Crate cjc_ad 

Automatic differentiation for CJC.

Provides forward-mode differentiation via dual numbers and reverse-mode differentiation via a computation tape. Supports grad(), jacobian(), and gradient graph construction for ML training loops.

Modules

pinn

Physics-Informed Neural Networks (PINN) and Physics-Informed ML (PIML)

Structs

Dual

Dual number for forward-mode automatic differentiation.

GradGraph

The reverse-mode AD tape/graph.

GradNode

A node in the reverse-mode AD graph.

Enums

GradOp

Operation recorded in the computation graph.

Functions

check_grad_finite_diff

Validate gradient using finite differences.