Deep Neural Networks: Theory of Frequency Principle Investigated

This paper rigorously investigates the Frequency Principle (F-Principle) in Deep Neural Networks (DNNs), which describes the tendency for DNNs to learn target functions from low to high frequencies during training. The authors provide theorems characterizing the F-Principle at initial, intermediate, and final training stages. These results are general, applying to various network architectures, data distributions, and loss functions, thereby establishing a theoretical foundation for understanding DNN training processes. AI

RANK_REASON Academic paper published on arXiv detailing theoretical investigation of a phenomenon in deep neural networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• arXiv
• Deep Neural Networks
• Zhiqin Xu

• paper
• other

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