paper-with-me

홈 › Papers

Understanding the wiring evolution in differentiable neural architecture search

2020-09-02 · Sirui Xie, Shoukang Hu, Xinjiang Wang, Chunxiao Liu, Jianping Shi, Xunying Liu, Dahua Lin

Controversy exists on whether differentiable neural architecture search methods discover wiring topology effectively. To understand how wiring topology evolves, we study the underlying mechanism of several existing differentiable NAS frameworks. Our investigation is motivated by three observed searching patterns of differentiable NAS: 1) they search by growing instead of pruning; 2) wider networks are more preferred than deeper ones; 3) no edges are selected in bi-level optimization. To anatomize these phenomena, we propose a unified view on searching algorithms of existing frameworks, transferring the global optimization to local cost minimization. Based on this reformulation, we conduct empirical and theoretical analyses, revealing implicit inductive biases in the cost's assignment mechanism and evolution dynamics that cause the observed phenomena. These biases indicate strong discrimination towards certain topologies. To this end, we pose questions that future differentiable methods for neural wiring discovery need to confront, hoping to evoke a discussion and rethinking on how much bias has been enforced implicitly in existing NAS methods.

📄 PDF Abstract BibTeX arXiv:2009.01272

Code (1)

SNAS-Series/SNAS-Series 공식 구현 pytorch

Tasks

global-optimizationNeural Architecture Search

Methods 이 논문이 사용한 방법론

Differentiable NAS 설명 없음

Similar Papers 제목 키워드 기반

A Deep Graph Neural Networks Architecture Design: From Global Pyramid-like Shrinkage Skeleton to Local Topology Link Rewiring

2020-12-16 · Gege Zhang

Expressivity plays a fundamental role in evaluating deep neural networks, and it is closely related to understanding the limit of performance improvement. In this paper, we propose a three-pipeline training framework bas…

ClusteringNode Classification

Differentiable Dynamic Wirings for Neural Networks

2021-01-01 · ICCV 2021 10 · Kun Yuan, Quanquan Li, Shaopeng Guo, Dapeng Chen 외

A standard practice of deploying deep neural networks is to apply the same architecture to all the input instances. However, a fixed architecture may not be suitable for different data with high diversity. To boost t…

object-detectionObject Detection

A Deep Graph Neural Networks Architecture Design: From Global Pyramid-like Shrinkage Skeleton to Local Link Rewiring

2021-01-01 · Gege Zhang, Gangwei Li, Weining Shen, Huixin Zhang 외

Expressivity plays a fundamental role in evaluating deep neural networks, and it is closely related to understanding the limit of performance improvement. In this paper, we propose a three-pipeline training framework ba…

ClusteringNode Classification

CP-CNN: Core-Periphery Principle Guided Convolutional Neural Network

2023-03-27 · Lin Zhao, Haixing Dai, Zihao Wu, Dajiang Zhu 외

The evolution of convolutional neural networks (CNNs) can be largely attributed to the design of its architecture, i.e., the network wiring pattern. Neural architecture search (NAS) advances this by automating the search…

Neural Architecture Search

DARTS: Differentiable Architecture Search

2018-06-24 · ICLR 2019 5 · Hanxiao Liu, Karen Simonyan, Yiming Yang

This paper addresses the scalability challenge of architecture search by formulating the task in a differentiable manner. Unlike conventional approaches of applying evolution or reinforcement learning over a discrete and…

General Classificationimage-classificationImage ClassificationLanguage Modeling+5