Private Zeroth-Order Nonsmooth Nonconvex Optimization

Authors: Qinzi Zhang, Hoang Tran, Ashok Cutkosky

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical We introduce a new zeroth-order algorithm for private stochastic optimization on nonconvex and nonsmooth objectives. Our algorithm satisfies (α, αρ2/2)-R enyi differential privacy (RDP) (Mironov, 2017) (which is approximately (ρ, γ)-DP) and finds a (δ, ϵ)-stationary point with O(dδ 1ϵ 3+d3/2ρ 1δ 1ϵ 2) data complexity. This paper presents a novel zeroth-order algorithm for private nonsmooth nonconvex optimization.
Researcher Affiliation Academia Qinzi Zhang, Hoang Tran & Ashok Cutkosky Department of Electrical and Computer Engineering Boston University Boston, MA, USA {qinziz,tranhp,cutkosky}@bu.edu
Pseudocode Yes Algorithm 1 Zeroth-order gradient oracle GRADf,δ(x, z1:b), Algorithm 2 Zeroth-order gradient difference oracle DIFFf,δ(x, y, z1:b), Algorithm 3 Online-to-Nonconvex Conversion, Algorithm 4 Private variance-reduced gradient oracle O, Algorithm 5 Tree Mechanism
Open Source Code No The paper does not provide concrete access to source code for the methodology described, nor does it include specific repository links or explicit code release statements.
Open Datasets No The paper focuses on theoretical analysis and algorithm design and does not describe experiments using datasets.
Dataset Splits No The paper is theoretical and does not describe experiments with dataset splits.
Hardware Specification No The paper is theoretical and does not describe experiments, therefore no hardware specifications are mentioned.
Software Dependencies No The paper is theoretical and does not describe experiments, therefore no specific ancillary software dependencies with version numbers are mentioned.
Experiment Setup No The paper focuses on theoretical analysis and algorithm design, and does not provide specific experimental setup details such as hyperparameters or training configurations.