Manuscript Psy01-A@1 · Psy01-A.tex · section · 94477fa3
Nested pruning and a quantitative learning-induced defect
Stability, hard-mask discontinuity, and schedule dependence › Nested pruning and a quantitative learning-induced defect · lines 1506–1512
Equation (eq:gradientcommutator) is a bound for the specified smooth gradient step. Adam and ReLU optimization require their own augmented-state assumptions; the formula is not an unproved Adam stability theorem.
Nearby in this section
- theorem: Theorem: Fixed-weight nesting and fixed-mask commutation
- section: Nested pruning and a quantitative learning-induced defect
- theorem: Theorem: Ranking stability after an update
- example: Example: An intervening update changes the final support
- proposition: Proposition: A projection–gradient commutator bound