摘要：
Informally,wedenethebiasofamodeltobetheexpected generalizationerrorevenifweweretofitittoavery(say,innitely)large trainingset.Thus,fortheproblemabove,thelinearmodelsuersfromlarge bias,andmayundert(i.e.,failtocapturestructureexhibitedby)thedata. Apartfrombias,there'sasecondcomponenttothegeneraliza 阅读全文