Semantic Networks

Topics Lexicon · Structure · Semantics · Represent & Reason · 2×1 Matrix · Guards & Prisoners · Weights
01 //

Overview

Semantic networks are our first knowledge representation. We use them to represent problems and reason over them—especially for visual analogy tasks like Raven's Matrices.

Represent → Reason
Represent Reason Solve

Encode knowledge & problem → Use it to find the answer

Knowledge representation

How we encode what we know so an agent can use it.

Problem-solving

Semantic nets work with Generate and Test, Means-Ends, Problem Reduction.

Analogy

Core use: A is to B as C is to ? — complete the pattern.

02 //

Structure of Semantic Networks

Every semantic network has three parts:

Lexicon

Nodes — objects or concepts (e.g. shapes, entities). The vocabulary.

Structure

Directed links — how nodes connect. Relationships between objects.

Semantics

Labels — what each link means (e.g. inside, above). Enables inference.

Example

For a visual analogy: nodes = objects (circle, square, triangle); links = spatial relations (inside, above); labels describe transformations between frames (unchanged, expanded, deleted).

03 //

Good Representations

What makes a representation useful?

04 //

2×1 Matrix Problems

A is to B, as C is to ? Build semantic networks for A→B and C→D. Pick the D that best matches the A→B transformation.

Approach
  1. 1. Represent A: nodes for each object, links for relations (inside, above, etc.)
  2. 2. Represent B: same structure, note transformations (unchanged, expanded, deleted)
  3. 3. Represent C and each candidate D
  4. 4. Pick the D whose net matches A→B best
Transformation vocabulary

Unchanged — same object, same relation. Expanded — object grows or multiplies. Deleted — object removed. Use consistently across A→B and C→D.

05 //

Guards & Prisoners

Classic state-space problem (also: cannibals & missionaries). Semantic nets model states and moves.

Why it matters

The network makes illegal and unproductive moves obvious. From initial state, ~5 moves are possible; only ~2 are both legal and productive. The representation exposes this without search.

06 //

Weights for Matching

When multiple candidates could fit, assign weights to transformations to prefer more specific or better matches.

Sum weights across transformations. Best D = highest total.

07 //

Cognitive Connection

Semantic networks mirror theories of human memory—especially spreading activation.

Spreading activation

Activating one node spreads to connected nodes. Explains why "John wanted to get rich. He got a gun." leads you to infer robbery—activation from rich and gun merges on paths through the net.

08 //

Summary