Istari Technologies, in cooperation with MIPT’s Department of Computer Science, has presented the following research works:

Semantic Web

Istari supports an active program of semantic web research and associated development projects that started 10 years ago. Research work makes it easier for us to follow the W3C recommendations and expand our core knowledge base. Members of the research team are involved in client projects. As the semantic web takes off, we will be more involved in application projects and consulting support for our clients. We look forward to future enhancements of the semantic web capability.

Intelligent Agents

The Intelligent Agents area of multidisciplinary research explores the way in which machine learning can create an advantage for web data mining. There are multiple perspectives in using multi-agent systems. The discipline includes elements across computer science areas such as machine learning, human-computer interaction, human-robot interaction, intelligent user interfaces, and cognitive science. We believe that the semantic web will be a key to the successful application of intelligent agents in everyday web systems. The ability for everyday users, not experts, to adapt agent and agent systems easily will be the key to their success. Machine learning techniques have had much success over the years when applied to agents, but machine learning techniques have not yet been specifically designed for learning from non-expert users and current techniques are generally not suited for it out of the box. Istari creates easy to implement agent systems and in our research we pay special attention to such issues as:

• How can everyday people solve the problem of teaching autonomous agents?
• What are the proper evaluation metrics for machine learning systems?
• What is the state of the art in human-agent systems?
• What are the grand challenges in building agents that learn from actions?

We work in the realm of software agents that learn from actions, experience and human social interactions.

Machine Reading through the semantic web

The majority of human knowledge is encoded in text. Much of this text is available in machine processable form on the web. But to machines, the knowledge encoded in the texts they read remains inaccessible. They process human texts but do not understand them. With the emerging semantic web significant progress has been made in machine reading, learning and knowledge acquisition. Given significant value in this topic, machine learning through the semantic web will be one of the crucial issues to large-scale deployment of BI and AI systems in the next ten years. Istari’s research effort focuses on machine learning through the semantic web.

Machine Reading through NLP

The traditional machine processing of natural language (NLP) has improved in the last few years because of significant progress in the areas of language processing as morphological analysis, syntactic parsing, proper name recognition, and advanced logical form extraction. Still, computers today are not yet able to generate semantic representations on a scale sufficient to support automatic knowledge acquisition. The goal of Istari research is to make texts semantically processable by machines and to create an advanced machine-learning process. Learning by reading, relates to automatically extracting machine-understandable knowledge from text. It is related to automating the process of knowledge extraction required to acquire new ontologies and lexicons just as humans do. Our topics of interest include:

• Extracting new ontologies from text and expanding ontologies (learning new concepts) by NLP
• Expanding lexicons (linking lexicons to ontologies) through automatic text processing
• Self-learning systems that learn by reading
• Challenges related to extracting knowledge from text gathered from the web
• Semantic integration and interoperability
• Machine learning from encyclopedias
• Goal - directed machine reading

Architecture and Experimental Design for Real-World Systems

As more artificial intelligence (AI) research is implemented in real-world data management software applications, the evaluation of systems designed for various tasks becomes critical. Designing itself can be incredibly challenging. Our research helps to evaluate different aspects of system and interface architecture.

Social semantic web 2.0 and web 3.0

The social web and the semantic web are related and complement each other. However the levels of sophistication are very different in user-submitted content and in the content with embedded semantics. In the case of semantic web, metadata are embedded and logic-backed. In the case of user tagging, there is just basic metadata. Because of this difference, there are still only limited ways for users to find, customize and reuse data between two webs. As a result, semantic web applications are still of limited use and impact. Istari is working on a new generation of applications that combine the strengths of these two approaches: the data flexibility and portability that is characteristic of the semantic web, and the dynamics and authorship advantages of the social web.

Predictive Analysis

Istari’s predictive analysis research will explore new methods for predictive analysis. We will implement emerging semantic web technologies and a multi-disciplinary approach to predictive modeling by leveraging knowledge from both the social and natural sciences. The research and modeling target the development and implementation of new methods and algorithms for predictive analysis and modeling.