5
Regularization
An Original Contribution
Standard ML view: regularization as prior injection, geometric regularization, spectral regularization, generalization bounds. Thesis view: regularization as dynamical modification (and why it is not bias), parameter-space and objective-space techniques, the cosine schedule and two-stage protocol, and empirical behavior under shot noise.
Lessons
5.1
Regularization as Prior Injection
5.2
Geometric Regularization
5.3
Spectral Regularization of Parameterized Circuits
5.4
Regularization and Generalization Bounds
5.5
Regularization as Dynamical Modification (and Why It Is Not Bias)
5.6
Parameter-Space Regularization Techniques
5.7
Objective-Space Regularization Techniques
5.8
The Cosine Schedule and Two-Stage Protocol
5.9
Empirical Behavior Under Shot Noise
Guided instruction and research mentorship are offered separately → hbar.work