Hardware Architectures for Flexible Component-Based Hybrid Cognitive Systems

Creating a realization of cognitive processes, reasoning, and behavior is a challenging problem due to its size, scope and complexity. Current realizations of cognitive systems are software based and typically implemented on conventional von Neumann processors. However, performance is a constraining factor in applications that require extensive testing, training and learning as well as in the implementation cognitive processes that require high memory bandwidth or are computationally intensive. In such cases, hardware processing elements that are tailored or made suitable for cognitive functions will be required. However, due to the rapid evolution and development of cognitive system requirements, missions and architectures, any hardware-based solutions must remain flexible and capable of supporting a variety of cognitive architectures and functions. At the same time, new cognitive system design and architectures should be build out of modules, each which can be specialized to particular hardware platforms without disturbing the other parts of the design. We propose to define a set of detailed requirements and designs for flexible hardware platforms that support both existing cognitive architectures and the next generation of component-based cognitive agents. The requirements and specifications of the hardware will be driven by the existing set of knowledge representations, reasoning representations, and cognitive models and architectures. We will also explore design methodologies for novel component-based cognitive architectures that can be used to exploit hardware processing elements and cognitive processing hardware architectures while retaining the flexibility, modularity and reusability of software implementations on COTS processors.