• ☰
  • Home
  • Introduction
  • An overview
  • Vectorization approaches
  • Factorization
  • Neural embeddings
  • Bayesian clustering
  • Relational approaches
  • Hybrid approaches
  • Related approaches
  • Relational Learning Primer
  • Deep Learning Primer

Combining relational and deep learning has yielded several distinct ideas. We will systematically compare them long several dimensions:

• supervision: are representations learned in a supervised or unsupervised way?
• preservation of relational information: to which extent is the relational information preserved?
• declarativness: how easy is it to explain latent features?
• scalability

By far the largest category are the vectorization approaches. The core idea behind these approaches is to map relational entities to a low-dimensional vector spaces. Thus, each entity becomes a vector, while their relationships are represented with matrices.

Next chapter: Vectorization approaches