Alexander G. Schwing
Alexander G. Schwing
Home Projects Publications Teaching Links

 
Most of the fundamental ideas of science
are essentially simple,
and may, as a rule,
be expressed in a language comprehensible to everyone.
(Albert Einstein)
Distributed Structured Prediction with and without Latent Variables for General Graphical Models
This project subsumes work for learning parameters of structured distributions. Subsequently we detail
  • 1. a method for efficient structured prediction with latent variables for general graphical models
  • 2. distributed learning of structured predictors
  • 3. distribtued cutting plane learning of structured predictors
Thanks for reporting suggestions, bugs and successful use-cases.
 
1. Efficient Structured Prediction with Latent Variables for General Graphical Models
by: A.G. Schwing, T. Hazan, M. Pollefeys and R. Urtasun
In this paper we propose a unified framework for structured prediction with latent variables which includes hidden conditional random fields and latent structured support vector machines as special cases. We describe a local entropy approximation for this general formulation using duality, and derive an efficient message passing algorithm that is guaranteed to converge. We demonstrate its effectiveness in the tasks of image segmentation as well as 3D indoor scene understanding from single images, showing that our approach is superior to latent structured support vector machines and hidden conditional random fields.
Preprint:
Supplementary:
 
The ICML 2012 talk explaining this work:
Source Code: Please see below for publicly available implementations.
 
2. Distributed Learning of Structured Predictors
by: A.G. Schwing, T. Hazan, M. Pollefeys and R. Urtasun
Main Features:
  • latent variable models
  • parallelized for high-performance computing
  • sparse feature vectors
  • any form of higher order regions
  • arbitrary counting numbers
  • ...
License:
distributed Structured Prediction is licensed under GPL v3 (or higher). Upon request, other licensing options are available, e.g., if you want to use distributed Structured Prediction in a closed-source product.
Citations:
If you write a paper describing research that made use of this library, please cite the following papers describing distributed Structured Prediction.

Alexander G. Schwing, Tamir Hazan, Marc Pollefeys and Raquel Urtasun;
Distributed Structured Prediction for Big Data;
In Proc. NIPS Workshop on Big Learning 2012

Alexander G. Schwing, Tamir Hazan, Marc Pollefeys and Raquel Urtasun;
Distributed Message Passing for Large Scale Graphical Models;
In Proc. CVPR 2011;

Tamir Hazan and Raquel Urtasun;
A Primal-Dual Message-Passing Algorithm for Approximated Large Scale Structured Prediction;
In Proc. NIPS 2010

 
@inproceedings{SchwingNIPSWSBigLearn2012,
  author = {A.~G. Schwing and T. Hazan and M. Pollefeys and R. Urtasun},
  title = {{Distributed Structured Prediction for Big Data}},
  booktitle = {Proc. NIPS Workshop on Big Learning},
  year = {2012},
}
@inproceedings{SchwingCVPR2011a,
  author = {A.~G. Schwing and T. Hazan and M. Pollefeys and R. Urtasun},
  title = {{Distributed Message Passing for Large Scale Graphical Models}},
  booktitle = {Proc. CVPR},
  year = {2011},
}
@inproceedings{HazanUrtasunNIPS2010,
  author = {T. Hazan and R. Urtasun},
  title = {{A Primal-Dual Message-Passing Algorithm for Approximated Large Scale Structured Prediction}},
  booktitle = {Proc. NIPS},
  year = {2010},
}
Downloads:
  • New release: (Feb. 09, 2014) Source Code v5.2
  • Previous release: (published upon request, not maintained/supported anymore) Source Code v3
  • All other versions were used internally or by a restricted set of people.
 
3. Distributed Cutting Plane Learning of Structured Predictors
A re-implementation of cutting plane learning for structured predictors suitable for distributed computing.
License:
Structured Prediction is licensed under GPL v3 (or higher). Upon request, other licensing options are available, e.g., if you want to use Structured Prediction in a closed-source product.
Citations:
If you use this software package please cite appropriate papers from T. Joachims and B. Taskar and collaborators (see README for details). If you feel like acknowledging my work on this implementation please refer to the work for which this implementation was created, i.e.,

Alexander G. Schwing, Sanja Fidler, Marc Pollefeys and Raquel Urtasun;
Box In the Box: Joint 3D Layout and Object Reasoning from Single Images;
In Proc. ICCV 2013

@inproceedings{SchwingICCV2013,
  author = {A.~G. Schwing and S. Fidler and M. Pollefeys and R. Urtasun},
  title = {{Box In the Box: Joint 3D Layout and Object Reasoning from Single Images}},
  booktitle = {Proc. ICCV},
  year = {2013},
}
Downloads:
  • Latent Variable Version: (Feb. 20, 2014) Source Code
  • Standard Version: (Feb. 20, 2014) Source Code
  • CMakeLists file: File generously provided by Gellert Mattyus from DLR. For questions please directly contact .
 
 
© 2009 - 2017 Alexander G. Schwing