Active Pointillistic Pattern Search. Ma, Y.; Sutherland, D., J.; Garnett, R.; and Schneider, J. In AISTATS, pages 672–680, 2, 2015.
Active Pointillistic Pattern Search [pdf]Website  abstract   bibtex   
We introduce the problem of active pointillistic pattern search (APPS), which seeks to discover regions of a domain exhibiting desired behavior with limited observations. Unusually, the patterns we consider are defined by large-scale proper-ties of an underlying function that we can only observe at a limited number of points. Given a description of the desired patterns (in the form of a classifier taking functional inputs), we se-quentially decide where to query function values to identify as many regions matching the pattern as possible, with high confience. For one broad class of models the expected reward of each un-observed point can be computed analytically. We demonstrate the proposed algorithm on three dif-ficult search problems: locating polluted regions in a lake via mobile sensors, forecasting winning electoral districts with minimal polling, and iden-tifying vortices in a fluid flow simulation.
@inproceedings{
 title = {Active Pointillistic Pattern Search},
 type = {inproceedings},
 year = {2015},
 identifiers = {[object Object]},
 pages = {672–680},
 websites = {http://proceedings.mlr.press/v38/ma15.html,http://jmlr.org/proceedings/papers/v38/ma15.pdf},
 month = {2},
 day = {21},
 city = {San Diego, USA},
 id = {c58c62ac-2d45-3847-8526-68ef0221f9a6},
 created = {2019-09-11T02:53:54.962Z},
 accessed = {2018-07-22},
 file_attached = {false},
 profile_id = {ad469375-6e0c-3e76-b763-9d9fdd9285a3},
 group_id = {302cd49e-47c4-3446-9da2-2f3bd6678dd3},
 last_modified = {2019-09-11T02:53:54.962Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {Ma2015},
 private_publication = {false},
 abstract = {We introduce the problem of active pointillistic pattern search (APPS), which seeks to discover regions of a domain exhibiting desired behavior with limited observations. Unusually, the patterns we consider are defined by large-scale proper-ties of an underlying function that we can only observe at a limited number of points. Given a description of the desired patterns (in the form of a classifier taking functional inputs), we se-quentially decide where to query function values to identify as many regions matching the pattern as possible, with high confience. For one broad class of models the expected reward of each un-observed point can be computed analytically. We demonstrate the proposed algorithm on three dif-ficult search problems: locating polluted regions in a lake via mobile sensors, forecasting winning electoral districts with minimal polling, and iden-tifying vortices in a fluid flow simulation.},
 bibtype = {inproceedings},
 author = {Ma, Yifei and Sutherland, Dougal J and Garnett, Roman and Schneider, Jeff},
 booktitle = {AISTATS}
}
Downloads: 0