International Journal of Computational Intelligence Research

Open Access Open Access  Restricted Access Subscription Access
Open Access Open Access Open Access  Restricted Access Restricted Access Subscription Access

Multi-Tree Classification for Uncertain Markov Random Fields


Affiliations
1 Dept. of Computer Science and Engineering, JNTUK UCEV, Vizianagaram
2 Dept. of Information Technology, JNTUK UCEV, Vizianagaram
     

   Subscribe/Renew Journal


Feature generation algorithms for searching globally useful features using traditional Markov network structures is now a days in wide practice. The composition of a Markov network can be represented one of two ways. The first approach is to treat this task as a global search problem. However, these algorithms are slow as they require running the expensive operation of weight learning many times. The second approach involves learning a set of local models and then combining them into a global model. However, it can be computationally expensive to learn the local models for datasets that contain a large number of variables and/or examples.

In this paper, we follow a third approach that views Markov network structure learning as a feature generation problem. For this we used an algorithm DTSL (Decision Tree Structured Learner) which combines a data-driven, adhoc-to-generic search strategy with randomization for quickly generating a large set of candidate features that all have support in the data. In addition to that it uses weight learning, using forest of uncertain decision trees to select a subset of generated features for making feature generation process more accurate and effective.


Keywords

Markov Networks, Rough Decision Trees, Likelihood, DTSL, Inference
Subscription Login to verify subscription
User
Notifications
Font Size


  • Daniel Lowd and Jesse Davis” Learning Markov Network Structure with Decision Trees”, IEEE International Conference on Data Mining, 2010, 334-343.

  • Daniel Lowd and Pedro Domingos, “Efficient weight learning for Markov logic networks”, Eleventh European Conference on Principles and Practice of Knowledge Discovery in Databases, 2007, 200—211.

  • A. Kulesza and F. Pereira, “Structured learning with approximate inference,” in NIPS 20, 2007, pp. 785–792.

  • J. Davis and P. Domingos, “Bottom-up learning of Markov network structure,” in Proceedings of the Twenty-Seventh International Conference on Machine Learning. Haifa, Israel: ACM Press, 2010.

  • H. Poon and P. Domingos, “Sound and efficient inference with probabilistic and deterministic dependencies,” in Proc.AAAI’06, 2006, pp. 458–463.

  • S. Della Pietra, V. Della Pietra, and J. Lafferty, “Inducing features of random fields,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 19, pp. 380–392, 1997.

  • P. Ravikumar, M. J. Wainwright, and J. Lafferty, “Highdimensionalising model selection using L1-regularized logistic regression,” Annals of Statistics, 2009.

  • Stanley Kok, Pedro Domingos “Learning the structure of Markov logic networks” 22nd international conference on Machine learning, 441 – 448.

  • http://archive.ics.uci.edu/ml/datasets/Mushroom.


Abstract Views: 542

PDF Views: 2




  • Multi-Tree Classification for Uncertain Markov Random Fields

Abstract Views: 542  |  PDF Views: 2

Authors

P. Suresh Kumar
Dept. of Computer Science and Engineering, JNTUK UCEV, Vizianagaram
M.H.M. Krishna Prasad
Dept. of Information Technology, JNTUK UCEV, Vizianagaram

Abstract


Feature generation algorithms for searching globally useful features using traditional Markov network structures is now a days in wide practice. The composition of a Markov network can be represented one of two ways. The first approach is to treat this task as a global search problem. However, these algorithms are slow as they require running the expensive operation of weight learning many times. The second approach involves learning a set of local models and then combining them into a global model. However, it can be computationally expensive to learn the local models for datasets that contain a large number of variables and/or examples.

In this paper, we follow a third approach that views Markov network structure learning as a feature generation problem. For this we used an algorithm DTSL (Decision Tree Structured Learner) which combines a data-driven, adhoc-to-generic search strategy with randomization for quickly generating a large set of candidate features that all have support in the data. In addition to that it uses weight learning, using forest of uncertain decision trees to select a subset of generated features for making feature generation process more accurate and effective.


Keywords


Markov Networks, Rough Decision Trees, Likelihood, DTSL, Inference

References