Abstract: The article focuses on enhancing sigma-pi neural networks by using polynomial activation functions to address the vanishing gradient problem common in deep networks with linear activations.