Duality

There is a duality between SVM and Boosting. This is why people still think that they have not understood boosting properly. Everything related to the norm two of the classifier and the norm 2 of the data in SVM can be replaced with norm 1 of the hypothesis and norm infinity of the data. And then you will get exactly the same result! However, where is the place of the random projection in this case? We have a very tiny intuition about the meaning of the projection in boosting (and also winnow) but that is far from what we expect to know. People do believe that SVM and boosting are dual of each other with the same power. However, by this Johnson-Lindenstrauss lemma we have a somehow good understanding about the roll of random projection is SVMs but people know almost nothing about the counterpart in norm 1- norm infinity (boosting style).

Love it

ساعت شش و بیست و چهار دقیقه بامداد. تموم شد . کل شب رو با این بودم و واقعا چه زود صبح شد. دومین شب متوالیه که این اتفاق می افته. زیاده حرفی نیست امروز