Generating spatial embedding by integrating proxy motion and optimizing prediction objectives

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a spatially embedded neural network configured to process data characterizing motion of an agent interacting with an environment to generate a spatially embedded. In one aspect, the method...

Full description

Saved in:
Bibliographic Details
Main Authors BANINO ANDREA, URIA-MARTINEZ BENIGNO, IBARZ GABARDOS BORJA, ZAMBARDI, VITTORIO, BLUNDELL CHARLES
Format Patent
LanguageChinese
English
Published 08.11.2022
Subjects
CALCULATING
COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
COMPUTING
COUNTING
PHYSICS
Online AccessGet full text

Cover

More Information
Summary:Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a spatially embedded neural network configured to process data characterizing motion of an agent interacting with an environment to generate a spatially embedded. In one aspect, the method includes processing data characterizing motion of an agent in an environment in a current time step using a spatial embedding neural network to generate a current spatial embedding of the current time step; determining a predicted score and a target score for each of a plurality of slots in the external memory, where each slot stores: (i) a representation of observations characterizing the environmental state and (ii) a spatial embedding; and determining an update to the value of the set of spatially embedded neural network parameters based on an error between the predicted score and the target score. 用于训练配置为处理表征与环境互动的代理的运动的数据以生成空间嵌入的空间嵌入神经网络的方法、系统和装置，包括编码在计算机存储介质上的计算机程序。在一个方面上，该方法包括：使用空间嵌入神经网络处理表征在当前时间步长中的代理在环境中的
Bibliography:Application Number: CN202180023280