Manuscript Psy03-D@1 · Psy03-D-v1.tex · section · 3e18951d

Stored and effective parameters

Masked gradients and reference-preserving repair › Stored and effective parameters · lines 1306–1311

Let $W$ be a stored weight tensor, $M$ a fixed binary mask, and $B = W \circ M$ the tensor used in the forward computation. Assume the loss depends on $W$ only through $B$ .

Nearby in this section

theorem: Theorem: Masked-gradient obstruction
section: Stored and effective parameters
proposition: Proposition: Ambient effective-weight scores remain available
section: Stored and effective parameters