{ "contributor-id" "5ca58a6a-aa46-4e5a-a2e1-b3fc6bc2efa6" "description" "A dropout uncertainty neural network (DUNN) potential for condensed-matter carbon systems with a dropout ratio of 0.3. This is an ensemble model consisting of 100 different network structures obtained by dropout. Before dropout, there are three hidden layers each containing 128 neurons; each neuron in the hidden layers has probability 0.3 of being removed from the network. By default, the model will run in the 'mean' mode where the output energy, forces, and virial are obtained by averaging over the 100 ensembles. If desired, one can set the 'active_member_id' to '0' to use the fully-connected structure or to '1, 2, ..., 100' to use a single ensemble member. When multiple ensemble members are used, the ensemble average of the energy and forces are what are ultimately returned for a given configuration." "developer" [ "5ca58a6a-aa46-4e5a-a2e1-b3fc6bc2efa6" "360c0aed-48ce-45f6-ba13-337f12a531e8" ] "doi" "10.25950/656f7a62" "domain" "openkim.org" "executables" [] "extended-id" "DUNN_WenTadmor_2019v3_C__MO_714772088128_000" "kim-api-version" "2.0" "maintainer-id" "5ca58a6a-aa46-4e5a-a2e1-b3fc6bc2efa6" "model-driver" "DUNN__MD_292677547454_000" "potential-type" "dunn" "publication-year" "2019" "source-citations" [ { "author" "Mingjian Wen and Ellad B. Tadmor" "title" "Uncertainty quantification in molecular simulations with dropout neural network potentials" "journal" "npj Computational Materials" "volume" "6" "issue" "1" "year" "2020" "doi" "10.1038/s41524-020-00390-8" "issn" "2057-3960" "recordkey" "MO_714772088128_000a" "recordprimary" "recordprimary" "recordtype" "article" } ] "species" [ "C" ] "title" "Dropout uncertainty neural network (DUNN) potential for condensed-matter carbon systems developed by Wen and Tadmor (2019) v000" }