We propose a neural network architecture grounded in high-dimensional hypercube topology. In contrast to conventional sequential or skip-connected designs, the proposed approach maps layers to the vertices of an $n$ -dimensional hypercube and establishes inter... ...

请注册登录后继续浏览