With the ever increasing application of Convolutional Neural Networks to customer products the need emerges for models to efficiently run on embedded, mobile hardware. Slimmer models have therefore become a hot research topic with various approaches which vary from binary networks to revised convolution layers. We offer our contribution to the latter and propose a novel convolution block which significantly reduces the computational burden while surpassing the current stateof-the-art. Our model, dubbed EffNet, is optimised for models which are slim to begin with and is created to tackle issues in existing models such as MobileNet and ShuffleNet.

Abstract. Multidimensional lossless networks are of special interest for use as reference structures for multidimensional wave digital filters [1]- [3]. The starting point of the presented synthesis procedure for two-dimensional representatives of the networks mentioned is a scattering matrix description of the desired multiport. This given matrix is assumed to have those properties which have turned out to be necessary [9], [ 10] for any scattering matrix of a multidimensional lossless network. The method presented for the synthesis of 2-D reactance m-ports is based mainly on known properties of block-companion matrices and the factorization of a univariable rational matrix which is discrete para-Hermitian and nonnegative definite on the unit circle.The resulting network always contains only a minimal number of frequencydependent building elements. No restrictions are made concerning the coefficients of the rational entries of the scattering matrix; they may be either real or complex, so as to include even complex networks which are of special interest for multidimensional wave digital filters [3].

No abstract

No abstract

No abstract