{"success":1,"msg":"","color":"rgb(28, 35, 49)","title":"Meshed Gaussian Processes for efficient Bayesian inference of big data spatial regression models<\/b>","description":"webinar","title2":"","start":"2021-02-03 03:30","end":"2021-02-03 04:30","responsable":"Donatello Telesca <\/i><\/a>","speaker":"Michele Peruzzi (Duke University)","id":"22","type":"webinar","timezone":"America\/Los_Angeles","activity":"Join Zoom Meeting\r\nhttps:\/\/ucla.zoom.us\/j\/93422849767\r\n\r\nMeeting ID: 934 2284 9767\r\nOne tap mobile\r\n+16692192599,,93422849767# US (San Jose)\r\n+12133388477,,93422849767# US (Los Angeles)\r\n\r\nDial by your location\r\n +1 669 219 2599 US (San Jose)\r\n +1 213 338 8477 US (Los Angeles)\r\n +1 720 928 9299 US (Denver)\r\n +1 971 247 1195 US (Portland)\r\n +1 253 215 8782 US (Tacoma)\r\n +1 346 248 7799 US (Houston)\r\n +1 602 753 0140 US (Phoenix)\r\n +1 786 635 1003 US (Miami)\r\n +1 301 715 8592 US (Washington D.C)\r\n +1 312 626 6799 US (Chicago)\r\n +1 470 250 9358 US (Atlanta)\r\n +1 646 558 8656 US (New York)\r\n +1 651 372 8299 US (Minnesota)\r\nMeeting ID: 934 2284 9767\r\nFind your local number: https:\/\/ucla.zoom.us\/u\/adCYieihog\r\n\r\nJoin by SIP\r\n93422849767@zoomcrc.com\r\n\r\nJoin by H.323\r\n162.255.37.11 (US West)\r\n162.255.36.11 (US East)\r\n115.114.131.7 (India Mumbai)\r\n115.114.115.7 (India Hyderabad)\r\n213.19.144.110 (Amsterdam Netherlands)\r\n213.244.140.110 (Germany)\r\n103.122.166.55 (Australia)\r\n149.137.40.110 (Singapore)\r\n64.211.144.160 (Brazil)\r\n69.174.57.160 (Canada)\r\n207.226.132.110 (Japan)\r\nMeeting ID: 934 2284 9767","abstract":"Big spatial data are now routinely collected in massive amounts in diverse scientific and data-driven industrial applications including, but not limited to, natural and environmental sciences; economics; climate science; ecology; forestry; and public health. In this talk, I will introduce Meshed Gaussian Processes (MGPs) for scalable Bayesian regression modeling of spatial Big Data. The underlying idea combines concepts on high-dimensional geostatistics by partitioning the spatial domain and modeling the regions in the partition using a sparsity-inducing directed acyclic graph (DAG). Unlike other methods, MGPs consider the DAG as an explicit design choice -- rather than building the DAG based on some criterion (e.g. limiting conditional dependence to the m nearest neighbors), one chooses a DAG because of its known properties. The DAG is linked to groups of spatial locations, arising e.g. from domain tiling, tessellations, or other partitioning strategies. In particular, one may consider two particularly convenient DAGs and the corresponding domain partitioning strategies: (1) a recursive tree, (2) a \"cubic\" mesh. I will focus on the latter and show that the resulting \"cubic\" MGP (QMGP) corresponds to efficient parallel MCMC sampling of the latent spatial process, even with spatiotemporal data at more than ten million locations. I will then mention refinements, improvements and extensions of MGPs and QMGPs in particular:\r\n\r\n(1) MCMC for QMGPs may exhibit slow convergence for irregularly spaced data and\/or in estimating the covariance parameters a posteriori. I will resolve these issues by showing that a Grid-Parametrize-Split (GriPS) strategy results in massively more efficient MCMC. \r\n\r\n(2) Why MCMC though? In some scenarios, it may be possible to fix some covariance parameters at some reasonable value; then, MCMC may be avoided. I will outline the possible computational advantages of QMGPs in these settings, compared to existing alternatives.\r\n\r\n(3) The idea of fixing the DAG allows one to devise tailor-made MCMC algorithms for sampling specific MGPs. As a result, MGPs may facilitate computations for more general regression models on (multivariate) non-Gaussian outcomes."}