New algorithm connects independently trained neural network modes

Researchers have developed a novel empirical algorithm to establish continuous low-loss paths between independently trained neural network models, a phenomenon known as mode connectivity. This new method demonstrates broader applicability than previous techniques, successfully connecting a wider range of architectures including MobileNet, EfficientNet, and Compact Convolutional Transformers (CCT). The algorithm also provides more consistent connectivity paths and supports linking modes trained with different hyperparameters. AI

RANK_REASON This is a research paper detailing a new algorithm for neural network mode connectivity. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Compact Convolutional Transformers
• Deep Layer Aggregation
• EfficientNet
• MobileNet
• ResNet
• ShuffleNet
• Yongding Tian

• paper
• model release

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