Ontology of Synthesis A

Generalized-Confined Subspace Learning (GCSL) is a popular technique that aims at learning a subspace for a given input [@tolstikhin2016generalized; @tol2018generalized] by constraining the data distribution to be a sub-Gaussian distribution. GCSL has been widely used in various tasks such as image classification [@zhao2016learning] and video classification [citation], and has demonstrated promising results in many recent studies [@li2019gcsl]. In most studies, the GCSL model is initialized with the same dataset structure used in [@chen2018learning; @li2019deep; @yang2019gccsd; @zhang2019learning; ] and the architecture is optimized with the *GCS* loss. The main difference between these models is that the GDSL models are optimized in a *GDC* way while the GSCDL models are trained in a GDC way. The GDSSL models use more hyperbolic systems than the usual two-dimensional GISSL models. The reason is that we need to obtain the hyperbola for the three-dimensional system. The second step of the GDSS is to estimate the hyperbolae for the two- and three-dimentional systems. For the two dimensional case, we have used the same method of the one-dimensional case.