Probabilistic graphical models such as (Dynamic) Bayesian networks (BN), Gaussian Bayesian Networks and Markov Random Networks have been increasingly applied to various fields such as computer vision and computational biology. Current mainstream network structure and parameter learning approaches are derived by focusing on the quantitative data in general. The data-driven approaches suffer when the required training data is inadequate in either quantity or quality. Moreover, they cannot effectively incorporate different kinds of priori information readily available in an application domain. It is well known that efficient use of prior information becomes critical and indispensable to achieve a good estimation on network structure and parameters as well as on the prediction results based on the learned network. This observation is particularly true when the amount of training data is scarce. Moreover, proper way of incorporating domain-specific prior knowledge into the learning algorithm has been shown to effectively alleviate the dependence on the training data, while maintaining the performance accuracy.
The critical usage of prior knowledge to compensate for the scarcity of the data has been widely recognized in many fields. For instance, in computational biology research, deciphering the biological networks underlying the complex phenotypic traits such as human diseases is no doubt crucial to understanding the molecular mechanisms and to developing effective therapeutics. Because of the complexity of the networks and the small number of available functional experiments, in-silico modeling of the underpinning molecular network is challenging. Beyond the structure, the eventual prediction of a gene or protein’s function requires accurate mathematical models to parameterize the structure. However, current mainstream algorithms inevitably bear the following disadvantages: i) Structure learning accuracy is limited due to the availability of functional data. ii) Inability to make accurate quantitative predictions based on the learned network structure and parameters.
Thus, it is crucial to develop novel algorithms addressing one or more of the following problems:
i) improving networks structure learning accuracy, in particular under scarce training data by integrating various types of prior knowledge, and
ii) Producing accurate quantitative predictions with a relative large network.
To addresses these issues, this special session especially encourages submissions on the following (but not limited to) topics
§ Novel algorithms on graphical model structure and/or parameter learning based on integrating training data with the available domain knowledge.
§ Novel algorithms that can capture and represent prior knowledge of different formats for effective graphical model learning;
§ Novel algorithms for systematically incorporating multiple data resources in graphical model learning.
§ Application studies, i.e. integrates a specific form of prior knowledge in the author’s particularly interested domain into the data learning process. In this case, the addressed domain should be of general interests including (but not limited to) science, engineering, economics, politics, sociality, humanities and ethnic, cultural studies.
The submitted papers will be rigorously reviewed. All accepted papers will be published in Conference Proceedings. We are also considering the potential publications as special issues in one (or more) of the following journals (to be confirmed):
§ BMC Bioinformatics,
§ IEEE/ACM Transactions on Computational Biology and Bioinformatics (TCBB)
§ Image and Vision Computing
§ IEEE Transactions on Systems, Man and Cybernetics
This special session will be held as part of the ICMLA’11 conference. Authors should submit papers through the main conference submission website. Papers must correspond to the requirements detailed in the instructions to authors. All conference submissions will be handled electronically. Detailed instructions for submitting the papers are provided on the conference home page at:
Accepted papers should be presented by one of the authors to be published in the conference proceeding. If you have any questions, do not hesitate to direct your questions to the session organizers.
Special Sesssion Chairs
Rui Chang, University of California at San Diego (UCSD) (email@example.com)
Qiang Ji, Rensselaer Polytechnic Institute (RPI) (firstname.lastname@example.org)