Accurate and fast regression model based on solid mathematical foundation.
The code represents the implementation of the Kolmogorov-Arnold regression model and two methods of the model identification. This approach has recently gained popularity under name "KAN: Kolmogorov-Arnold Networks".
The code is written in MATLAB.
The code is a result of the collaborative project between myself and Andrew Polar conducted in 2019-2023. The basic concept of this approach has been published in Ref.1 in 2021. Our latest paper, Ref.2 published as a preprint in May 2023 (first version), has the most general formulation of the approach. Updated Ref.2 published as a preprint in June 2024 (second version) also includes data-driven solution of partial differential equations.
The main highlights of the code are:
Main script mainTriang runs the code. In the script, flag modelMethod selects the model type (splines or piecewise-linear) and the parameter estimation method (GN or NK). The latest addition is the accelerated NK method (see Ref.2, section on parallelisation/vectorisation). In the model building functions, flag printProgr switches the printout of the progress in the Command Window.
The code builds the model and plots log(RMSE) as a function of the number of passes through the data. The model constitutes two matrices with the parameters: fnB and fnT.
The obtained model can be used to make a prediction on a new dataset. For the spline version of the model,
y = modelKA_basisC( x, xmin, xmax, ymin, ymax, fnB, fnT );
should be executed, where x is the input data in the same format as in script mainTriang and y will be the predicted output data. For the piecewise-linear version of the model, the function that makes the prediction is modelKA_linear (it has the same format as above).
The computational example is a synthetic dataset - for each record, the inputs are the coordinates of three points in 2D and the output is the area of the triangle that is formed by the points. The points belong to the unit square. The default example takes approximately 14 seconds on a laptop with 11th Gen Intel Core i5.
There are three scripts with the units tests: testBasis for verifying the derivatives of the basis functions, testDeriv for verifying the derivatives of the model output by the inputs and by the parameters, and testSpline for comparing the implemented splines with the built-in MATLAB splines.
Data-driven solution of partial differential equations is implemented separately in script solvePDE_NK. The computational example is a second-order PDE; the details are given in Ref.2, the updated version from June 2024. The default example takes approximately 15 seconds on a laptop with 11th Gen Intel Core i5.
Script mainMIT runs the same example as in a recent Python implementation by other researchers (link in the file). It is added only for reproducing the latest benchmark results reported in Ref.2.
The code has been developed by Dr Michael Poluektov (University of Dundee, Department of Mathematical Sciences and Computational Physics).
The author would like to acknowledge the great help of Dr Andrew Polar, who contributed equally to the research project and maintains separate implementations of the approach in C# and C++ on his GitHub page.
A. Polar and M. Poluektov, Eng. Appl. Artif. Intell., 99:104137, 2021, link. ↩
M. Poluektov and A. Polar, arXiv:2305.08194, 2023, link. ↩ ↩2 ↩3 ↩4 ↩5
