Singular Learning Theory course

A course on Singular Learning Theory, exploring its applications in machine learning and mathematical AI alignment.

The course covers the mathematical foundations of SLT and touches on recent developments in how its tools are used in alignment research. The core focus is the role of degeneracies and singularities in governing the behaviour and training dynamics of neural networks. By the end of the course you will be familiar with:

• how degeneracies appear in neural networks;
• the basics of algebraic and analytic geometry, including blow-ups and resolution of singularities;
• the learning coefficient and its mathematical origins;

• Watanabe's free energy formula and its consequences.

  Prerequisites

• The course aims to be self-contained, but it assumes some mathematical maturity: multivariable calculus, linear algebra, real and complex analysis and introductory probability and statistics. Familiarity with machine learning, algebraic and analytic geometry or integral transforms is helpful but not required.

• The course is aimed at master's students, DPhil students, postdocs and faculty who want a grounding in the fundamentals of SLT.

  Schedule

• All lectures are on Tuesdays, 16:00–18:00.

If you’re interested, take a look at the course announcement for more information.https://kaiogden.net/slt/