3. 実装など
ü?[John Yu et al. 2009]のSSVMの実装のソースコード
が以下で公開されている
§? http://www.cs.cornell.edu/ cnyu/latentssvm/
§? written in C
§? Cornell Univ.のProf. Thorsten の研究グループが沢山論文
を出してるっぽい http://www.cs.cornell.edu/People/tj/
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.
12. 1-Slack Formulation
Theorem1. Any solution ?? of 1-slack OP is also a solution
of N-slack OP (and vice versa), with ξ? = ∑ ξ?
?
?
?. (prove later)
Proof sketch.
ü?optimal n-slack
ü?optimal 1-slack
Therefore, the objective functions are equal for any ?
17. Loss functions
概要
ü?SSVMでは通常のhinge lossでは不十分な場合が多い
§? For example, in natural language parsing, a parse tree that is almost
correct and differs from the correct parse in only one or a few
nodes should be treated differently from a parse tree that is
completely different. [Tsochantaridis et al. 2005]
ü?margin-rescaling, slack-rescalingという2つの方法
でloss functionを導入できる
ü?本にはあまり詳しく書いていないので
[Tsochantaridis et al. 2005] を参照するほうが良い
かも
23. Application: learning to rank
ü?この損失関数と特徴ベクトルを用いて最適化が可能
ü?ただし,最も違反している制約を高速に見つける必要
がある[Yue et al. 2007]
24. References
ü? 竹内一郎,烏山昌幸, (2015),サポートベクトルマシン,講談社
ü? Yu, C.-N., Joachims, T., & Elber, R. (2006). Training Protein Threading
Models Using Structural SVMs. ICML Workshop on Learning in
Structured Output Spaces.
ü? Joachims, T., Finley, T., & Yu, C. N. J. (2009). Cutting-plane training of
structural SVMs. Machine Learning. doi:10.1007/s10994-009-5108-8
ü? John Yu, C.-N., & Joachims, T. (2009). Learning Structural SVMs with
Latent Variables.
ü? Tsochantaridis, I., Joachims, T., Hofmann, T., Altun, Y., & Org, A.-C.
(2005). Large Margin Methods for Structured and Interdependent
Output Variables. Journal of Machine Learning Research, 6, 1453?
1484.
ü? Yue, Y., Finley, T., Radlinski, F., & Joachims, T. (2007). A support
vector method for optimizing average precision. Proceedings of the
30th annual international ACM SIGIR conference on Research and
development in information retrieval , Amsterdam, 271.
doi:10.1145/1277741.1277790