FPGA Simulation for Computing Pseudoinverse Matrices

Many scientific and engineering processing models of big data deal with classification tasks. In particular, current intelligent approaches work efficiently for them, as they use neural techniques with shallow architectures. The Extreme Learning Machine (ELM), which is a one-hidden layer feedforward neural network, is capable of achieving high accuracy in these tasks. Moreover, one of its weighting matrices is the pseudoinverse of the hidden data. In this sense, this work presents meaningful results of computing the Moore-Penrose pseudoinverse for the ELM, using a recurrent model. The numerical simulations were based on an FPGA framework used to design a pseudoinverse-computing core.