Efficient Kernel Sparse Coding Via First-Order Smooth Optimization

We consider the problem of dictionary learning and sparse coding, where the task is to find a concise set of basis vectors that accurately represent the observation data with only small numbers of active bases. Typically formulated as an L1-regularized least-squares problem, the problem incurs compu...

Celý popis

Uloženo v:

Podrobná bibliografie
Vydáno v:	IEEE transaction on neural networks and learning systems Ročník 25; číslo 8; s. 1447 - 1459
Hlavní autor:	Kim, Minyoung
Médium:	Journal Article
Jazyk:	angličtina
Vydáno:	New York, NY IEEE 01.08.2014 Institute of Electrical and Electronics Engineers The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Témata:	Algorithmics. Computability. Computer arithmetics Algorithms Applied sciences Approximation algorithms Approximation methods Artificial intelligence Coding Coding, codes Computer science; control theory; systems Data processing. List processing. Character string processing Dictionaries Dictionary learning Encoding Exact sciences and technology Exact solutions first-order smooth optimization Information, signal and communications theory Kernel kernel representer theorem Kernels Learning Memory organisation. Data processing Neural networks Optimization Optimization techniques Pattern recognition. Digital image processing. Computational geometry Signal and communications theory Software sparse coding Telecommunications and information theory Theorems Theoretical computing Vectors kernel representer theorem sparse coding first-order smooth optimization Dictionary learning Dictionaries Sparse coding Computational complexity Optimization Kernel method Kernel function Linear coding Efficiency Least squares method Optimal code Observation data Sparse representation Representation theory Large scale Learning algorithm Artificial intelligence
ISSN:	2162-237X, 2162-2388, 2162-2388
On-line přístup:	Získat plný text
Tagy:	Přidat tag Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!

Popis
Shrnutí:	We consider the problem of dictionary learning and sparse coding, where the task is to find a concise set of basis vectors that accurately represent the observation data with only small numbers of active bases. Typically formulated as an L1-regularized least-squares problem, the problem incurs computational difficulty originating from the nondifferentiable objective. Recent approaches to sparse coding thus have mainly focused on acceleration of the learning algorithm. In this paper, we propose an even more efficient and scalable sparse coding algorithm based on the first-order smooth optimization technique. The algorithm finds the theoretically guaranteed optimal sparse codes of the epsilon-approximate problem in a series of optimization subproblems, where each subproblem admits analytic solution, hence very fast and scalable with large-scale data. We further extend it to nonlinear sparse coding using kernel trick by showing that the representer theorem holds for the kernel sparse coding problem. This allows us to apply dual optimization, which essentially results in the same linear sparse coding problem in dual variables, highly beneficial compared with the existing methods that suffer from local minima and restricted forms of kernel function. The efficiency of our algorithms is demonstrated for natural stimuli data sets and several image classification problems.
Bibliografie:	ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 14 content type line 23 ObjectType-Article-2 ObjectType-Feature-1
ISSN:	2162-237X 2162-2388 2162-2388
DOI:	10.1109/TNNLS.2013.2294059