Noise supplement learning algorithm for associative memories using quantized multilayer perceptrons and sparsely interconnected neural networks

摘要

Recently, we have proposed associative memories composed of multilayer perceptrons (MLPs) and sparsely interconnected neural networks (SINNs), named MLP-SINN. Our proposed associative memories are suitable for hardware implementation and have better performance than both MLPs and SINNs. However, if MLP and SINN are synthesized independently, their capabilities are not effectively used in our MLP-SINN. Therefore, we have proposed the noise supplement learning for MLP-SINN associative memories to improve them. In this report, we investigate the effectiveness of the noise supplement learning for our MLP-SINN composed of quantized MLP and SINN.