LEARNING FROM MULTIMEDIA DATA
Many datasets contain more than just a single type of data. The existing algorithms can usually only cope with a single type data. How can we design methods that can take on multimedia data from multiple modalities? Are we to apply separate learning algorithms to each data modality, and combine their results, or are there to be algorithms that can handle multimedia multiple modalities on a feature-level. Learning casual relationships among visual stored data is another important area, which can benefit from our project. Most existing learning algorithms detect only correlations, but are unable to model causality and hence fail to predict the effect of external controls. Visualization and interactive discovery data mining is a process, which involves automated data analysis and control decisions by an expert of the domain. For example, patterns in many large- scale business system databases might be discovered interactively, by a human expert looking at the data, as it is done with medical data. Data visualization is specifically difficult when data is high dimensional, specifically when it involves non-numerical data such as text. The projects might be a basis for designing interactive business tools.
COMPUTABLE WORLD MODELS
The technique in Nourani (1991, 1995a, 1999g) for model building as applied to the problem of AI reasoning allows us to build and extend models by diagrams. This requires us to define the notion of generalized or generic diagram. The G-diagrams are used to build models with a minimal family of generalized Skolem functions. The minimal sets of function symbols are those with which a model can be built inductively. We focus our attention on such models, since they are computable (Bailey et al., 1995). The G-diagram methods applied and further developed here, allow us to formulate AI world descriptions, theories, and models in a minimal computable manner. It further allows us to view the world from only the relevant functions. Thus models and proofs for the specified problems can be characterized by models computable by a set of functions. The G-diagram functions can define IM objects and be applied with the Morph Gentzen logic.