Author:
Balashov M. V., ,Biglov K. Z.,Tremba A. A., , ,
Abstract
We consider some problems with a set-valued mapping, which can be reduced to minimization of a homogeneous Lipschitz function on the unit sphere. Latter problem can be solved in some cases with a first order algorithm—the gradient projection method. As one of the examples, the case when set-valued mapping is the reachable set of a linear autonomous controlled system is considered. In several settings, the linear convergence is proven. The methods used in proofs follow those introduced by B.T. Polyak for the case where Lezanski– Polyak–Lojasiewicz condition holds. Unlike algorithms that use approximation of the reachable set, the proposed algorithms depend far less on dimension and other parameters of the problem. Efficient error estimation is possible. Numerical experiments confirm the effectiveness of the considered approach. This approach can also be applied to various set-theoretical problems with general set-valued mappings.
Publisher
The Russian Academy of Sciences
Reference30 articles.
1. 1. Ioffe, A.D., Metric regularity - a survey Part I and II, J. Aust. Math. Soc., 2016, vol. 101, pp. 188-243, pp. 376-417.
2. 2. Luke, D.R., Finding best approximation pairs relative to a convex and prox-regular set in a Hilbert space, SIAM J. Optim., 2008, vol. 19, no. 2, pp. 714-739.
3. 3. Gr¨unew¨alder, S., Compact convex projections, J. Mach. Learn. Res., 2018, vol. 18, no. 2019, pp. 1-43.
4. 4. Sosa, W. and Raupp, F.M.P, An algorithm for projecting a point onto a level set of a quadratic function, Optimization, 2022, vol. 71, no. 1, pp. 71-89.
5. 5. Bregman, L.M., Censor, Y., Reich, S., and Zepkowitz-Malachi, Y., Finding the projection of a point onto the intersection of convex sets via projections onto half-spaces, J. Approx. Theory., 2003, vol. 124, no. 2, pp. 194-218.