Models formulated as graphs can be used in many fields such as machine learning, statistics, bioinformatics, medical diagnosis, speech processing and recognition, image processing, computer vision, pattern recognition, computer networks, telecommunications and control theory. The attractive feature of graph-based methods is that, once the problem at hand has been abstracted to a relational structure, techniques from linear algebra, statistics, probability theory and spectral theory can be used for purposes of analysis.
In computer vision and pattern recognition, techniques from graph theory have been used to develop algorithms for purposes of segmentation and grouping, tracking, image database indexing and retrieval and simulation where the uncertain relationships between inputs, intermediate representations, and outputs of the algorithms are taken into account by probability theory. In the other way around, complex probability models for real-world applications often involve millions of random variables and intractable density functions, so probabilities cannot be computed using straightforward approaches such as algebraic manipulations and it is advantegeous to augment the analysis using diagrammatic representation of probability distributions. Thus, probabilistic graphical models bring graph theory and probability theory together for multivariate statistical modelling. For instance, graphical models provide powerful computational support for the Bayesian approach to computer vision, which has become a standard framework for addressing vision problems. Some other well known graphical modelling tools include Kalman filters for tracking problems, nonlinear Bayesian filters for probabilistic control systems and probabilistic robotics, Bayesian networks for decentralized (distributed) systems, bioinformatics and genetics, Markov Random Fields for speech processing, various generative and discriminative graphical models for machine learning etc. Also, each year new models have been developed. More importantly, the graphical model formalism makes it possible to generalize these tools and develop novel statistical representations and associated algorithms for inference and learning. In most of the cases direct inference in probabilistic graphical models may be intractable due to the computational load. Thus, in recent years some methods for approximate inference techniques have been developed but there are still many open problems and unresolved issues. Similarly there are various learning approaches applied to probabilistic graphical models.
In this session, we want to cover various graphical modeling approaches with their inference and learning methods for computer vision and pattern recognition problems. Some of the theoretical topics and applications are listed below but other related studies are more than welcome.
Authors should submit a six page manuscript in double-column format, A4 page size, including authors' names, affiliations, and an abstract. Submissions formating style should adhere to the ISPA author instructions. Papers should be submitted online using the ISPA electronic submission form.
For further information prospective authors should contact special session organizer and chair.
Professor Ilkay Ulusoy, Middle Eastern Technical University, Turkey