- Title
- On model selection criteria in reverse engineering gene networks using using RNN model
- Creator
- Noman, Nasimul; Palafox, Leon; Iba, Hitoshi
- Relation
- 6th International Conference on Convergence and Hybrid Information Technology (ICHIT 2012). Convergence and Hybrid Information Technology: 6th International Conference, ICHIT 2012: Proceedings (Daejeon, South Korea 23-25 August, 2012) p. 155-164
- Publisher Link
- http://dx.doi.org/10.1007/978-3-642-32645-5_20
- Publisher
- Springer
- Resource Type
- conference paper
- Date
- 2012
- Description
- Recurrent Neural Network (RNN) is a comparatively newer modeling approach to gene regulation process. Several reverse engineering algorithms came out since its proposal and many of them are evolutionary algorithm based approaches. Almost all of these works have used mean square error (MSE) function for fitness evaluation of the alternative gene network models. Akaike’s Information Criteria (AIC) is a well established technique for discriminating the true and the estimated models. In this work we systematically compare the these two alternative model evaluation approaches for reverse engineering genetic networks using RNN formalism. We used Differential Evolution (DE), an elegant evolutionary algorithm for the reverse engineering task and used the above mentioned fitness evaluation techniques with it. We compared MSE and AIC based fitness functions using gene networks of different dimensions and characteristics under both noise-free ideal condition and noisy real condition.
- Subject
- recurrent neural network; Akaike's information criteria; mean square error; differential evolution
- Identifier
- http://hdl.handle.net/1959.13/1057596
- Identifier
- uon:16218
- Identifier
- ISBN:9783642326448
- Language
- eng
- Reviewed
- Hits: 1335
- Visitors: 1323
- Downloads: 0
Thumbnail | File | Description | Size | Format |
---|