New SWAP-Score metric evaluates neural networks without training

Researchers have introduced SWAP-Score, a novel zero-shot metric designed to evaluate neural networks without requiring training. This method measures a network's expressivity using sample-wise activation patterns and demonstrates strong predictive performance across various architectures, including CNNs and Transformers. SWAP-Score significantly outperforms existing metrics in computer vision and natural language processing tasks, showing high correlations with ground-truth performance and enabling faster neural architecture search. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Enables faster and more accurate neural architecture search by reducing computational overhead in model evaluation.

RANK_REASON The cluster contains an academic paper introducing a new method for evaluating neural networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• SWAP-Score
• Sample-Wise Activation Patterns
• CNNs
• Transformers
• CIFAR-10
• DARTS
• FlexiBERT
• GLUE
• ImageNet

• paper
• other