Agent Based Computing Machine

This SBIR will develop a new class of computer architecture called an “agent based computing” module (ABC Machine) that enables “cognitive computing” algorithms to be implemented effectively on a large scale. The ABC Machine is a biologically inspired architecture derived from the field of “membrane computing” and is also based upon “statistical dataflow computing”. It operates in local contexts over string operators. The ABC Machine is motivated by analyzing the biochemical processing in cells. The architecture is suited for computing problems not easily solved by traditional machines. It has the properties of very high parallelism, distributed and redundant processing, and graceful degradation. Phase 1 found the ABC Machine to be both feasible and attractive for connectionist problems, symbolic computing problems with fuzzy and deep search spaces, and for machine learning to these problems. For Phase 2, the ABC Machine will be emulated on a High Performance Cluster (HPC) machine to achieve near term results on complex cognitive problems. Phase 1 demonstrated advantages over traditional AI algorithms on cenventional machines in the following ways: 1) pattern recognition problems with very large numbers of classes and noisy data perform well without training, 2) production systems with deep search spaces using heuristics or a "best first" search strategy generate good results with graceful degradation, 3) complex systems simulation with properties of "emergence" can be effectively modeled on a large scale.