The p erceptro n algo rithm ceases to update the para meters only when all the training

images ar e classified corr e ctly (no mistakes, no update s). So, if the training images are

p ossible to clas sify co rrectly with a linear c la ss ifier, will the p erceptr on al gorithm find such

a classifier? Yes , it does, and it w ill co nverge to such a classifier in a finite numb er of

updates (mistakes ). We’ll sho w this in lectur e 2.-The　p erceptro n　algo rithm　 ceases　 to　update　 the　para meters　 only　 when　all　the　training

images　 ar e　classified　corr e ctly　(no　 mistakes,　no　update s).　So,　if　 the　training　images　 are　

p ossible　 to　 clas sify　 co rrectly　 with　 a　linear　 c la ss ifier,　 will　the　 p erceptr on　al gorithm　find　 such

a　classifier?　Yes ,　it　 does,　and　it　 w ill　 co nverge　to　such　a　classifier　in　 a　finite　numb er　 of　

updates　(mistakes ).　We’ll　sho w　this　 in　 lectur e　2.