We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

stat.ML

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Machine Learning

Title: End-to-End Kernel Learning with Supervised Convolutional Kernel Networks

Authors: Julien Mairal
Abstract: In this paper, we introduce a new image representation based on a multilayer kernel machine. Unlike traditional kernel methods where data representation is decoupled from the prediction task, we learn how to shape the kernel with supervision. We proceed by first proposing improvements of the recently-introduced convolutional kernel networks (CKNs) in the context of unsupervised learning; then, we derive backpropagation rules to take advantage of labeled training data. The resulting model is a new type of convolutional neural network, where optimizing the filters at each layer is equivalent to learning a linear subspace in a reproducing kernel Hilbert space (RKHS). We show that our method achieves reasonably competitive performance for image classification on some standard "deep learning" datasets such as CIFAR-10 and SVHN, and also for image super-resolution, demonstrating the applicability of our approach to a large variety of image-related tasks.
Comments: to appear in Advances in Neural Information Processing Systems (NIPS)
Subjects: Machine Learning (stat.ML); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
Cite as: arXiv:1605.06265 [stat.ML]
  (or arXiv:1605.06265v2 [stat.ML] for this version)

Submission history

From: Julien Mairal [view email]
[v1] Fri, 20 May 2016 09:52:14 GMT (9700kb,D)
[v2] Tue, 25 Oct 2016 12:52:50 GMT (9700kb,D)

Link back to: arXiv, form interface, contact.