KANLLPR: KAN with Last-Layer Prediction Rigidity

In this video we create a new sklearn-compatible version of the Kolmogorov-Arnold Networks with LLPR (Last-Layer Prediction Rigidity): Principled uncertainty quantification. This is more efficient than the DPOSE version if you have homoskedastic (constant) noise). https://github.com/jkitchin/pycse/blob/master/src/pycse/examples/KANLLPR_Demo.ipynb