New research details complexity of synthesizing ReLU networks from parameters

A new arXiv paper explores the computational complexity of synthesizing ReLU networks when parameter values are provided. The research investigates how difficult it is to construct a network with a given set of parameters that achieves a specific target behavior. The paper details that matching two labeled examples is already NP-complete, and further complexities arise when the target must hold universally or on average, with specific problems being Sigma_2^P-complete or DP-complete depending on the neuron count and input domain. AI

IMPACT This research contributes to understanding the theoretical limits of neural network construction, potentially informing future algorithm design.

RANK_REASON Academic paper on computational complexity of neural network synthesis. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Dimitrios Myrisiotis
• $DP$
• $NP^{PP}$
• ReLU network
• $Sigma_2^P$

• paper
• other

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