tag:blogger.com,1999:blog-6141980.post6988018045782573814..comments2019-05-24T02:33:04.117-05:00Comments on Nuit Blanche: GraphLab workshop, Why should you care ?Igorhttp://www.blogger.com/profile/17474880327699002140noreply@blogger.comBlogger6125tag:blogger.com,1999:blog-6141980.post-76211419937414720012012-09-21T09:59:12.713-05:002012-09-21T09:59:12.713-05:00Dear Igor,
Thanks for the post. I have some forma...Dear Igor,<br /><br />Thanks for the post. I have some formal training in machine learning and I'm pretty comfortable with Python. So, is graphlab a good entry point to large scale or distributed machine learning?<br /><br />Also, I think the links that you provided to MATLAB and the Python/Jython API are both broken. Could you please update those?<br /><br />ThanksJaidev Deshpandehttps://www.blogger.com/profile/01082878986852695651noreply@blogger.comtag:blogger.com,1999:blog-6141980.post-63266353690633945632012-03-15T13:11:29.631-05:002012-03-15T13:11:29.631-05:00Yes, exactly, graphlab is designed to support spar...Yes, exactly, graphlab is designed to support sparse graphs better than dense graph. This is because that the amount of correlation between variables/factors/etc is significantly reduced in sparse graphs and thus allowing better parallelism.Danny Bicksonhttps://www.blogger.com/profile/01517237836051035400noreply@blogger.comtag:blogger.com,1999:blog-6141980.post-18760153346002772182012-03-15T13:00:59.382-05:002012-03-15T13:00:59.382-05:00yes the LDPC matrices should also work. My underst...yes the LDPC matrices should also work. My understanding is that GraphLab allows the "chunking" of the work in iterative algorithms because at every step, the multiplication is performed with a sparse measurement matrix (thereby reducing not nulling) the communication between cpus. So to a certain extent the sparsity of the matrix being multiplied at every step of the iterative algorithm "maps" into the graph between cpus. Danny ?Igorhttps://www.blogger.com/profile/17474880327699002140noreply@blogger.comtag:blogger.com,1999:blog-6141980.post-1136792383932119252012-03-15T12:48:49.385-05:002012-03-15T12:48:49.385-05:00Thanks for enlightening us. Now that you mention s...Thanks for enlightening us. Now that you mention sparse measurement matrices, LDPC matrices have shown some promise in a couple of papers under way recently. Could they be relevant here?<br />I wonder about the somewhat sparse measurement matrices of the random demodulator as well?Thomas Arildsenhttps://www.blogger.com/profile/00461210615375168098noreply@blogger.comtag:blogger.com,1999:blog-6141980.post-31821910748472204882012-03-15T12:32:59.115-05:002012-03-15T12:32:59.115-05:00Igor, you are amazing! I am sure now we get a much...Igor, you are amazing! I am sure now we get a much improved visibility for the CS/ matrix factorization community.Danny Bicksonhttps://www.blogger.com/profile/01517237836051035400noreply@blogger.comtag:blogger.com,1999:blog-6141980.post-31665824990989169062012-03-15T12:30:21.581-05:002012-03-15T12:30:21.581-05:00Thanks for making a complex idea understandable.Thanks for making a complex idea understandable.Robhttp://www.merchantpages.comnoreply@blogger.com