About

This is an implementation of latent structural SVM accompanying the ICML '09 paper "Learning Latent Structural SVMs with Latent Variables". It was developed under Linux and compiles under gcc, built upon the SVM^light software by Thorsten Joachims. There are two versions available. The standalone version using the SVM^light QP solver is available below. Another version using the Mosek quadratic program solver is also available. It has been developed and tested for a longer period of time but requires the separate installation of the solver.

Please send comments and questions to Chun-Nam Yu.

Source Code for Latent Structural SVM API.

The Latent Structural SVM API follows the API design of SVM^struct. To implement the your own application, there are two files that you need to modify. Similar to SVM^struct, you need to supply the data type definitions in 'svm_struct_latent_api_types.h' and the inference function definitions in 'svm_struct_latent_api.c'.

  • latentssvm_v0.12.tar.gz - which includes the API code for implementing your own application. Please follow README.TXT for instructions on compilation and usage.
  • latentssvm_v0.11.mosek.tar.gz - which includes the API code for implementing your own application (Mosek version). Please follow README.TXT for instructions on compilation and usage.

Example Implementations - Discriminative Motif Finding

  • latentmotif_v0.12.tar.gz - the discriminative motif finding application described in the paper. Please follow README.TXT for instructions on compilation and usage.
  • latentmotif_v0.11.mosek.tar.gz - the discriminative motif finding application described in the paper (Mosek version). Please follow README.TXT for instructions on compilation and usage.

Example Implementations - Noun Phrase Coreference Resolution

  • latentnpcoref_v0.12.tar.gz - the noun phrase coreference resolution application described in the paper. Please follow README.TXT for instructions on compilation and usage.

Version History.

  • Feb 8 2010: v0.12 - Fixed convergence issue due to shrinking heuristic for the versions using the SVM^light solver. (thanks to Pawan and Ben Packer for raising the issue)

References.

    • Learning Structural SVMs with Latent Variables (pdf)
    • C.-N. Yu and T. Joachims
    • International Conference on Machine Learning (ICML), 2009