Learning spatio-temporally invariant representations from video | IEEE Conference Publication | IEEE Xplore