Learning theory

This is the curated entrypoint for TorchLean's learning-theory layer. We group the files here around three themes:

• robustness specifications and executable Float diagnostics;
• algorithmic stability definitions, dynamics stability, and a fully proved 1D ridge-regression stability theorem;
• differential privacy definitions and closure lemmas.

The files keep theorem statements explicit about their level. Specification files define Prop-level notions; runtime files compute finite diagnostics; worked examples such as ridge regression prove concrete bounds with explicit assumptions.

References:

• Bousquet and Elisseeff, "Stability and Generalization", JMLR 2002.
• Dwork and Roth, The Algorithmic Foundations of Differential Privacy, 2014.
• Goodfellow, Shlens, and Szegedy, "Explaining and Harnessing Adversarial Examples", ICLR 2015.
• Hardt, Recht, and Singer, "Train faster, generalize better", ICML 2016.