BriefGPT.xyz
Jun, 2020
探索层次化学习:神经表征的好处
Towards Understanding Hierarchical Learning: Benefits of Neural Representations
HTML
PDF
Minshuo Chen, Yu Bai, Jason D. Lee, Tuo Zhao, Huan Wang...
TL;DR
本文指出了中间神经表现添加了深度学习网络的灵活性并且在原始输入上具有优势,并阐述了与浅学习者,如卷积内核的神经表现的关系。通过学习低秩的多项式,中间神经表现可以实现比原始输入更少的样本复杂度,并且在神经可切内核的情况下,本文还提出了神经表现不利的限制。
Abstract
deep neural networks
can empirically perform efficient
hierarchical learning
, in which the layers learn useful representations of the data. However, how they make use of the intermediate representations are not e
→