Singular Learning Theory (SLT)

What is SLT?#

Singular Learning Theory (SLT) is a mathematical framework developed by Sumio Watanabe for analyzing statistical learning when the model is singular (non-regular).

Key Concepts#

  • RLCT (Real Log Canonical Threshold): A key quantity that determines generalization error
  • Free Energy: Measures model complexity in Bayesian inference
  • Singular Models: Models where the Fisher information matrix is degenerate

References#

  • Watanabe, S. (2009). Algebraic Geometry and Statistical Learning Theory
  • Watanabe, S. (2018). Mathematical Theory of Bayesian Statistics