OmniVec: Learning robust representations with cross modal sharing | IEEE Conference Publication | IEEE Xplore