Matrix backpropagation for deep networks with structured layers

Research output: Chapter in Book/Report/Conference proceedingPaper in conference proceeding

Abstract

Deep neural network architectures have recently produced excellent results in a variety of areas in artificial intelligence and visual recognition, well surpassing traditional shallow architectures trained using hand-designed features. The power of deep networks stems both from their ability to perform local computations followed by pointwise non-linearities over increasingly larger receptive fields, and from the simplicity and scalability of the gradient-descent training procedure based on backpropagation. An open problem is the inclusion of layers that perform global, structured matrix computations like segmentation (e.g. normalized cuts) or higher-order pooling (e.g. log-tangent space metrics defined over the manifold of symmetric positive definite matrices) while preserving the validity and efficiency of an end-to-end deep training framework. In this paper we propose a sound mathematical apparatus to formally integrate global structured computation into deep computation architectures. At the heart of our methodology is the development of the theory and practice of backpropagation that generalizes to the calculus of adjoint matrix variations. We perform segmentation experiments using the BSDS and MSCOCO benchmarks and demonstrate that deep networks relying on second-order pooling and normalized cuts layers, trained end-to-end using matrix backpropagation, outperform counterparts that do not take advantage of such global layers.

Details

Authors
Organisations
External organisations
  • Institute of Mathematics of the Romanian Academy
  • University of Bonn
Research areas and keywords

Subject classification (UKÄ) – MANDATORY

  • Computer Vision and Robotics (Autonomous Systems)
Original languageEnglish
Title of host publicationProceedings - 2015 IEEE International Conference on Computer Vision, ICCV 2015
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages2965-2973
Number of pages9
Volume11-18-December-2015
ISBN (Electronic)9781467383912
StatePublished - 2016 Feb 17
Publication categoryResearch
Peer-reviewedYes
Event15th IEEE International Conference on Computer Vision, ICCV 2015 - Santiago, Chile
Duration: 2015 Dec 112015 Dec 18

Conference

Conference15th IEEE International Conference on Computer Vision, ICCV 2015
CountryChile
CitySantiago
Period2015/12/112015/12/18