This paper provides a fair comparison between Kolmogorov-Arnold Networks (KAN) and Multi-Layer Perceptrons (MLP) across various tasks, including machine learning, computer vision, natural language processing, audio processing, and symbolic formula representation. The study controls for the number of parameters and FLOPs to ensure a fair comparison. The main findings are that KAN outperforms MLP only in symbolic formula representation tasks, while MLP generally outperforms KAN in other tasks. The advantage of KAN in symbolic formula representation is attributed to its B-spline activation function. When B-spline is applied to MLP, its performance in symbolic formula representation matches or surpasses that of KAN. However, B-spline does not significantly improve MLP's performance in other tasks. Additionally, KAN exhibits more severe forgetting issues in continual learning tasks compared to MLP. The study also shows that KAN can be viewed as a special type of MLP with learnable B-spline activation functions. The results suggest that KAN and MLP are suited for different tasks, with KAN excelling in symbolic formula representation and MLP excelling in other tasks. The paper provides insights for future research on KAN and other MLP alternatives.This paper provides a fair comparison between Kolmogorov-Arnold Networks (KAN) and Multi-Layer Perceptrons (MLP) across various tasks, including machine learning, computer vision, natural language processing, audio processing, and symbolic formula representation. The study controls for the number of parameters and FLOPs to ensure a fair comparison. The main findings are that KAN outperforms MLP only in symbolic formula representation tasks, while MLP generally outperforms KAN in other tasks. The advantage of KAN in symbolic formula representation is attributed to its B-spline activation function. When B-spline is applied to MLP, its performance in symbolic formula representation matches or surpasses that of KAN. However, B-spline does not significantly improve MLP's performance in other tasks. Additionally, KAN exhibits more severe forgetting issues in continual learning tasks compared to MLP. The study also shows that KAN can be viewed as a special type of MLP with learnable B-spline activation functions. The results suggest that KAN and MLP are suited for different tasks, with KAN excelling in symbolic formula representation and MLP excelling in other tasks. The paper provides insights for future research on KAN and other MLP alternatives.