2024AI安全的关键概念机器学习中可靠的不确定性量化方法分析报告 英文版.docxVIP

Thispaperisthefifthinstallmentinaserieson“AIsafety,”anareaofmachinelearningresearchthataimstoidentifycausesofunintendedbehaviorinmachinelearningsystemsanddeveloptoolstoensurethesesystemsworksafelyandreliably.OtherpapersintheseriesdescribethreecategoriesofAIsafetyissues—problemsofrobustness,assurance,andspecification.Thispaperintroducestheideaofuncertaintyquantification,i.e.,trainingmachinelearningsystemsthat“knowwhattheydon’tknow.”Introduction

Thispaperisthefifthinstallmentinaserieson“AIsafety,”anareaof