Fast Inference in Sparse Coding Algorithms with Applications.pdfVIP

Fast Inference in Sparse Coding Algorithms with Applications to Object Recognition Koray Kavukcuoglu Marc’Aurelio Ranzato Yann LeCun Department of Computer Science Courant Institute of Mathematical Sciences New York University, New York, NY 10003 {koray,ranzato,yann}@ December 4, 2008 Computational and Biological Learning Laboratory Technical Report CBLL-TR-2008-12-01† Abstract Adaptive sparse coding methods learn a possibly overcomplete set of basis functions, such that natural image patches can be reconstructed by linearly combining a small subset of these bases. The applicability of these methods to visual object recognition tasks has been limited because of the prohibitive cost of the optimization algorithms required to compute the sparse representation. In this work we propose a simple and efficient algorithm to learn basis functions. After training, this model also provides a fast and smooth approximator to the optimal representation, achieving even better accuracy than exact sparse coding algorithms on visual object recognition tasks. 1 Introduction Object recognition is one of the most challenging tasks in computer vision. Most methods for visual recognition rely on handcrafted features to represent images. It has been shown that making these representations adaptive to image data can improve performance on vision tasks as demonstrated in [1] in a supervised †Presented at OPT 2008 Optimization for Machine Learning Workshop, Neural Informa- tion Processing Systems, 2008 1 1 INTRODUCTION