Granular neural network architecture search over low level primitives

• Main Authors: MANKE WOJCIECH ANDRZEJ, SHAZEER NOAM M, LIU HANXIAO, SOE DAVID RICHARD, LE QUOC V, DAI ZIHANG
• Format: Patent
• Language: Chinese; English
• Published: 24.11.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining a neural network architecture. One of the methods includes receiving initial neural network architecture data; generating search space data defining a plurality of sub-model architectures from the initial neural network architecture data, each sub-model architecture comprising an ordered set of primitive neural network operations, each primitive neural network operation associated with one or more operating parameters; and determining a final architecture of the neural network for performing the machine learning task, the method includes running an evolutionary architecture search algorithm on the search space data to identify a respective optimization value for each of one or more operating parameters of a primitive neural network operation in at least one sub-model mechanism of the plurality of sub-model architectures. 用于确定神经网络架构的方法、系统和装置，包括在计算机存储介质上编码的计算机程序。所述方法之一包括接收初始神经网络架构数据；根据初始神经网络架构数据生成定