Introduces hybrid noise and novel coupling analysis to achieve the first convergent hidden-state DP bound for zeroth-order optimization.
Then w⊥ 2 2 = R2 z2 1 Pd j=2 z2 j Pd j=1 z2 j 2 ≤ R2 z2 1 Pd j=1 z2 j Pd j=1 z2 j 2 = r2 z2 1Pd j=1 z2 j ≜ B, where B ∼ Beta 1 2 , d−1 2
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Privacy Amplification in Differentially Private Zeroth-Order Optimization with Hidden States
Introduces hybrid noise and novel coupling analysis to achieve the first convergent hidden-state DP bound for zeroth-order optimization.