arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1908.02831cs.LGcs.NEstat.ML
收藏

可视化神经网络的PHATE

Visualizing the PHATE of Neural Networks

Yale University(耶鲁大学);Princeton University(普林斯顿大学);University of California, San Diego(加利福尼亚大学圣迭戈分校)

Scott Gigante, Adam S. Charles, Smita Krishnaswamy, Gal Mishne

摘要:

理解某些神经网络为什么以及如何优于其他网络是指导网络架构和优化方法未来发展的关键。为此,我们引入了一种新的可视化算法,揭示了这种网络的内部几何结构:Multislice PHATE(M-PHATE),这是第一种明确设计用于可视化神经网络隐藏的数据表示在整个训练过程中如何演变的方法。我们证明,我们的可视化提供了直观的,详细的总结学习动态超越简单的全球措施(即,验证损失和准确性),而无需访问验证数据。此外,与基于标准降维方法(例如,ISOMAP,t-SNE)。我们用两个小插曲来演示M-PHATE:持续学习和泛化。在前者中,M-PHATE可视化显示了“灾难性遗忘“的机制,这是任务切换背景下学习的一个主要挑战。在后者中,我们的可视化揭示了隐藏单元之间增加的异质性如何与提高的泛化性能相关。M-PHATE的一个实现,以及复制本文中的图形的脚本,可以在https://github.com/scottgigante/M-PHATE上找到。

英文摘要:

Understanding why and how certain neural networks outperform others is key to guiding future development of network architectures and optimization methods. To this end, we introduce a novel visualization algorithm that reveals the internal geometry of such networks: Multislice PHATE (M-PHATE), the first method designed explicitly to visualize how a neural network's hidden representations of data evolve throughout the course of training. We demonstrate that our visualization provides intuitive, detailed summaries of the learning dynamics beyond simple global measures (i.e., validation loss and accuracy), without the need to access validation data. Furthermore, M-PHATE better captures both the dynamics and community structure of the hidden units as compared to visualization based on standard dimensionality reduction methods (e.g., ISOMAP, t-SNE). We demonstrate M-PHATE with two vignettes: continual learning and generalization. In the former, the M-PHATE visualizations display the mechanism of "catastrophic forgetting" which is a major challenge for learning in task-switching contexts. In the latter, our visualizations reveal how increased heterogeneity among hidden units correlates with improved generalization performance. An implementation of M-PHATE, along with scripts to reproduce the figures in this paper, is available at https://github.com/scottgigante/M-PHATE.

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。

↑