Multiplicative Neural Network

Multiplicative neural networks (MNNs) utilize multiplication operations as core components, offering potential advantages over traditional additive networks in specific applications. Current research focuses on developing efficient MNN architectures, such as those employing shift-and-add operations or recursive building blocks for polynomial approximation, and exploring their approximation capabilities for various function classes, including those relevant to signal processing and simulation metamodeling. These networks show promise in improving energy efficiency in hardware implementations and enhancing the accuracy of models in tasks like image classification and online portfolio selection, particularly when dealing with high-order polynomials or signals with specific frequency characteristics.