The paper introduces Kolmogorov-Arnold Networks (KANs) as an alternative to Multi-Layer Perceptrons (MLPs). Unlike MLPs, which use fixed activation functions on nodes, KANs use learnable activation functions on edges, replacing linear weights with univariate functions parametrized as splines. This approach enhances both accuracy and interpretability. The authors demonstrate that KANs can achieve comparable or better accuracy than larger MLPs in function fitting tasks and exhibit faster neural scaling laws. They also show that KANs can be visually intuitive and easily interacted with human users, making them useful for scientific discoveries in mathematics and physics. The paper includes theoretical guarantees, empirical experiments, and comparisons with other models to support these claims.The paper introduces Kolmogorov-Arnold Networks (KANs) as an alternative to Multi-Layer Perceptrons (MLPs). Unlike MLPs, which use fixed activation functions on nodes, KANs use learnable activation functions on edges, replacing linear weights with univariate functions parametrized as splines. This approach enhances both accuracy and interpretability. The authors demonstrate that KANs can achieve comparable or better accuracy than larger MLPs in function fitting tasks and exhibit faster neural scaling laws. They also show that KANs can be visually intuitive and easily interacted with human users, making them useful for scientific discoveries in mathematics and physics. The paper includes theoretical guarantees, empirical experiments, and comparisons with other models to support these claims.