tag:blogger.com,1999:blog-6555947.post7647922661287765114..comments2014-01-12T10:46:48.153-07:00Comments on The Geomblog: Core Sets As GridsSuresh Venkatasubramanianhttps://plus.google.com/112165457714968997350noreply@blogger.comBlogger5125tag:blogger.com,1999:blog-6555947.post-12663563279418180192008-01-15T09:46:00.000-07:002008-01-15T09:46:00.000-07:00Have to have *something* to do at 4 am :)Have to have *something* to do at 4 am :)Sureshhttp://www.blogger.com/profile/15898357513326041822noreply@blogger.comtag:blogger.com,1999:blog-6555947.post-29989333923869056362008-01-15T06:01:00.000-07:002008-01-15T06:01:00.000-07:00Congrats on the new 6-pounder Suresh! I am amazed ...Congrats on the new 6-pounder Suresh! I am amazed you are still able to spend some cycles entertaining your <EM>Geomblog</EM> clientele!<BR/><BR/>aravindAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-6555947.post-89576777909874385372008-01-14T15:31:00.000-07:002008-01-14T15:31:00.000-07:00While some of the coreset stuff is only theoretica...While some of the coreset stuff is only theoretically useful, some of it is very useful in practice. See the work by Agarwal and Yu (and maybe Varadarajan was also involved). The line of attack is multi stages: (i) Is there a coreset for this problem? (ii) Can this coreset be computed efficient? And (iii) can it be computed incrementally (i.e., these leads to useful construction)? <BR/><BR/>Philosophically, if there is a coreset for a problem, then most of the data given is redundant and can be ignored. The question is how to do that efficiently in practice...Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-6555947.post-35302660950688529272008-01-14T13:48:00.000-07:002008-01-14T13:48:00.000-07:00Practically, repeatedly applying the core sets ide...Practically, repeatedly applying the core sets ideas does not work (like using many balls, each is from a core set of a subset of the input points, to cover a set of points in 3D). I have one of my graduate students to implement that, after trying for a few days, he refused to continue ("it is too slow and it is a garbage"). <BR/><BR/>Computing core set once is ok (in fact, beautiful), but is only meaningful theoretically.<BR/><BR/>In the long run, the whole core set stuff will be like parametric search, IMHO.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-6555947.post-55361396562786327702008-01-14T07:29:00.000-07:002008-01-14T07:29:00.000-07:00Congratulations!JeffPCongratulations!<BR/><BR/>JeffPJeffPnoreply@blogger.com