Open XiaohuiZhang1996 opened 4 months ago
I haven't conducted parameter tuning experiments on the two frameworks (MLP
and KAN
) yet. Currently, for the tasks presented in the demo(sklearn.make_regression
and fashion-mnist
), the performance achievable by the two models might be similar(and KAN
uses fewer parameters as its non-linear ability is transferred to the calculation of B-spline basis).
Was curious about the performance in your two example demo and whether KAN outperforms the MLP.