arXiv:2605.06729v1 Announce Type: new Abstract: We present the E-MHC-Geo Transformer, a novel architecture that unifies Manifold-Constrained Hyper-Connections (mHC), Deep Delta Learning (DDL), and the Cayley transform to obtain input-adaptive, unconditionally orthogonal residual connections. Unlike DDL, whose Householder operator is orthogonal only at , our Data-Dependent Cayley rotation preserves orthogonality for all and all inputs. To handle negation, an eigenvalue case that Cayley provably…
arXiv:2605.06729v1 Announce Type: new Abstract: We present the E-MHC-Geo Transformer, a novel architecture that unifies Manifold-Constrained Hyper-Connections (mHC), Deep Delta Learning (DDL), and the Cayley transform to obtain input-adaptive, unconditionally orthogonal residual connections. Unlike DDL, whose Householder operator is orthogonal only at , our Data-Dependent Cayley rotation preserves orthogonality for all and all inputs. To handle negation, an eigenvalue case that Cayley provably…