Robotics & Machine Learning Daily News2024,Issue(Mar.13) :117-120.

Patent Issued for System and method for improved generalization from concept con strained dreams (USPTO 11907815)

Robotics & Machine Learning Daily News2024,Issue(Mar.13) :117-120.

Patent Issued for System and method for improved generalization from concept con strained dreams (USPTO 11907815)

扫码查看

Abstract

From the background information supplied by the inventors, news correspondents o btained the following quote: “BACKGROUND OF INVENTION”Supplementing the background information on this patent, NewsRx reporters also o btained the inventors’ summary information for this patent: “The present inventi on relates to a system for training an agent to handle unexpected situations, an d more particularly, to a system for training an agent to handle unexpected situ ations using concept constrained dreaming. The system comprises one or more proc essors and a non-transitory computer-readable medium having executable instructi ons encoded thereon such that when executed, the one or more processors perform multiple operations. The system extracts a set of concepts from an agent’s exper iences of an environment, the environment having associated semantics and dynami cs. The extracted concepts are consolidated into an episodic world model. Using the episodic world model, a dream sequence of prospective simulations comprising a selected set of concepts constrained by the environment’s semantics and dynam ics is generated. The sensor data format of the dream sequence is used for augme nted training of the agent to operate in the environment with improved generaliz ation to unanticipated changes in the environment.

Key words

Business/Cyborgs/Emerging Technologies/HRL Laboratories LLC/Machine Learning

引用本文复制引用

出版年

2024
Robotics & Machine Learning Daily News

Robotics & Machine Learning Daily News

ISSN:
段落导航相关论文