CIBRE Modeler Toolkit
While we often use high-fidelity architectures such as ACT-R
during the course of our modeling, we had the need for a
higher-performance alternative when interfacing with real-time
systems. The Cognitive Instance-Based Rule Engine (CIBRE) is a
lightweight cognitive architecture developed at AdCogSys to address
CIBRE is a hybrid system, integrating low-level statistical learning
capabilities with a high-level symbolic framework. The use of
this architecture gives us the ability to handle planning, inference,
and hierarchically structured behavior within a framework that also
supports learning directly from data, and high-speed perceptual-motor
interaction, thereby spanning a breadth of interaction needs in a
lightweight, scalable architecture suited to real-time
applications. Perhaps the single most compelling aspect of the
architecture, however, is the ability to develop a model by demonstrating the behavior rather than through programming.
We have been continuing to use and expand
CIBRE on a variety of projects, both in-house and for customers. To see CIBRE models created through demonstration rather than through programming in action, visit one of the following links:
The development of CIBRE was initially funded by an NSF phase I SBIR award to support the commercializaiton of cognitive frameworks that have a grounding in statistical learning theory. The architecture by itself, however, is only part of the toolkit used to develop CIBRE models. The entire CIBRE Modeler Toolkit includes the following components:
- CIBRE architecture: the core cognitive architecture with
facilities for planning, learning, clustering, and inferencing, plus
the support software to enable debugging and visualizing results, and
storing and accessing accumulated knowledge.
- Cognitive Agent Simulation Interaction Layer (CASIL): a middleware layer that supports interaction between arbitrary cognitive architectures and a range of simulation environments, including translation layers for vocabulary terms and knowledge, geometrical and spatial constructs and actions, that supports true portability in cognitive agents (see published paper here).
- Adaptive Mesh Refinement (AMR) tools: a software platform for
conducting experiments and simulations involving a cognitive agent that
allows a scientist or engineer to test a range of configurations for the
agent while minimizing and optimizing the use of computational
resources (see published paper here).
Taken together, these components make up a comprehensive suite of
modeling tools that has already seen successful deployment in real-time
domains. If you would like to know how CIBRE might help you address your modeling needs, please contact us for more information.