Research map / Learning theory and algorithms / Combinatorial and search optimization

On the Scalability and Memory Efficiency of Semidefinite Programs for Lipschitz Constant Estimation of Neural Networks

ICLR 2024 · Learning theory and algorithms › Combinatorial and search optimization · approach: Semidefinite relaxations for verification

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Studies how to make semidefinite-program (SDP) Lipschitz constant estimation for neural networks scalable and memory-efficient, addressing the memory and speed limits of standard SDP solvers.

Problem

Standard SDP-based Lipschitz constant estimation for neural networks is memory-intensive and slow, limiting its scalability despite yielding less conservative bounds than weight-norm products.

Approach

Abstract only states the motivation: SDPs give tighter polynomial-time Lipschitz bounds, but standard SDP algorithms consume too much memory and time; the paper aims to improve scalability and memory efficiency of this approach. Specific technique not stated in the provided text.

Closest papers by meaning

• A Quantitative Geometric Approach to Neural-Network Smoothness (NeurIPS 2022 · similarity 0.93)
• Efficient and Accurate Estimation of Lipschitz Constants for Deep Neural Networks (NeurIPS 2019 · similarity 0.93)
• Lipschitz constant estimation of Neural Networks via sparse polynomial optimization (ICLR 2020 · similarity 0.92)
• ECLipsE: Efficient Compositional Lipschitz Constant Estimation for Deep Neural Networks (NeurIPS 2024 · similarity 0.92)
• Novel Quadratic Constraints for Extending LipSDP beyond Slope-Restricted Activations (ICLR 2024 · similarity 0.89)
• A Rescaling-Invariant Lipschitz Bound Based on Path-Metrics for Modern ReLU Network Parameterizations (ICML 2025 · similarity 0.87)