New research explores symmetry transfer in neural network parameters

A new research paper explores the concept of symmetry within trained neural networks, investigating whether learned symmetries in function space can be translated to parameter space. The study proposes a method using parameter-wise functional sensitivity to identify directions that realize group actions in function space. Findings indicate that while these directions can induce predicted function-space motion locally, they may diverge after training, suggesting a complex relationship between parameterization and learned symmetries. AI

影响 This research could lead to a deeper understanding of how neural networks learn and represent symmetries, potentially influencing future model architectures and training methodologies.

排序理由 The cluster contains a single academic paper published on arXiv. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.LG 阅读 →

• arXiv
• Hugging Face
• machine learning
• Parameter-Level Attribution of Symmetry in Trained Networks Though Parameter-Wise Functional Sensitivity

• 论文
• 模型发布

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →