Neural Network Verification Complexity in Quantized Settings Explored

Researchers have analyzed the computational complexity of verifying feedforward neural networks (FNNs) when using quantized settings. They categorized FNNs into rational, quantized, and dynamically quantized types, and considered both linear programming (LP) and bit-vector (BV) specifications. For quantized FNNs with fixed precision, verification remains NP-complete for both LP and BV specifications. For dynamically quantized FNNs with BV specifications, new upper bounds were established, building on prior PSPACE-hardness findings. AI

RANK_REASON This is a research paper detailing theoretical computational complexity findings related to neural networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →