C. Bailey-Kellogg and F. Zhao, "Influence-based model decomposition", Proc. AAAI, 1999, 402-409. [preprint]

Recent rapid advances in MEMS and information processing technology have enabled a new generation of AI robotic systems — so-called Smart Matter systems — that are sensor rich and physically embedded. These systems range from decentralized control systems that regulate building temperature (smart buildings) to vehicle on-board diagnostic and control systems that interrogate large amounts of sensor data. One of the core tasks in the construction and operation of these Smart Matter systems is to synthesize optimal control policies using data rich models for the systems and environment. Unfortunately, these models may contain thousands of coupled real-valued variables and are prohibitively expensive to reason about using traditional optimization techniques such as neural nets and genetic algorithms. This paper introduces a general mechanism for automatically decomposing a large model into smaller subparts so that these subparts can be separately optimized and then combined. The mechanism decomposes a model using an influence graph that records the coupling strengths among constituents of the model. This paper demonstrates the mechanism in an application of decentralized optimization for a temperature regulation problem. Performance data has shown that the approach is much more efficient than the standard discrete optimization algorithms and achieves comparable accuracy.