Efficient F Measure Maximization via Weighted Maximum Likelihood

Abstract

The classification models obtained via maximum likelihood-based training do not necessarily reach the optimal $F_\beta $ F β -measure for some user’s choice of $\beta $ β that is achievable with the chosen parametrization. In this work we link the weighted maximum entropy and the optimization of the expected $F_\beta $ F β -measure, by viewing them in the framework of a general common multi-criteria optimization problem. As a result, each solution of the expected $F_\beta $ F β -measure maximization can be realized as a weighted maximum likelihood solution within the maximum entropy model - a well understood and behaved problem for which standard (off the shelf) gradient methods can be used. Based on this insight, we present an efficient algorithm for optimization of the expected $F_\beta $ F β using weighted maximum likelihood with dynamically adaptive weights.

Cite

Text

Dimitroff et al. "Efficient F Measure Maximization via Weighted Maximum Likelihood." Machine Learning, 2015. doi:10.1007/S10994-014-5439-Y

Markdown

[Dimitroff et al. "Efficient F Measure Maximization via Weighted Maximum Likelihood." Machine Learning, 2015.](https://mlanthology.org/mlj/2015/dimitroff2015mlj-efficient/) doi:10.1007/S10994-014-5439-Y

BibTeX

@article{dimitroff2015mlj-efficient,
  title     = {{Efficient F Measure Maximization via Weighted Maximum Likelihood}},
  author    = {Dimitroff, Georgi and Georgiev, Georgi and Tolosi, Laura and Popov, Borislav},
  journal   = {Machine Learning},
  year      = {2015},
  pages     = {435-454},
  doi       = {10.1007/S10994-014-5439-Y},
  volume    = {98},
  url       = {https://mlanthology.org/mlj/2015/dimitroff2015mlj-efficient/}
}