Knowledge Systems
Research areas we address:
- The effective use of knowledge to solve problems
- Online agents that function within a physical world
- Infoglut—when too much data is available, identifying the right information a person needs, and delivering it at the time they need it
The capture and appropriate use of knowledge has been a constant theme at Teknowledge since we were a premier expert-systems provider in the 1980s. Computer knowledge bases have become significantly larger since then, and the computing power considerably greater, But the basic problems remain the same. How does one best capture, represent, and employ knowledge in the most effective way for the problem at hand?
Teknowledge has extensive experience with traditional reasoning methods such as rule-based reasoning. We have built significant ontologies and employed formal methods such as theorem proving. However, we always strive to identify and master promising new approaches for solving our clients’ problems. Biologically inspired processing is one such emerging approach. The brain is able to solve problems that are still very difficult for computers after decades of AI research. One such technology we employ is Numenta’s Hierarchical Temporal Memory for learning classifications with human-like flexibility.
Decision makers have noticed that using satellite and World Wide Web information systems is like "standing in front of an information firehose." It is hard to take a sip without getting more information than is useful or necessary. Teknowledge's Knowledge Systems business unit focuses on providing the tools to capitalize on the digital information infrastructure.
Many knowledge-based systems to date have been purely online entities. This is a valuable class of problems, but to be increasingly useful and relevant to ordinary people, intelligent systems must function within the physical world that people do. Our ActionWeb intelligent agent platform obtains its data from data-driven sensor information and by proactively reading sensors. An ActionWeb sensor can be a traditional hardware sensor, programs and data sources available on the Web, or humans. ActionWeb’s inference engine reasons about this dynamic stream of input to decide what actions to take, at what times. Actions can set values in online data sources, but can also interact with people, and activate machinery. ActionWeb agents then apply knowledge within a closed loop—the agent monitors its world, reacts to changes, and attempts to take action to improve the situation it finds itself in.
Some recent, representative projects illustrate our capabilities:
User-Centered Communications Manager (UCCM)
Naval Air Systems Command
UCCM addresses the infoglut problem in aircraft
such as the Broad Area Maritime Surveillance UAS.
The BAMS sensor package will easily be able to generate
far more data than its maximum bandwidth can transmit,
and radio bandwidth can change from moment to moment.
UCCM applies heuristics about the content and context
of each transmission, doctrine, and user preferences
to prioritize transmissions. By dynamically managing
a priority queue, the aircraft can always send the
most valuable information at any time. UCCM is built
on top of the ActionWeb platform to calculate priorities
based on the changing conditions during a mission,
to switch between different communications strategies,
and to make the best use of onboard and bandwidth
resources.
Novel Inference from Massive Data
Advanced Research and Development Activity (ARDA)
Teknowledge completed work on the MILO (Mid-level
ontology) and several domain-level ontologies of
interest to the intelligence community. Given this
knowledge, the main thrust of the work was development
of an analogical reasoning capability for developing
novel hypotheses.
Semantic Search Engine
DARPA DAML Program
This search engine crawled the Semantic Web, indexing pages with OWL markup.
The goal of the search is not to return pages that happen to contain some or all of the
search string, but to pose a question and obtain reasoned answers to it based on
the indexed assertions.