This　paper　presents　a　multivalued　deep　neural　network　(DNN)　inverse　modeling　technique　and　its　applications　in　high-dimensional　microwave　modeling　for　parameter　extraction　of　microwave　filters.　DNNs　with　smooth　ReLUs　have　been　proven　to　have　significant　abilities　in　dealing　with　complex　design　challenges,　particularly　in　high-dimensional　microwave　forward　modeling.　However,　for　inverse　modeling,　the　conventional　DNNs　with　smooth　ReLUs　face　difficulties　because　they　cannot　solve　the　non-　uniquenessproblem　which　is　a　common　and　key　issue　in　inverse　modeling.　In　this　paper,　we　propose　a　high-dimensional　inverse　modeling　technique　using　multivalued　DNN　with　smooth　ReLUs　to　address　the　inverse　modeling　problem　with　high　complexity　and　non-uniqueness　issue.　Finally,　a　more　accurate　DNN　model　can　be　achieved　using　the　proposed　technique　compared　to　existing　DNN　modeling　techniques.　A　high-dimensional　inverse　modeling　example　for　parameter　extraction　of　a　microwave　filter　is　presented　to　validate　the　effectiveness　of　the　proposed　technique.　©　2023　IEEE.