Infinite kernel learning via infinite and semi-infinite programming

Weber, Gerhard Wilhelm
As data become heterogeneous, multiple kernel learning methods may help to classify them. To overcome the drawback lying in its (multiple) finite choice, we propose a novel method of 'infinite' kernel combinations for learning problems with the help of infinite and semi-infinite optimizations. Looking at all the infinitesimally fine convex combinations of the kernels from an infinite kernel set, the margin is maximized subject to an infinite number of constraints with a compact index set and an additional (Riemann-Stieltjes) integral constraint due to the combinations. After a parametrization in the space of probability measures, we get a semi-infinite programming problem. We analyse regularity conditions (reduction ansatz) and discuss the type of density functions in the constraints and the bilevel optimization problem derived. Our proposed approach is implemented with the conceptual reduction method and tested on homogeneous and heterogeneous data; this yields a better accuracy than a single-kernel learning for the heterogeneous data. We analyse the structure of problems obtained and discuss structural frontiers, trade-offs and research challenges.


Manguoğlu, Murat; Sameh, Ahmed H.; Grama, Ananth (Society for Industrial & Applied Mathematics (SIAM), 2010-01-01)
The emergence of multicore architectures and highly scalable platforms motivates the development of novel algorithms and techniques that emphasize concurrency and are tolerant of deep memory hierarchies, as opposed to minimizing raw FLOP counts. While direct solvers are reliable, they are often slow and memory-intensive for large problems. Iterative solvers, on the other hand, are more efficient but, in the absence of robust preconditioners, lack reliability. While preconditioners based on incomplete factor...
Effective optimization with weighted automata on decomposable trees
Ravve, E. V.; Volkovich, Z.; Weber, Gerhard Wilhelm (Informa UK Limited, 2014-01-02)
In this paper, we consider quantitative optimization problems on decomposable discrete systems. We restrict ourselves to labeled trees as the description of the systems and we use weighted automata on them as our computational model. We introduce a new kind of labeled decomposable trees, sum-like weighted labeled trees, and propose a method, which allows us to reduce the solution of an optimization problem, defined in a fragment of Weighted Monadic Second Order Logic, on such a tree to the solution of effec...
INAN, K (Springer Science and Business Media LLC, 1992-01-01)
Supervisory control problems are formulated in terms of a process model where the mechanism of control is expressed in terms of an algebraic operator with the plant and supervision processes as its arguments. The solution subspaces for supervisory processes restrict the observation and the control capability of supervision. The main result corresponds to decentralized marked supervision under partial observations, and specific cases are derived from this result in a unified, algebraic way. The result and it...
Seven, Ahmet İrfan (American Mathematical Society (AMS), 2011-05-01)
There is a particular analogy between combinatorial aspects of cluster algebras and Kac-Moody algebras: roughly speaking, cluster algebras are associated with skew-symmetrizable matrices while Kac-Moody algebras correspond to (symmetrizable) generalized Cartan matrices. Both classes of algebras and the associated matrices have the same classification of finite type objects by the well-known Cartan-Killing types. In this paper, we study an extension of this correspondence to the affine type. In particular, w...
Gleason's problem and homogeneous interpolation in Hardy and Dirichlet-type spaces of the ball
Alpay, D; Kaptanoglu, HT (Elsevier BV, 2002-12-15)
We solve Gleason's problem in the reproducing kernel Hilbert spaces with reproducing kernels 1/(1 - Sigma(1)(N) z(j) (W) over bar (j))(r) for real r > 0 and their counterparts for r less than or equal to 0, and study the, I homogeneous interpolation problem in these spaces. (C) 2002 Elsevier Science (USA). All rights reserved.
Citation Formats
S. AKYÜZ and G. W. Weber, “Infinite kernel learning via infinite and semi-infinite programming,” OPTIMIZATION METHODS & SOFTWARE, pp. 937–970, 2010, Accessed: 00, 2020. [Online]. Available: