Gaussian conditional random fields for classification
Апстракт
Gaussian conditional random fields (GCRF) are a well-known structured model for continuous outputs that uses multiple unstructured predictors to form its features and at the same time exploits dependence structure among outputs, which is provided by a similarity measure. In this paper, a Gaussian conditional random field model for structured binary classification (GCRFBC) is proposed. The model is applicable to classification problems with undirected graphs, intractable for standard classification CRFs. The model representation of GCRFBC is extended by latent variables which yield some appealing properties. Thanks to the GCRF latent structure, the model becomes tractable, efficient and open to improvements previously applied to GCRF regression models. In addition, the model allows for reduction of noise, that might appear if structures were defined directly between discrete outputs. Two different forms of the algorithm are presented: GCRFBCb (GCRGBC - Bayesian) and GCRFBCnb (GCRFBC - n...on-Bayesian). The extended method of local variational approximation of sigmoid function is used for solving empirical Bayes in Bayesian GCRFBCb variant, whereas MAP value of latent variables is the basis for learning and inference in the GCRFBCnb variant. The inference in GCRFBCb is solved by Newton-Cotes formulas for one-dimensional integration. Both models are evaluated on synthetic data and real-world data. We show that both models achieve better prediction performance than unstructured predictors. Furthermore, computational and memory complexity is evaluated. Advantages and disadvantages of the proposed GCRFBCb and GCRFBCnb are discussed in detail.
Кључне речи:
Structured classification / Local variational approximation / Gaussian conditional random fields / Empirical Bayes / Discriminative graph-based modelИзвор:
Expert Systems with Applications, 2023, 212Издавач:
- Pergamon-Elsevier Science Ltd, Oxford
Финансирање / пројекти:
- ONR/ONR Global, US [N62909-19-1-2008]
- company Saga New Frontier Group Belgrade
- Аутоматско резоновање и истраживање података (RS-MESTD-Basic Research (BR or ON)-174021)
- Интеракција етиопатогенетских механизама пародонтопатије и паериимплантитиса са системским болестима данашњице (RS-MESTD-Integrated and Interdisciplinary Research (IIR or III)-41008)
DOI: 10.1016/j.eswa.2022.118728
ISSN: 0957-4174
WoS: 000875503900013
Scopus: 2-s2.0-85138167650
Институција/група
Fakultet organizacionih naukaTY - JOUR AU - Petrović, Andrija AU - Nikolić, Mladen AU - Jovanović, Miloš AU - Delibašić, Boris PY - 2023 UR - https://rfos.fon.bg.ac.rs/handle/123456789/2481 AB - Gaussian conditional random fields (GCRF) are a well-known structured model for continuous outputs that uses multiple unstructured predictors to form its features and at the same time exploits dependence structure among outputs, which is provided by a similarity measure. In this paper, a Gaussian conditional random field model for structured binary classification (GCRFBC) is proposed. The model is applicable to classification problems with undirected graphs, intractable for standard classification CRFs. The model representation of GCRFBC is extended by latent variables which yield some appealing properties. Thanks to the GCRF latent structure, the model becomes tractable, efficient and open to improvements previously applied to GCRF regression models. In addition, the model allows for reduction of noise, that might appear if structures were defined directly between discrete outputs. Two different forms of the algorithm are presented: GCRFBCb (GCRGBC - Bayesian) and GCRFBCnb (GCRFBC - non-Bayesian). The extended method of local variational approximation of sigmoid function is used for solving empirical Bayes in Bayesian GCRFBCb variant, whereas MAP value of latent variables is the basis for learning and inference in the GCRFBCnb variant. The inference in GCRFBCb is solved by Newton-Cotes formulas for one-dimensional integration. Both models are evaluated on synthetic data and real-world data. We show that both models achieve better prediction performance than unstructured predictors. Furthermore, computational and memory complexity is evaluated. Advantages and disadvantages of the proposed GCRFBCb and GCRFBCnb are discussed in detail. PB - Pergamon-Elsevier Science Ltd, Oxford T2 - Expert Systems with Applications T1 - Gaussian conditional random fields for classification VL - 212 DO - 10.1016/j.eswa.2022.118728 UR - conv_2788 ER -
@article{ author = "Petrović, Andrija and Nikolić, Mladen and Jovanović, Miloš and Delibašić, Boris", year = "2023", abstract = "Gaussian conditional random fields (GCRF) are a well-known structured model for continuous outputs that uses multiple unstructured predictors to form its features and at the same time exploits dependence structure among outputs, which is provided by a similarity measure. In this paper, a Gaussian conditional random field model for structured binary classification (GCRFBC) is proposed. The model is applicable to classification problems with undirected graphs, intractable for standard classification CRFs. The model representation of GCRFBC is extended by latent variables which yield some appealing properties. Thanks to the GCRF latent structure, the model becomes tractable, efficient and open to improvements previously applied to GCRF regression models. In addition, the model allows for reduction of noise, that might appear if structures were defined directly between discrete outputs. Two different forms of the algorithm are presented: GCRFBCb (GCRGBC - Bayesian) and GCRFBCnb (GCRFBC - non-Bayesian). The extended method of local variational approximation of sigmoid function is used for solving empirical Bayes in Bayesian GCRFBCb variant, whereas MAP value of latent variables is the basis for learning and inference in the GCRFBCnb variant. The inference in GCRFBCb is solved by Newton-Cotes formulas for one-dimensional integration. Both models are evaluated on synthetic data and real-world data. We show that both models achieve better prediction performance than unstructured predictors. Furthermore, computational and memory complexity is evaluated. Advantages and disadvantages of the proposed GCRFBCb and GCRFBCnb are discussed in detail.", publisher = "Pergamon-Elsevier Science Ltd, Oxford", journal = "Expert Systems with Applications", title = "Gaussian conditional random fields for classification", volume = "212", doi = "10.1016/j.eswa.2022.118728", url = "conv_2788" }
Petrović, A., Nikolić, M., Jovanović, M.,& Delibašić, B.. (2023). Gaussian conditional random fields for classification. in Expert Systems with Applications Pergamon-Elsevier Science Ltd, Oxford., 212. https://doi.org/10.1016/j.eswa.2022.118728 conv_2788
Petrović A, Nikolić M, Jovanović M, Delibašić B. Gaussian conditional random fields for classification. in Expert Systems with Applications. 2023;212. doi:10.1016/j.eswa.2022.118728 conv_2788 .
Petrović, Andrija, Nikolić, Mladen, Jovanović, Miloš, Delibašić, Boris, "Gaussian conditional random fields for classification" in Expert Systems with Applications, 212 (2023), https://doi.org/10.1016/j.eswa.2022.118728 ., conv_2788 .