Abstract
Forward stagewise and Frank Wolfe are popular gradient based projection free optimization algorithms which both require convex constraints. We propose a method to extend the applicability of these algorithms to problems of the form \(\min _x f(x) \quad s.t. \quad g(x) \le \kappa \) where f(x) is an invex (Invexity is a generalization of convexity and ensures that all local optima are also global optima.) objective function and g(x) is a non-convex constraint. We provide a theorem which defines a class of monotone component-wise transformation functions \(x_i = h(z_i)\). These transformations lead to a convex constraint function \(G(z) = g(h(z))\). Assuming invexity of the original function f(x) that same transformation \(x_i = h(z_i)\) will lead to a transformed objective function \(F(z) = f(h(z))\) which is also invex. For algorithms that rely on a non-zero gradient \(\nabla F\) to produce new update steps invexity ensures that these algorithms will move forward as long as a descent direction exists.
S. M. Keller and D. Murezzan—Equal contribution.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
In the supplementary materials, we provide an analytical proof which shows that for least squares regression problems the first active variable will always be correctly identified by the forward-stagewise method independent of the constraint. For other functions, the correctness of the first active variable can easily be empirically verified.
- 2.
This is a crucial property in the context of sparse regression, as only then the sparsity patterns in x- and z-space are identical.
References
Ben-Israel, A., Mond, B.: What is invexity? J. Aust. Math. Soc. Ser. B. Appl. Math. 28, 1–9 (1986)
Chechik, G., Globerson, A., Tishby, N., Weiss, Y.: Information bottleneck for Gaussian variables. J. Mach. Learn. Res. 6(1), 165–188 (2005)
Dinuzzo, F., Ong, C.S., Pillonetto, G., Gehler, P.V.: Learning output kernels with block coordinate descent. In: Proceedings of the 28th International Conference on Machine Learning (ICML-2011), pp. 49–56 (2011)
Fan, J., Li, R.: Variable selection via nonconcave penalized likelihood and its oracle properties. J. Am. Stat. Assoc. 96(456), 1348–1360 (2001)
Frank, M., Wolfe, P.: An algorithm for quadratic programming. Naval Res. Logistics (NRL) 3(1–2), 95–110 (1956)
Friedman, J.H.: Fast sparse regression and classification. Int. J. Forecast. 28(3), 722–738 (2012)
Gasso, G., Rakotomamonjy, A., Canu, S.: Solving non-convex lasso type problems with dc programming. In: IEEE Workshop on Machine Learning for Signal Processing, MLSP 2008, pp. 450–455. IEEE (2008)
Giorgi, G.: On first order sufficient conditions for constrained optima. In: Maruyama, T., Takahashi, W. (eds.) Nonlinear and Convex Analysis in Economic Theory, pp. 53–66. Springer, Heidelberg (1995). https://doi.org/10.1007/978-3-642-48719-4_5
Giorgi, G.: On some generalizations of preinvex functions 49 (2008)
Gorodnitsky, I.F., Rao, B.D.: Sparse signal reconstruction from limited data using focuss: a re-weighted minimum norm algorithm. IEEE Trans. Signal Process. 45(3), 600–616 (1997)
Hastie, T., Taylor, J., Tibshirani, R., Walther, G.: Forward stagewise regression and the monotone lasso. Electron. J. Stat. 1, 1–29 (2007). https://doi.org/10.1214/07-EJS004
Jaggi, M.: Revisiting Frank-Wolfe: projection-free sparse convex optimization. In: ICML, vol. 1, pp. 427–435 (2013)
Lanza, A., Morigi, S., Sgallari, F.: Convex image denoising via non-convex regularization. In: Aujol, J.-F., Nikolova, M., Papadakis, N. (eds.) SSVM 2015. LNCS, vol. 9087, pp. 666–677. Springer, Cham (2015). https://doi.org/10.1007/978-3-319-18461-6_53
Li, G., Yan, Z., Wang, J.: A one-layer recurrent neural network for constrained nonsmooth invex optimization. Neural Netw. 50, 79–89 (2014)
Li, X., Zhao, T., Zhang, T., Liu, H.: The picasso package for nonconvex regularized M-estimation in high dimensions in R. Technical report (2015)
Mazumder, R., Friedman, J.H., Hastie, T.: SparseNet: coordinate descent with nonconvex penalties. J. Am. Stat. Assoc. 106(495), 1125–1138 (2011)
Mishra, S., Giorgi, G.: Invexity and Optimization. Nonconvex Optimization and Its Applications. Springer, Heidelberg (2008). https://doi.org/10.1007/978-3-540-78562-0
Pedregosa, F., et al.: Scikit-learn: machine learning in python. J. Mach. Learn. Res. 12, 2825–2830 (2011)
Rey, M., Fuchs, T., Roth, V.: Sparse meta-Gaussian information bottleneck. In: Proceedings of the 31st International Conference on Machine Learning (ICML-2014), pp. 910–918 (2014)
Rey, M., Roth, V.: Meta-Gaussian information bottleneck. In: Advances in Neural Information Processing Systems-NIPS 25 (2012)
Tibshirani, R.J.: A general framework for fast stagewise algorithms. J. Mach. Learn. Res. 16, 2543–2588 (2015)
Tishby, N., Pereira, F.C., Bialek, W.: The information bottleneck method. arXiv preprint physics/0004057 (2000)
Zhang, C.H.: Nearly unbiased variable selection under minimax concave penalty. Ann. Stat. 38, 894–942 (2010)
Acknowledgements
This project is supported by the the Swiss National Science Foundation project CR32I2 159682.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
1 Electronic supplementary material
Below is the link to the electronic supplementary material.
Rights and permissions
Copyright information
© 2019 Springer Nature Switzerland AG
About this paper
Cite this paper
Keller, S.M., Murezzan, D., Roth, V. (2019). Invexity Preserving Transformations for Projection Free Optimization with Sparsity Inducing Non-convex Constraints. In: Brox, T., Bruhn, A., Fritz, M. (eds) Pattern Recognition. GCPR 2018. Lecture Notes in Computer Science(), vol 11269. Springer, Cham. https://doi.org/10.1007/978-3-030-12939-2_47
Download citation
DOI: https://doi.org/10.1007/978-3-030-12939-2_47
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-12938-5
Online ISBN: 978-3-030-12939-2
eBook Packages: Computer ScienceComputer Science (R0)