研究人员已定量证明了深度神经网络前向传播与重整化群（RG）流之间的类比。他们对MLP残差网络的研究表明，残差流的有效秩随深度增加而减小，这表明无关数据被逐步整合。这种秩崩溃是选择性的，取决于输入分布的相关长度，并且网络仅保留相关的自由度。研究结果表明，MLP实现了一个由输入光谱结构决定的选择性粗粒化过程，网络的大部分运行接近一个不动点。 AI
影响 为理解MLP如何处理信息提供了一个量化框架，可能指导未来的架构设计。
排序理由 这是一篇详细介绍MLP网络内部工作机制新发现的研究论文。
在 Hugging Face Daily Papers 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
The analogy between deep neural network forward passes and renormalization group (RG) flows has been repeatedly noted in the literature, but existing treatments remain qualitative: depth is described as a coarse-graining scale, attention is likened to a partition function, and re…
arXiv:2606.10324v1 Announce Type: cross Abstract: The analogy between deep neural network forward passes and renormalization group (RG) flows has been repeatedly noted in the literature, but existing treatments remain qualitative: depth is described as a coarse-graining scale, at…
The analogy between deep neural network forward passes and renormalization group (RG) flows has been repeatedly noted in the literature, but existing treatments remain qualitative: depth is described as a coarse-graining scale, attention is likened to a partition function, and re…