arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2306.03440cs.LG
收藏

量化神经网络的变异性坍缩

Quantifying the Variability Collapse of Neural Networks

Tsinghua University(清华大学)

Jing Xu, Haoxiong Liu

AI总结:

针对神经坍缩中类内特征变化难以度量的问题,提出 Variability Collapse Index(VCI),并证明其与线性探测损失相关且具备不变性和数值稳定性,可用于评估预训练网络的变异性坍缩与可迁移性。

摘要:

近期研究从经验上证明了神经网络的可迁移性与最后一层特征的类内变化之间存在正相关关系。最近发现的 Neural Collapse(NC,神经坍缩)现象为理解神经网络的这种最后一层几何结构提供了新视角。本文提出一种名为 Variability Collapse Index(VCI,变异性坍缩指数)的新指标,用于量化 NC 范式中的变异性坍缩现象。VCI 指标具有充分动机,并与最后一层特征上的线性探测损失存在内在联系。此外,它具备理想的理论和经验性质,包括在可逆线性变换下的不变性以及数值稳定性,这使其区别于以往指标。我们的实验验证了 VCI 能够指示预训练神经网络的变异性坍缩和可迁移性。

英文摘要:

Recent studies empirically demonstrate the positive relationship between the transferability of neural networks and the within-class variation of the last layer features. The recently discovered Neural Collapse (NC) phenomenon provides a new perspective of understanding such last layer geometry of neural networks. In this paper, we propose a novel metric, named Variability Collapse Index (VCI), to quantify the variability collapse phenomenon in the NC paradigm. The VCI metric is well-motivated and intrinsically related to the linear probing loss on the last layer features. Moreover, it enjoys desired theoretical and empirical properties, including invariance under invertible linear transformations and numerical stability, that distinguishes it from previous metrics. Our experiments verify that VCI is indicative of the variability collapse and the transferability of pretrained neural networks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑