@inproceedings {icaps16-225,
    track    = {​Main Track},
    title    = {OGA-UCT: On-the-Go Abstractions in UCT},
    url      = {http://www.aaai.org/ocs/index.php/ICAPS/ICAPS16/paper/view/13060},
    author   = {Ankit Anand and  Ritesh Noothigattu and   Mausam and  Parag Singla},
    abstract = {Recent work has begun exploring the value of domain
abstractions in Monte-Carlo Tree Search (MCTS) algo-
rithms for probabilistic planning. These algorithms au-
tomatically aggregate symmetric search nodes (states or
state-action pairs) saving valuable planning time. Exis-
ting algorithms alternate between two phases: (1) ab-
straction computation for computing node aggregations,
and (2) modified MCTS that use aggregate nodes. We
believe that these algorithms do not achieve the full po-
tential of abstractions because of disjoint phases – e.g.,
recovery from erroneous abstractions can be slow.

In response, we propose On the Go Abstractions
(OGA), a novel approach in which abstraction compu-
tation is tightly integrated into the MCTS algorithm. We
implement these on top of UCT and name the resulting
algorithm OGA-UCT. It has several desirable proper-
ties, including (1) rapid use of new information in mod-
ifying existing abstractions, (2) elimination of the ex-
pensive batch abstraction computation phase, and (3)
focusing abstraction computation on important part of
the sampled search space,. We experimentally compare
OGA-UCT against ASAP-UCT, a recent state-of-the-
art MDP algorithm as well as vanilla UCT algorithm.
We find that across a suite of planning competition and
other MDP domains, OGA-UCT performs better or at
par with the best technique on each domain obtaining
up to 18 % quality improvements.},
    keywords = {Probabilistic planning; MDPs and POMDPs}
}

{"_id":"845QXq3cfZfxA9DLT","bibbaseid":"anand-noothigattu-mausam-singla-ogauctonthegoabstractionsinuct","downloads":0,"creationDate":"2016-03-09T03:04:32.912Z","title":"OGA-UCT: On-the-Go Abstractions in UCT","author_short":["Anand, A.","Noothigattu, R.","Mausam","Singla, P."],"year":null,"bibtype":"inproceedings","biburl":"icaps16.icaps-conference.org/papers.bib","bibdata":{"bibtype":"inproceedings","type":"inproceedings","track":"​Main Track","title":"OGA-UCT: On-the-Go Abstractions in UCT","url":"http://www.aaai.org/ocs/index.php/ICAPS/ICAPS16/paper/view/13060","author":[{"firstnames":["Ankit"],"propositions":[],"lastnames":["Anand"],"suffixes":[]},{"firstnames":["Ritesh"],"propositions":[],"lastnames":["Noothigattu"],"suffixes":[]},{"firstnames":[],"propositions":[],"lastnames":["Mausam"],"suffixes":[]},{"firstnames":["Parag"],"propositions":[],"lastnames":["Singla"],"suffixes":[]}],"abstract":"Recent work has begun exploring the value of domain abstractions in Monte-Carlo Tree Search (MCTS) algo- rithms for probabilistic planning. These algorithms au- tomatically aggregate symmetric search nodes (states or state-action pairs) saving valuable planning time. Exis- ting algorithms alternate between two phases: (1) ab- straction computation for computing node aggregations, and (2) modified MCTS that use aggregate nodes. We believe that these algorithms do not achieve the full po- tential of abstractions because of disjoint phases – e.g., recovery from erroneous abstractions can be slow. In response, we propose On the Go Abstractions (OGA), a novel approach in which abstraction compu- tation is tightly integrated into the MCTS algorithm. We implement these on top of UCT and name the resulting algorithm OGA-UCT. It has several desirable proper- ties, including (1) rapid use of new information in mod- ifying existing abstractions, (2) elimination of the ex- pensive batch abstraction computation phase, and (3) focusing abstraction computation on important part of the sampled search space,. We experimentally compare OGA-UCT against ASAP-UCT, a recent state-of-the- art MDP algorithm as well as vanilla UCT algorithm. We find that across a suite of planning competition and other MDP domains, OGA-UCT performs better or at par with the best technique on each domain obtaining up to 18 % quality improvements.","keywords":"Probabilistic planning; MDPs and POMDPs","bibtex":"@inproceedings {icaps16-225,\r\n track = {​Main Track},\r\n title = {OGA-UCT: On-the-Go Abstractions in UCT},\r\n url = {http://www.aaai.org/ocs/index.php/ICAPS/ICAPS16/paper/view/13060},\r\n author = {Ankit Anand and Ritesh Noothigattu and Mausam and Parag Singla},\r\n abstract = {Recent work has begun exploring the value of domain\r\nabstractions in Monte-Carlo Tree Search (MCTS) algo-\r\nrithms for probabilistic planning. These algorithms au-\r\ntomatically aggregate symmetric search nodes (states or\r\nstate-action pairs) saving valuable planning time. Exis-\r\nting algorithms alternate between two phases: (1) ab-\r\nstraction computation for computing node aggregations,\r\nand (2) modified MCTS that use aggregate nodes. We\r\nbelieve that these algorithms do not achieve the full po-\r\ntential of abstractions because of disjoint phases – e.g.,\r\nrecovery from erroneous abstractions can be slow.\r\n\r\nIn response, we propose On the Go Abstractions\r\n(OGA), a novel approach in which abstraction compu-\r\ntation is tightly integrated into the MCTS algorithm. We\r\nimplement these on top of UCT and name the resulting\r\nalgorithm OGA-UCT. It has several desirable proper-\r\nties, including (1) rapid use of new information in mod-\r\nifying existing abstractions, (2) elimination of the ex-\r\npensive batch abstraction computation phase, and (3)\r\nfocusing abstraction computation on important part of\r\nthe sampled search space,. We experimentally compare\r\nOGA-UCT against ASAP-UCT, a recent state-of-the-\r\nart MDP algorithm as well as vanilla UCT algorithm.\r\nWe find that across a suite of planning competition and\r\nother MDP domains, OGA-UCT performs better or at\r\npar with the best technique on each domain obtaining\r\nup to 18 % quality improvements.},\r\n keywords = {Probabilistic planning; MDPs and POMDPs}\r\n}\r\n\r\n","author_short":["Anand, A.","Noothigattu, R.","Mausam","Singla, P."],"key":"icaps16-225","id":"icaps16-225","bibbaseid":"anand-noothigattu-mausam-singla-ogauctonthegoabstractionsinuct","role":"author","urls":{"Paper":"http://www.aaai.org/ocs/index.php/ICAPS/ICAPS16/paper/view/13060"},"keyword":["Probabilistic planning; MDPs and POMDPs"],"metadata":{"authorlinks":{}},"downloads":0,"html":""},"search_terms":["oga","uct","abstractions","uct","anand","noothigattu","mausam","singla"],"keywords":["probabilistic planning; mdps and pomdps"],"authorIDs":[],"dataSources":["iMkx859KiXcegwsin","EZtZjCTnxcdTTyeij"]}