sampling_derivatives#
Jet machinery: symmetric directional derivatives of probe/apply/entries, all in one place.
The *_derivatives ops strictly generalize the plain sampling ops with a leading order axis
(order 0 == the plain op). The jets of the helper chain (compute_*_jets,
binomial_combine_tensor) are shared across the three sampling types, which is why the
derivative layer is one module rather than three. Math: docs/symmetric_probe_derivatives.tex;
costs/usage: docs/entries_apply_probe.md.
Functions#
|
Symmetric derivatives of probing a Tucker tensor train, in one repeated direction. |
|
Structural check (hard error): the perturbation |
|
Structural check (hard error): the perturbation |
|
Binomial tensor driving every jet convolution. |
|
Input jets: stack each (value, direction) pair on a leading order axis. |
|
Left derivative-pushthrough jets. |
|
Left derivative-pushthrough jets (standard fused-recurrence form). |
|
Right derivative-pushthrough jets. |
|
Right derivative-pushthrough jets (standard recurrence form). |
|
Combine the left and right jets at each free mode via the binomial jet-product. |
|
Combine the left and right jets at each free mode (standard order-scan form). |
|
Lift the combined jets back to the ambient modes through the Tucker cores (order by order). |
|
The apply-derivative frame sweep (lean): the up-index jet |
|
The entries-derivative frame sweep (lean): like |
|
The probe-derivative frame sweep (full): the jet-ified twin of |
|
Variation-leftward edge-variable jets sigma (the jet-ified Algorithm-7 sigma recursion). |
|
Variation-rightward edge-variable jets tau -- the mirror of |
|
Variation-downward edge-variable jets deta (the jet-ified Algorithm-7 deta combine). |
|
Tangent combine at each free mode (standard order-scan form). |
|
Variation-leftward edge-variable jets sigma (standard banded-recurrence form). |
|
Variation-rightward edge-variable jets tau (standard banded-recurrence form) -- sigma-banded on |
|
Assemble tangent-probe-derivative jets: |
|
Variation half of |
|
Symmetric derivatives of probing a tangent vector, in one repeated direction (Riemannian J^(s)). |
|
Symmetric derivatives of applying a Tucker tensor train in all modes, in one repeated direction. |
|
Variation half of |
|
Symmetric derivatives of applying a tangent vector in all modes, in one repeated direction. |
|
Symmetric derivatives of an entry of a Tucker tensor train, in one repeated direction. |
|
Variation half of |
|
Symmetric derivatives of an entry of a tangent vector, in one repeated direction. |
|
Adjoint-up edge-variable jets: |
|
Adjoint-var-rightward edge-variable jets (jet-ified probing.compute_tau_tilde). |
|
Adjoint-var-leftward edge-variable jets -- the mirror (reverse) of compute_tau_tilde_jets_trs. |
|
Adjoint-var-leftward edge-variable jets tau_tilde (standard order-scan form). Dense reference: |
|
Adjoint-var-rightward edge-variable jets sigma_tilde (standard order-scan form; reverse of |
|
Adjoint-var-down edge-variable jets (jet-ified probing.compute_dxi_tilde): two adjoint-hooked |
|
Assemble Tucker-core variation gradients (the 1-edge, plain-order-sum case): |
|
Assemble the Tucker variation gradient (standard W-chunked form; dense reference |
|
Assemble TT-core variation gradients (the 3-edge, trs case): three order-less trs outer products |
|
Assemble the TT variation gradient (standard W-chunked form; dense reference |
|
A memory-balanced |
|
The largest |
|
Variation gradient of |
|
Transpose of |
|
Propagation-only adjoint sweep via Q, seeded at the terminal bond by the residual jet |
|
Variation gradient of |
|
Transpose of |
|
Variation gradient of |
|
Transpose of |
|
Corewise (non-manifold) transpose of |
|
Corewise transpose of |
|
Corewise transpose of |
|
Exact dense symmetric probe derivatives, by the multilinear subset expansion (test oracle). |
|
Exact dense symmetric apply derivatives, by the all-modes multilinear subset expansion (oracle). |
|
Exact dense symmetric entry derivatives (oracle): apply-derivatives with one-hot basis vectors |