torch_numopt.curvature

Curvature estimators for second-order optimization.

This package provides classes that approximate or compute the Hessian matrix (and its products) in various ways:

• Exact Hessian (full or block-diagonal)

• Gauss-Newton approximation (full or block)

• Hutchinson diagonal approximation (via random projections)

• Identity (no curvature)

All estimators inherit from CurvatureEstimator and implement the scaling_matrix, hvp, and quadratic_form methods.

Modules

exact_block_hessian	Block-diagonal exact Hessian.
exact_hessian	Exact Hessian computation using torch.func.
gauss_newton_approximation	Gauss-Newton approximation of the Hessian.
gauss_newton_block_approximation	Block-diagonal Gauss-Newton approximation.
hutchinson_diagonal_approximation	Hutchinson diagonal Hessian approximation.
naive_identity	Identity curvature estimator (no curvature).