Figure 5
A bar graph compares error rates between single-layer and two-layer networks for different activation functions. The horizontal axis lists the activation functions: Tanh, Sigmoid, ReLu, GeLu, and Swish. The vertical axis represents the error rate, ranging from 0 to 18. The graph features two sets of vertical bars for each activation function, with blue bars representing single-layer networks and orange bars representing two-layer networks. For Tanh, the error rates are approximately 3 for single-layer and 11 for two-layer networks. For Sigmoid, the error rates are approximately 7 for single-layer and 16 for two-layer networks. For ReLu, the error rates are approximately 5 for single-layer and 6 for two-layer networks. For GeLu, the error rates are approximately 3 for single-layer and 4 for two-layer networks. For Swish, the error rates are approximately 3 for single-layer and 11 for two-layer networks.A comparison of error rates between single-layer (blue) and two-layer (orange) networks, demonstrating the superiority of the simpler architecture for all tested activation functions