation-learning algorithm that learns node embeddings from biased random walks and their context relationships.
• embeddingkind of · A numerical vector representation of an item or input that encodes features or relationships useful for computational tasks.
• vectorizationkind of · Vectorization is the representation of data or computations as vectors or arrays so they can be processed numerically or collectively.

This explainer was generated by Manning's knowledge graph from the text of Knowledge Graphs and LLMs in Action. The section links point to the parts of the book it draws on.