L-curve
From HandWiki
L-curve is a visualization method used in the field of regularization in numerical analysis and mathematical optimization.[1] It represents a logarithmic plot where the norm of a regularized solution is plotted against the norm of the corresponding residual norm. It is useful for picking an appropriate regularization parameter for the given data.[2]
This method can be applied on methods of regularization of least-square problems, such as Tikhonov regularization and the Truncated SVD,[2] and iterative methods of solving ill-posed inverse problems, such as the Landweber algorithm, Modified Richardson iteration and Conjugate gradient method.
References
- ↑ "L-Curve and Curvature Bounds for Tikhonov Regulairzation" (PDF). https://www.math.kent.edu/~reichel/publications/tikcrvL.pdf.
- ↑ 2.0 2.1 Hansen, P. C. (2001). "The L-curve and its use in the numerical treatment of inverse problems". in Johnston, P. R.. Computational Inverse Problems in Electrocardiography. WIT Press. pp. 119–142. ISBN 978-1-85312-614-7. https://www.sintef.no/globalassets/project/evitameeting/2005/lcurve.pdf.
- Hanke, Martin. "Limitations of the L-curve method in ill-posed problems." BIT Numerical Mathematics 36.2 (1996): 287-301.
- Engl, Heinz W., and Wilhelm Grever. "Using the L--curve for determining optimal regularization parameters." Numerische Mathematik 69.1 (1994): 25-31.
 |  |
