arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2403.07965cs.LGcs.AI
收藏

神经网络中的条件计算:原理与研究趋势

Conditional computation in neural networks: principles and research trends

Sapienza University of Rome(罗马第一大学);KU Leuven(鲁汶大学);University of Edinburgh(爱丁堡大学)

Simone Scardapane, Alessandro Baiocchi, Alessio Devoto, Valerio Marsocci, Pasquale Minervini, Jary Pomponi

AI总结:

本文综述神经网络条件计算的统一形式化与三类实现(MoE、token选择、早退网络),并分析其在效率、可解释性和迁移学习上的优势及新兴应用。

摘要:

本文总结了将“条件计算”方法应用于神经网络设计这一新兴领域的原理与思想。我们特别关注能够根据输入条件性地动态激活或停用其部分计算图的神经网络。示例包括对输入 token、层(或层集合)以及每层内部子模块(如卷积滤波器中的通道)等进行动态选择。我们首先给出一个通用形式化框架,以统一方式描述这些技术。随后,介绍这些原理的三种重要实现:mixture-of-experts(MoE,混合专家)网络、token 选择机制以及 early-exit(早退)神经网络。本文旨在为这一不断发展的领域提供类似教程的入门介绍。为此,我们分析这些模块化设计在效率、可解释性和迁移学习方面的优势,并重点关注从自动化科学发现到语义通信等新兴应用领域。

英文摘要:

This article summarizes principles and ideas from the emerging area of applying \textit{conditional computation} methods to the design of neural networks. In particular, we focus on neural networks that can dynamically activate or de-activate parts of their computational graph conditionally on their input. Examples include the dynamic selection of, e.g., input tokens, layers (or sets of layers), and sub-modules inside each layer (e.g., channels in a convolutional filter). We first provide a general formalism to describe these techniques in an uniform way. Then, we introduce three notable implementations of these principles: mixture-of-experts (MoEs) networks, token selection mechanisms, and early-exit neural networks. The paper aims to provide a tutorial-like introduction to this growing field. To this end, we analyze the benefits of these modular designs in terms of efficiency, explainability, and transfer learning, with a focus on emerging applicative areas ranging from automated scientific discovery to semantic communication.

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。

↑